|
|
||
|
|
||
    |
||
|
|
||
 |
|
 |
DataSpeaks, Inc. Issues a Call for Leadership Advancing the Next Big Idea in Software
The next really big new idea in software will help enable humanity to develop practical scientific understanding of mechanisms by which complex systems such as biological systems, patients, and economies work, i.e., function internally, respond to their environments, and act as agents on their environments. The same software also will help enable us to describe how complex systems change and adapt.
It appears that our ability to understand complex adaptive systems has been, until now, about where our understanding of infectious diseases was before the discovery of germs. We as a people have made enormous scientific and practical progress in understanding relatively simple time invariant systems. But our understanding of complex adaptive systems remains in its infancy.
The discovery required to speed practical scientific understanding of complex adaptive systems appears to have been made. This discovery also has the potential to jumpstart economic productivity and human welfare. This discovery is a computational method for using time ordered data to measure temporal contingencies between events as required to understand complex adaptive systems. The events are defined on variables and sets of variables. The computational method also has been described as measuring interactions or longitudinal associations. The value of the required software rests on the fundamental value of measurement in science.
Contingencies matter. Those who landed the Spirit and Opportunity rovers on Mars did not violate any natural laws as they advanced human will to explore the universe. They used the laws of nature by managing to account for and control many contingencies. We may need to measure and account for contingencies - what follows what - in order to understand the nature of complex adaptive systems, including ourselves, and to control our destinies (see responsible agency). Measurement of temporal contingencies helps make them a new subject matter for scientific investigations.
I discovered the required software technology primarily by serendipity while working on a specific problem. The technology is well described as temporal contingency analysis. I founded DataSpeaks, Inc., a one-man startup with technical expertise, prototype software, proof of concept demonstrations, knowledge about how the technology can address fundamental unmet needs in huge markets, and an intellectual property portfolio.
I am calling for leadership to create a company to advance DataSpeaks’ software. DataSpeaks must overcome the awesome power of the statistical establishment, apparently our primary competition, in matters where it counts.
The statistical method is great for describing groups and making inferences from representative samples to populations. The statistical method does not account for individuality and time as required to measure, describe, elucidate, and visualize mechanisms by which complex adaptive systems work,change, and adapt. For lack of better computational methods to understand mechanisms, we are getting swamped in data about systems, not realizing the potential of many great data collection and computer technologies, and often suffering needlessly. In addition, we are hampered in developing artificial systems that learn.
This is a Call for Leadership - business, scientific, academic, and political leadership - as well as leadership in ethics.
DataSpeaks’ software might be as consequential and valuable as the Web browser. The software is basically simple, computationally intensive, and often disruptive. Early adopters can achieve huge competitive advantages. Business leadership needs to be strong and experienced enough to create and manage a company that thrives without being Netscaped. This is a public call for leadership in diverse markets, a strategy that may help keep DataSpeaks from being Netscaped. The new technology is covered by issued patents.
The new technology helps enable sciences of various types of systems. DataSpeaks calls for scientists who investigate complex adaptive systems to lead with the new software technology primarily through demonstrations, discoveries, and publications.
Many great technologies have been invented outside academia. Leading universities established departments of electrical engineering after Edison. They started courses in aerodynamics after the Wright Brothers. Similarly, this new technology calls for academic leaders to establish new departments, grow the intellectual mass of the new methodology, educate tomorrow’s leaders, and provide better counsel about how to process data. Life sciences centers that lead with the new technology would have advantages in becoming the leading life sciences centers in the world.
The new software technology can make clinical research more ethical. In addition, this new technology appears to support a new scientific worldview in which individuals can be held accountable as responsible agents. The new technology also can be misused to threaten humanity. Therefore this also is a call for leadership in ethics.
I also call for political leadership that can help deliver the benefits of this new software technology to the people.
Please peruse DataSpeaks.com for more information. The information is presented from a somewhat personal and historical perspective. I hope that you can share some of my passion and excitement for this adventure. Then consider responding to this Call for Leadership by DataSpeaks, Inc. if you are qualified, think big, and have the will and the resources required to help advance DataSpeaks’ software as the next big idea in software.
DataSpeaks’ Software: What it Does, How it is New, and Why it is Valuable
Curtis A. Bagne, Ph.D. Scientist, Inventor, and Founder of DataSpeaks, Inc.
2. Why We Are Getting Swamped in Data 2.1. Hypothesis-Driven Science and Data-Driven Discovery Science 2.2. The Data Snapshots/Data Movie Analogy 3.2. Benefits of Developing Data Movies 3.3. Data Integration Is Not Sufficient 5. Opportunities and Challenges of Being First 5.1.1. The Statistical Establishment 5.1.1.2. Deference toward Statisticians 5.1.2. MQALA and the Statistical Method - Technical Differentiation 5.2. Stephen Wolfram and Forrest Gump 5.3. Departments of Empirical Induction 6. Eight Selected Market Opportunities 6.1. Revitalizing the Pharmaceutical Industry 6.1.1. Revitalizing Drug Discovery 6.1.2. Re-engineering Clinical Research 6.1.3. Competing Visions for Clinical Research and Practice 6.1.3.1. The Old Vision for Clinical Research and Practice 6.1.3.2. A New Vision for Clinical Research and Practice 6.1.4. Opportunities and Challenges 6.2.1.1.2. Treatment Evaluation 6.2.1.2. Health Care Administrators 6.2.2.1. Health Status Measures and Measurement of Benefit/Harm 6.2.2.2. Individualization, Treatment Guidelines, and Payment Policies 6.2.3. Patients, Potential Patients, and Lay Caregivers 6.2.4. Consumer Driven, Market Oriented Health Care Reform 6.4. Visualizing How Brains Work, Change, and Adapt 6.4.1. Visible Brain, Visible Human 6.5. Improving Prediction of Economies and Capital Markets 6.7. Advancing Responsible Agency 6.8. Reinvigorating Machine Learning and Artificial Intelligence APPENDIX A: How to Develop Data Movies - A Primer on How DataSpeaks Interactions® Works APPENDIX B: Three Proof-of-Concept Demonstrations Demonstration 1 - Reproductive Endocrinology Demonstration 2 - Economic Time Series Demonstration 3 - Functional Brain Image Analysis
DataSpeaks’ Software: The Next Big Idea in Software
Curtis A. Bagne, Ph.D. Scientist, Inventor, and Founder of DataSpeaks, Inc.
DataSpeaks, Inc. offers a major new category of software that empowers users to make discoveries, act more intelligently, and provide better services. In addition, users of health care, financial, and other knowledge and information intensive services should demand that service providers use DataSpeaks’ software because the software often enables patients to receive better care and clients in areas such as financial services to receive more intelligent service.
We are getting swamped in data about complex adaptive systems. DataSpeaks can help overcome this problem by making data useful through empirical induction. We make data speak. ® We make data speak more effectively to human interests and needs.
DataSpeaks’ software will help users understand how complex systems (1) work, (2) change, and (3) adapt. We are, each of us individually, a system. Each of us is made up of subsystems - a nervous system, a cardiovascular system, an immune system, a metabolic system, etc. Many systems are nested.
Each of us is part of larger systems - entire populations, social systems, economic and financial systems, ecosystems, etc. People create systems for business and production.
Complex adaptive systems exist in environments. Both systems and environments are assumed to have parts and attributes that can be measured repeatedly. Furthermore, it is assumed that at least some measured variables fluctuate in level over time in a more or less coordinated manner. Coordinated activity helps define complex adaptive systems. This coordination can be described by measuring interactions for individuals over time.
Complex adaptive systems (1) function internally, (2) respond to their environments including treatments, and (3) act as agents on their environments. Together, these three types of mechanism - function, response, and agency - will be said to describe how systems work.
Furthermore, mechanisms by which complex systems work can change through processes such as development and aging. In addition, systems can adapt through mechanisms such as evolution and learning. Systems can become disordered and respond to interventions such as medical treatments as well as economic and health care policies. Systems work, change, and adapt at different levels and types of system organization and understanding - e.g., physical, chemical, biological, psychological, social, economic, and cultural. Systems such as capital markets change as participants adapt and behave differently.
Emergence is adaptation that crosses thresholds such as speciation to form new types of systems. Biological systems emerged several billion years ago and continued to evolve. New types of systems continue to emerge. Additional levels are being added to hierarchies of control and coordination. Designed systems result from human agency.
Science is an advanced expression of human adaptation.
We come to understand complex systems scientifically as we discover and describe mechanisms by which systems work and adapt as well as how these mechanisms change over time. These mechanisms involve interactions and temporal contingencies that describe coordination.
DataSpeaks’ software is new, unique, and valuable because it appears to be the first software system that actually and effectively measures interactions or temporal contingencies over time for individual systems. DataSpeaks’ software product is called DataSpeaks Interactions®. The software applies to time ordered data from measures and yields values of new measures. DataSpeaks Interactions® elucidates mechanisms. It helps tell us how systems work. It also measures, apparently for the first time, apparent benefit/harm as an interaction over time between repeated measurements of treatments and repeated measurements of health for the same individual.
The value of DataSpeaks Interactions® rests on the fundamental value of measurement in science. One essential feature of DataSpeaks Interactions® is especially notable because it both contrasts sharply with prevailing practice in processing dimensional data and has major scientific import. When applied to dimensional time ordered data, DataSpeaks Interactions® first defines potentially large numbers of discrete independent and dependent events and determines their presence or absence on most repeated measurement occasions. Then it measures interactions, temporal contingencies, or longitudinal associations between various types of independent events and various types of dependent events. The resulting measures of interaction are new dimensional variables.
This method of going from dimensional variables to new dimensional variables through discrete events appears to have major scientific import. Scientists often seek natural laws such as e=mc2, which describe functional relationships involving measured quantities such as energy, mass, and the speed of light. Many scientists use sets of functional relationships to form mathematical models.
DataSpeaks Interactions® provides fundamentally new measures of interaction that can be used to form mathematical models of the mechanisms by which complex systems work and adapt as well as how these mechanisms change.
The new measures of interaction provided by DataSpeaks Interactions® essentially are measures of temporal contingency. By measuring temporal contingencies, temporal contingencies can be investigated scientifically. The new measures of temporal contingency appear to be a key for a new method to advance scientific investigations from simple time invariant systems to complex adaptive systems. The new measures of interaction will help shape scientific worldviews.
In addition, DataSpeaks Interactions® appears to advance a union of apparent opposites (determinacy versus mere contingency) by making something as apparently ephemeral and non-consequential as temporal contingencies a subject matter for functional relationships, mathematical models and scientific laws. Furthermore, as we shall see, temporal contingencies appear to be productive in nature through mechanisms such as natural selection and learning - mechanisms that help shape and characterize complex adaptive systems.
DataSpeaks Interactions® helps enable systems science. Sciences of various systems can empower and motivate us to act more intelligently. It can help us make better predictions and decisions. Sciences of systems can help us discover new products such as drugs, develop such products more efficiently, improve services such as health care and investment advice and help us design systems that meet human needs. DataSpeaks Interactions® will improve economic productivity and enhance human welfare.
But first, in order to accomplish all of this, leaders need to understand why we are getting swamped in data about systems and what we can do to overcome this problem.
2. Why We Are Getting Swamped in Data
We are getting swamped in data about systems primarily because of two interdependent problems. One problem is the data problem. Much of the data that we currently have and continue to collect is of limited value because the prevailing data collection design provides little of the information about time and individuality that is needed to understand complex adaptive systems. I will introduce this problem and its solution with an analogy that illustrates why data movies are better than data snapshots for understanding mechanisms.
The other problem is the software problem. Current software does not measure the interactions that describe the mechanisms by which individual complex systems work and adapt as well as how these mechanisms change. Furthermore, current software does not measure benefit/harm in evaluative investigations such as clinical trials as illustrated in Appendix A. This is the fundamental problem that is solved by DataSpeaks Interactions®.
I will introduce the problem of getting swamped in data in the context of hypothesis-driven science versus data driven discovery science for biological systems. Similar problems exist in disciplines that deal with other types of systems. Then I will present the analogy. I also will anticipate and address reasons why some people will delay progress and resist getting out of data swamps.
2.1. Hypothesis-Driven Science and Data-Driven Discovery Science
Science is evolving to advance objective understanding of nature. Traditionally, many scientists have tested hypotheses involving a few variables at a time that form small parts of more complex systems. Scientists were admonished by statisticians not to collect data except insofar as it could be used to test specific hypotheses. These aspects of hypothesis-driven science, together with a lack of high throughput measurement technologies and a lack of extensive data collection and storage infrastructures, helped keep scientists out of data swamps. But these limitations also impeded scientific progress.
Recently, scientific practice has been evolving toward discovery science, which encourages scientists to collect data whether or not the data are to be used for testing specific hypotheses. Discovery science has been said to be data-driven instead of hypothesis-driven. Completion of the Human Genome Project (HGP) is a major achievement of discovery science (http://www.nhgri.nih.gov/). The HGP and related gene sequencing projects have and continue to produce vast quantities of useful data.
Sequence data describe genes, which are relatively static or timeless compared to the dynamic mechanisms of life. Sequence data have been described figuratively with terms such as maps, parts lists, and genetic snapshots. DataSpeaks Interactions® was not developed for such timeless data. But it can help make timeless data, including gene data, useful.
“Genomes to Life” is a follow-up program on HGP (http://doegenomestolife.org/). As the name suggests, it is intended to address the dynamic mechanisms of living systems. Using the terminology that I introduced above, the Genomes to Life Program can be described as elucidating how biological systems work, change, and adapt. The Human Proteome Project (HPP) of the Human Proteome Organization (HUPO) is another major follow-up program, which seeks to identify both proteins and their mechanisms (http://www.hupo.org/).
The data processing methods and software for hypothesis-driven science, which still prevail, are not well suited for discovery science. As a result, discovery science is heavy on data and short on discovery. This problem is illustrated by the low and declining productivity of the pharmaceutical industry, even after HGP.
Hypothesis-driven science has had a glorious and productive history largely because scientists did most of the intellectual heavy lifting. Scientists form theories and hypotheses in their heads. This can involve thought experiments. Then the key data processing step is to reject null hypotheses. But methods for rejecting null hypotheses are not sufficient to make scientific discoveries or to convert data into scientific information and knowledge - to make data speak through empirical induction.
Hypothesis-driven science works best for relatively simple time invariant systems such as planetary systems that have been well described with few variables. Part of the reason why discovery science has not been more productive for entire biological systems is that biological systems involve data about more variables than scientists can process well in their heads.
The data processing methods and software used by hypothesis driven science have additional limitations for discovery science. Many hypotheses attempt to address mechanisms. But for lack of software to measure mechanisms over time for individuals, most hypotheses are about the levels of variables at particular times. This problem involving levels is discussed more fully elsewhere in the context of drug discovery.
It is difficult to understand mechanisms scientifically without measuring mechanisms. Demonstration 1 measures mechanisms in the context of reproductive endinocrinology. Similarly, it often is difficult to test hypotheses about the safety and efficacy of treatments without measuring the benefit/harm of treatments. Appendix A illustrates the measurement of benefit/harm. Failure to measure mechanisms and failure to measure benefit/harm are standard operating procedures in academe and industry.
Another limitation involves individuality. Decoding the human genome was a great achievement. But describing it as “the human genome” is an oversimplification because there are more human genomes than people who are not identical twins. Small differences in genomes can make important differences. The human and mouse genomes are about 98% similar.
About 1.8 million single nucleotide polymorphisms have been identified by the SNP Consortium (http://snp.cshl.org/). These SNPs can exist in many combinations to help account for human differences. Many of these differences are relevant to human health, health disorders and responses to treatments. Statistical methods of data processing that are based on measures of group variability and central tendency tend to obscure, rather than elucidate, individuality. It often is valuable to distinguish individual differences from measurement errors.
Humans appear to be more complex than mice or corn not because humans have more genes but because the products of gene expression and other substances are involved in more complex and higher orders of control and coordination, including the control of gene expression. This suggests that it will be valuable to measure the interactions that describe emergent mechanisms and coordination in addition to decoding genomes and measuring the levels of gene expression.
Personalized medicine is one of the great promises of the modern era in biology (http://www.ornl.gov/sci/techresources/Human_Genome/medicine/medicine.shtml). A major reason why this promise remains elusive is that key data processing methods of hypothesis-driven science continue to dominate but are largely incompatible with personalized medicine because conventional methods do not measure interactions or adequately account for time and individuality. Individuals often are not well represented by group averages.
Hypothesis testing will remain an important part of science. But scientists should demand more from software to help them understand complex adaptive systems as required for making discoveries and advancing their careers.
2.2. The Data Snapshots/Data Movie Analogy
I will introduce the solution to the data problem with an analogy that begins by distinguishing “data snapshots” from “data movies.” Start by thinking of data as recorded experience.
Consider the business and financial pages of your favorite newspaper. The information consists primarily of prices for stocks and mutual funds as well as the levels of various indices, economic time series, and measures of business performance together with changes from previous days or other periods.
The quantitative information in the business pages for a particular day illustrates a “data snapshot.” The corresponding “data movie” consists of the same information for a series of days over a period of time, preferably a prolonged period with many repeated measurements. Each variable in a data movie is a time series variable.
Data movies provide more information about time and individuality than data snapshots. Scientists often acknowledge that data movies would be superior for certain problems. However, most scientists continue to collect data snapshots. In addition, clinical trials that collect repeated measurements data often are analyzed as it the data were collections of data snapshots. Leaders in science, technology, business, industry, and academia need to understand why they are not experiencing the advantages of data movies.
Please exercise your imagination to appreciate the “information advantage” of data movies. Imagine that you don’t know anything about American football and that you are being challenged to discover the rules of the game from recorded experience without a rule book or instruction - much as investigators such as biologists are being challenged to use data to discover the mechanisms by which complex systems work and adapt and how these mechanisms change.
You have a choice - investigators often really do - as you seek to discover the rules. Assume that one frame in a movie is the same as a snapshot and that 100,000 frames are adequate for a movie of one entire game. Your first option is to choose a collection of 100,000 snapshots, one snapshot from each of 100,000 individual games. Each snapshot could be of a play of any type, a huddle, a timeout or a half-time show. The snapshots are of 100,000 games and there is no information about temporal order. This first option is called the “extensive design” because it involves the collection of data from each of many individuals, usually only a little data from each.
Your second option is to choose a movie of one entire game. This second option is called the “intensive design” because it involves the collection of a lot of data from only one individual.
Assume that the collection of snapshots and the movie for one entire game occupy the same amount of space in computer memory or on a CD. In this regard, both options appear to have the same amount of information. Which option would you choose? Why?
I suggest that the movie would be a better choice because it includes information about temporal contingencies in the game. For example, the team with the ball has four consecutive chances to score or advance the ball at least 10 yards or else they give up the ball where they are on the field. Information about temporal contingencies is valuable for discovering how nature works, changes, and adapts as well as for evaluating the temporal criterion of causal and other predictive interactions.
Snapshots are timeless. Timeless data are great for showing structures of systems in space. Movies are superior because they provide the additional information required to understand dynamic mechanisms.
Statistical methods are great for investigating measures of team and player performance such as average yards rushing, average yards passing and average number of sacks. But statistical methods by themselves are of limited value for understanding the rules of football or the dynamic mechanisms of individual complex adaptive systems.
To help further appreciate the information advantage of data in temporal order, imagine trying to understand football if the frames in the game movie were shown in random order. The “additional information” advantage of data being in temporal order is an underutilized world of evidence to gain scientific understanding. People rely on temporal order to help make sense of experience, but most of our data processing software works best for timeless data.
Data snapshots are not good for elucidating mechanisms or for quantifying the benefit/harm of treatments. Data movies are apt to become the gold standard for collecting data to help understand how complex systems work, change, and adapt. However, there are sources of resistance.
Given that intensive designs and data movies can provide an information advantage, why aren’t people using them more often and effectively? Here are some reasons. Some are good. Some reasons are bad.
First, here is a good reason for not collecting data movies. In some cases, the objective is to investigate treatments that affect the survival of systems such as patients. Here the treatment is considered to be either present or absent for each individual and the objective is to investigate time until a largely irreversible change in a system such as death. In survival investigations, time is mostly irrelevant except for length. Most of us care about survival. But survival investigations reveal little about dynamic mechanisms or the benefit/harm of treatments as it becomes evident over time for individuals.
In some cases, it is necessary to destroy systems in order to collect data. For example, it might be necessary to sacrifice animals. When this is necessary, data movies are not possible. However, many new nondestructive and minimally invasive measurement technologies are being developed that can monitor systems over time. For example, it is becoming possible to image activities in individual cells without destroying the cells. A major value of these technologies is that they allow collection of data movies.
An often specious reason for not collecting data movies is that they are considered to be more expensive than collections of data snapshots. But, returning to our analogy, would it be more expensive to obtain a movie of one entire game or to form a collection of 100,000 data snapshots, one from each of 100,000 games? It can be expensive to recruit, screen, and enroll large samples of individual subjects for extensive designs. In addition, there are the opportunity costs of failing to understand complex adaptive systems. Often it would be best to allocate scarce data collection resources to some combination of intensive and extensive designs - moderate numbers of repeated measurements for each individual in samples with moderate numbers of individuals.
A nexus of reasons for resisting intensive designs and data movies derive from habits learned while accommodating limitations of dominant methods better suited for investigations of simple time invariant systems than for complex adaptive systems. Such habits can be impediments to progress. Progress often requires changing the status quo. Brilliant people are not immune to bad habits and the status quo, perhaps especially when the habits and the status quo help define professional identities and when the habits have been enormously successful for limited but important classes of problems. Sources of resistance based on the status quo may be some of the most difficult challenges to overcome while advancing DataSpeaks Interactions®.
Science makes generalizations and looks for patterns. It appears that methods best suited for time invariant systems often have led investigators to favor generalizations across individuals at the expense of making generalizations and looking for patterns over time. A comprehensive understanding of nature, especially of complex adaptive systems, appears to require both types of generalization. Methods for generalizing across individuals and methods for generalizing over time are different but complementary, not antithetical. For example, DataSpeaks Interactions®, which generalizes and describes patterns over time, often is complementary to software for statistical analyses.
Another important habit and source of resistance involves experimental control. Randomization is important for achieving experimental control and isolating the effects of particular variables. However, it also appears that methods best suited for time invariant systems often have led investigators to favor randomization of individuals to different treatment groups at the expense of randomization of treatments to different periods of time for the same individual. However, Gordon H. Guyatt, a leader in evidence-based medicine, has identified randomized N-of-1 clinical trials as the gold standard for evidence based medicine under many circumstances (see, for example, http://www.cche.net/usersguides/applying.asp ). As with generalization, both types of randomization often can be complementary. A good choice, enabled by DataSpeaks Interactions®, often would be to use both types of randomization simultaneously in particular investigations. This option is described as a “double randomization design” in Section 2.5 of Patent 6,317,700.
I have had investigators express concern that results obtained from intensive designs may not apply to populations. This is true. But one way toward understanding both individuals and populations is to collect data movies from samples of individuals that represent populations. This approach also is an excellent way to investigate individual differences in mechanisms, disorders, and responses to treatments. There often is value in understanding individuals such as people, patients, populations, economies, capital markets, and ecosystems whether or not they can be sampled across individuals.
Some investigators collect or advocate the collection of lots of data snapshots because they think they have no other choice. Call this the first case. Other investigators use the intensive design and collect data movies because they have no other choice, the second case. How is it that some investigators think they have no other choice but to collect data snapshots while other investigators essentially have to collect data movies if they are to collect data at all?
This exemplifies the first case. The editor on an online journal, editorializing about “individuality and medicine,” recently said that we have no other choice for the sake of preventive medicine but to invest in what essentially are large collections of data snapshots, perhaps with a survival component. His conviction that there is no other choice testifies to the power of the statistical establishment to limit options. This case characterizes much of “large scale biology.” Important money is riding on extensive data collection designs as if they were the best choice for personalizing medicine.
Large scale biology has value. But with tens of thousands of genes and hundreds of thousands of proteins and other biologically active substances, and at least hundreds of thousands of variations in both sets, and both sets working in nearly unlimited numbers of combinations together in people with different histories in different environments to cause both normal function and disorder, it is an open question if there are enough people in the world for extensive designs alone to achieve their planned promise of understanding human health and disease. Furthermore, understandings of existent systems are apt to become outdated because people are agents creating new agents and our future.
Biological mechanisms are complex compared to the rules of football as the latter was illustrated in our analogy. How much will it really help to go from 100 individuals (games or patients) to 100,000 individuals or perhaps 1,000,000 or more individuals if we don’t capture and use more information about temporal contingencies? Current conceptions of large scale biology might not be efficient strategies for understanding biological systems on the way to preventing health disorders.
This exemplifies the second case. Those who investigate individuals such as economies and capital markets are prone to collect data movies largely because individuals are so inclusive and unique that sampling of individuals has been largely precluded and often is considered irrelevant. If such investigators are going to continue collecting data, data almost must be collected repeatedly. Investigators of unique systems have to accept individuality and try to account for time by generalizing and seeking patterns over time.
Collection of data movies is not sufficient for understanding complex adaptive systems. After all, data swamps include lots of data about economies and capital markets. Scientific understanding does not just flow from data movies, especially when they have many variables.
This introduces the primary problem that hinders our understanding of complex adaptive systems and keeps us in data swamps. This is the software problem. Current software does not measure interactions, temporal contingencies, or longitudinal associations. If economists and investors can not do a better job with data movies given the financial incentives for better prediction, why should investigators such as biologists collect data movies?
This has been a good reason to avoid intensive data collection designs. It has been a good reason not to collect data movies - until now.
Investigators can get out of data swamps by taking two major steps. The first step is to collect more data movies, preferably under conditions of experimental control. Fortunately, our capability to collect data movies is growing rapidly. Microarrays allow collection of data on thousands of variables at each time. Technologies such as functional brain imaging and Web-enabled monitoring devices are increasing the collection of time series data by orders of magnitude.
Data movies should be collected using time series experimental designs whenever feasible. Such designs vary independent variables over time for individuals and preferably randomize different levels of independent variables to different periods of time for the same individual. Such randomization contrasts with randomization of individuals to different treatment groups. Both forms of randomization can be used together when it is feasible and desirable to use samples of individuals to make inferences about populations.
Experimental control and randomization would help assure that values of the measures of interaction obtained with DataSpeaks Interactions® are valid measures of causal interactions. DataSpeaks Interactions® is exceptionally well-suited to evaluate the temporal criterion of causal interactions with or without experimental control and randomization.
Measures can not be valid unless they are reliable. There are two major aspects of reliability when considering measures obtained with DataSpeaks Interactions®. In general, the reliability of the new measures of interaction can be improved by collecting data from more repeated measurements. This first aspect of reliability works in a manner somewhat analogous to how large samples increase statistical power. More repeated measurements help overcome unreliability of measurement in data that are processed with DataSpeaks Interactions®.
The second aspect involves the reliability of computation. Given the input data and a scoring protocol, measures obtained with DataSpeaks Interactions® are as reliable as computation.
Collecting data movies can be counterproductive without DataSpeaks Interactions® because one data movie has more data than one data snapshot with the same number of variables.
The second major step for getting out of data swamps is solve the software problem by processing data movies with DataSpeaks Interactions®. DataSpeaks Interactions® can be viewed as a new category of software - computational measurement software - to “develop” data movies, which is described more fully in the patents section. Development consists of measuring patterns of interaction that describe how complex adaptive systems work, change, and adapt. Development makes data movies useful.
DataSpeaks Interactions® is the heretofore missing step for using computation to benefit from the “information advantage” of data movies as compared to data snapshots. DataSpeaks Interactions® enables more and better use of data movies. Fortunately, computing infrastructure is beginning to have sufficient power to develop data movies of complex adaptive systems.
Appendix A is a brief primer on how DataSpeaks Interactions® works.
3.2. Benefits of Developing Data Movies
An old maxim is that if one wants to investigate something scientifically, measure it. DataSpeaks Interactions® measures the interactions that describe the mechanisms by which complex systems work and adapt as well as how these mechanisms change. Thus, in a very real way, DataSpeaks Interactions® helps enable systems science. DataSpeaks Interactions® provides operational definitions of “interaction,” an ill-defined but increasingly used concept in science. The value of DataSpeaks Interactions® rests on one of the very foundations of science, which is measurement.
Measurement also enables visualization. Values of measures can be graphed and visualized in various ways. Visualization aids understanding. By developing data movies, DataSpeaks Interactions® makes data and interactions more visual. An old proverb is that a picture is worth a thousand words. Similarly, a visual display can do more to aid understanding than a computer memory full of data.
To the best of my knowledge, no one else has ever measured and visualized interactions between variables over time for individual systems as functions of the relevant analysis parameters. This statement is supported by the whole long process that led to my patents.
Development of data movies with DataSpeaks Interactions® appears to be a critical missing step to help understand many complex adaptive systems. The 2003 Nobel Memorial Prize in Economic Science went to Robert F. Engle and Clive W. J. Granger for their methodological work involving volatility of time series. Apparently this is the first time this prize has gone to econometricians. This speaks to the growing recognition of the importance of methods for processing time series data. However, such methods remain in their infancy, which is evidenced by the depth of controversy about what time series data say about economic policies and investment decisions.
Visualization of economic and capital market data movies still is limited primarily to showing trends. When different trends are shown side by side, people are largely left to form subjective impressions about how different variables interact in ways that might be predictive. The task of forming reproducible subjective impressions, both within and across individual people, becomes enormously difficult as the number of variables increases. Demonstration 2 illustrates the measurement of interactions between economic time series.
To appreciate the difficulty of visualizing predictive interactions, consider all the economic and financial variables that are reported in your daily newspaper. Imagine an enormous chart showing trends for hundreds of variables over a prolonged period of time. Imagine trying to understand this chart. No wonder different investigators - and individual investigators at different times - often arrive at different conclusions and make different predictions from the same data. They often focus on different parts of the entire system, draw different conclusions and make different predictions. And so people long for one-handed economists and suffer the consequences of polarizing disputes about economic policies. DataSpeaks Interactions® measures interactions by computation and yields dimensional measures of direction and degree. These measures are potential antidotes to excessively polarized rhetoric.
Mathematical modeling may be viewed as an alternative to forming subjective impressions about predictive interactions. Mathematical models can be extremely valuable. One major problem is that such models currently are formed without measuring the interactions that describe mechanisms by which complex systems work and adapt as well as how these mechanisms change. Thus the models are subject to the limitations of data processing methods that do not adequately account for time and individuality as well as the limitations of subjective impressions of those who model. DataSpeaks Interactions® can help inform the development of mathematical models, including models used in computational biology and economics.
DataSpeaks Interactions® measures interactions by computation. Computation is superior to subjective impressions because results can be obtained with transparent and reproducible procedures that can be expressed in protocols and applied to large complicated databases with many variables. Although measurement of interactions for large data movies or sets of data movies will remain a major task, DataSpeaks Interactions® provides many new options for addressing such tasks systematically with the power of computing. Many data processing strategy options can be automated to seek predictive patterns with little human intervention. Application of DataSpeaks Interactions® to data movies will help get us out of data swamps.
3.3. Data Integration Is Not Sufficient
Biological systems have many parts and subsystems. We need to understand the mechanisms by which all these parts and subsystems work, change, and adapt together. This has been called “integrative biology.” A recent Google search on this phrase yielded 35,000 hits.
This two-step strategy for getting out of data swamps - collecting more data movies and developing them with DataSpeaks Interactions® - supplements a prevailing trend for understanding systems, which is to integrate data of many types from different sources.
Data integration is important for understanding systems. Data integration does create new opportunities for scientific insight. But data integration is not integrative biology. In order to get from data to knowledge and understanding more efficiently, we need the right kind of data (more data movies) and a means of processing the data (DataSpeaks Interactions®) that is better because it measures and visualizes interactions to elucidate mechanisms.
Data collection and communication infrastructures are being developed that are fast, widely accessible, and have huge memories. Now we need better software to process the data so that investigators do not have to do so much of the intellectual heavy lifting. DataSpeaks Interactions® has great potential to capitalize on this opportunity.
Anticipate both that collections of data snapshots will be largely outmoded by collections of data movies and that those data movies will be processed primarily by DataSpeaks Interactions®.
Two U.S. patents have been issued to me, Founder of DataSpeaks, Inc. Foreign patents are pending. These patents claim features, applications, and uses of the methodology embodied by DataSpeaks Interactions®.
Patent 6,317,700 - “Computational Method and System to Perform Empirical Induction” - was issued on 11/13/2001. All 104 claims that were sought were approved. My understanding is that this can be characterized as a broad foundational patent.
I use the term “empirical induction” to describe procedures for drawing generalized conclusions and making predictions from data. My patents and this Web site are limited primarily to computational methods of empirical induction - methods that can be implemented as computer algorithms and performed by computers. Computer algorithms for empirical induction can be used to draw generalized conclusions and make predictions from data that records experience.
“Empirical induction” is intended to be a broad and growing category that encompasses a number of different methods and algorithms. Computational methods of empirical induction that have been embodied in various ways by software include the statistical method, neural networks, genetic algorithms, cellular automata, and chaos theory. These include artificial systems, inspired by biology, whose inner workings and results can appear to be as indeterminate as those of real systems, which they model.
Patent 6,317,700 claims a new computational method of empirical induction that is called the Method for the Quantitative Analysis of Longitudinal Associations or MQALA. “Longitudinal associations” is another way to describe interactions or temporal contingencies, both causal and non-causal.
MQALA can be described as a suite of computational tools specifically designed to measure, discover, analyze, synthesize, describe, and visualize patterns of temporal contingency in data movies for individual complex adaptive systems. Such patterns include mechanisms by which individual complex systems work, change, and adapt.
MQALA is a computational measurement method of empirical induction. This exemplifies a computational measurement method: density equals the mass of a substance per unit volume. Given two variables, mass and volume, Archimedes essentially discovered or invented a mathematical method to compute values of a new measure, density. Similarly, given data for two variables or sets of variables, I have discovered or invented a mathematical method or algorithm to compute values of new measures of interaction over time between variables and sets of variables for individual systems.
The results of both of these measurement procedures are descriptive rather than inferential. Measures of density describe a relatively static property of materials. Measures of interaction describe dynamic aspects of systems. Compared to the method or algorithm for measuring density, MQALA is substantially more complex. But MQALA is a computational measurement method none-the-less.
Measures often are useful. Measures of density have proven useful for problems such as those involving buoyancy and for the classification and identification of materials. For example, Archimedes developed the concept of density while trying to determine if a crown was made of pure gold. I invented what came to be MQALA while trying to analyze health diary data.
The concept of density and Archimedes’ measurement method has proven to be useful in science and practical affairs. Similarly, I anticipate that the concept of interaction as operationally defined by MQALA and embodied by DataSpeaks Interactions® will withstand the test of time.
Measures of interaction obtained with DataSpeaks Interactions® are useful in that they are quantitative conclusions, generalized over time ordered data, about systems. To illustrate, consider the N-of-1 clinical trial for the “drug for blood pressure” example that is presented in Appendix A. The second part of this example used data for drug dose and 20 health variables collected daily over 100 days. Assume that the resulting overall benefit/harm score was large and positive. Such a score quantifies the conclusion, generalized over all the data, that the drug was beneficial.
Generalized conclusions of this type can be used to help make predictions that can be acted on accordingly. For example, the patient investigated, knowing the results of her N-of-1 clinical trial were as just described, could take the drug on the 101st day with some confidence that the drug would be beneficial. Data from the 101st day could either strengthen or weaken this conclusion. Data could be collected daily and the growing mass of data could be analyzed daily to monitor any change in response.
As illustrated in Appendix A and demonstrated in Appendix B, DataSpeaks Interactions® can be used together with the statistical method to extend generalizations from the individuals actually investigated to populations that samples of individuals represent.
U.S. Patent 6,516,288 - “Method and System to Construct Action Coordination Profiles” - was issued 2/4/03. All 111 claims that were sought were approved. This patent extends the claims of 6,317,700.
The concept of action coordination profiles was inspired by exposure to motion capture technology and by the analogy of coordinated motion. Basically, sets of action coordination profiles show how every variable or selected sets of variables in a data movie for a system interacts with every other variable or selected sets of variables for that system. Profiles could be used to characterize different patterns of coordinated activity such as walk, trot, canter, and gallop for horses. In addition, action coordination profiles can be used to measure the amount and strength of evidence for coordinated motion. For example, I would anticipate that a series of repeated golf drives by an expert golfer would be more coordinated than the same series for a beginner.
Patent 6,516,288 also effectively extends the analogy of coordinated motion to additional types of action and additional types of systems. As examples, the actions can be physical, chemical, biological, behavioral, mental, or social. The systems can be objects of investigation such as brains, organisms, patients, economies, investment markets, populations, machines, and processes. In addition, the concept of coordinated action can be extended to how two or more individual systems interact.
Issued patents can help provide a competitive advantage in business.
5. Opportunities and Challenges of Being First
This call for leadership involves an opportunity to be first with a major new category of software that has great potential to improve economic productivity and human welfare.
This opportunity calls for great leadership because it involves great challenges. Expect rewards to be commensurate with the opportunity and performance.
First, here is some of the evidence that DataSpeaks is first in its category.
“Temporal contingency analysis” may be the best way to describe what MQALA does as described in Appendix A and demonstrated in Appendix B. MQALA is the methodology embodied by DataSpeaks Interactions®. However a recent Google search on the phrase “temporal contingency analysis” yielded zero hits. This result is surprising especially because temporal contingencies appear to be the primary way that people and other organisms learn directly from experience - what follows what. Temporal contingencies seem to capture the essence of what has been called the school of hard knocks. Temporal contingencies also are the source of much scientific inspiration and invention.
“Computational measurement software” may be the best way to describe the new category of software that DataSpeaks Interactions® represents. However a recent Google search on the phrase “computational measurement software” also yielded zero hits. This result is surprising because science and many practical applications of scientific understanding are grounded on measurement. In addition, values of many measures are derived by computation from other measures.
I followed many such descriptors for years before and after Google existed and before and after filing for my patents. I find it amazing that this opportunity still remains. Perhaps the scientific quest for the immutable laws of nature is deterring us from investigating anything as ephemeral and apparently inconsequential as the temporal contingencies that describe how complex systems work, change, and adapt.
Einstein insisted that nature does not play with dice. But perhaps dice are a part of nature, natural and man made. Contingency and chance might be important after all. E=mc2 was instrumental in the development of the atomic bomb. But we may need to recognize the importance of contingencies if we are to understand the behavior of those who might use the bomb.
Some business people fear that being first means that there is no market. The potential market for DataSpeaks Interactions® comprises all those who could benefit from better understanding of complex adaptive systems and from products and services enabled by such understanding.
This definition of a market is too broad to help develop the financials of a business plan. But it is helpful in dealing with the competition and overcoming specific sources of resistance.
Identifying and understanding the competition has proven to be more difficult for me than discovering, inventing, and patenting MQALA. But identifying and understanding the competition, provocative and disturbing as it might be for some, appears to be crucial for progress and success.
Basically, the problem of identifying and understanding the competition is simple, once it is recognized. Statisticians rule, at least where it counts. But the statistical method is not well suited to account for individuality and time as required to efficiently understand mechanisms. Over reliance on the statistical method, perhaps more than anything else, is what is making it difficult for people to benefit from better understanding of complex adaptive systems such as people, patients, organisms, brains, populations, societies, economies, ecosystems, and productive processes.
A deeper understanding of the problem and challenge will make it easier to succeed. The competitive issues cut to the heart of scientific methods for understanding nature, or more specifically, for understanding complex adaptive systems. These issues call for greater study and dialog. Here is some additional detail to help stimulate inquiry.
The statistical establishment is the primary competition for DataSpeaks Interactions®. Statisticians are the primary source of resistance to MQALA. In addition, everyone who defers too much to statisticians on issues related to drawing generalized conclusions and making predictions from data contributes to the problem of understanding complex adaptive systems.
This is not as bad as it may seem. The “statistical method,” as I have chosen to use this descriptor here in the context of complex adaptive systems, is the best method of empirical induction for what it does well - describing groups and making inferences from representative samples of individuals to populations. Inferential statistics is one great way to overcome measurement error by using data for multiple individuals. But the statistical method is not a way to account for individuality and time or to describe mechanisms as illustrated with the analogy. I know of nothing wrong about the statistical method for what it does well.
Unlike the statistical method, MQALA is great to account for individuality and time as well as to describe mechanisms. MQALA also is a great way to account for measurement error within individuals. But MQALA has nothing to say about describing groups as collections of separate individuals or about making inferences from samples of individuals to populations. The two methods are entirely different. I am not aware of any inherent conflict between these two methods. The two methods work best for different types of data and for different types of problems.
Furthermore, MQALA and the statistical method often are complementary. MQALA provides measures of the mechanisms by which individual systems work and adapt as well as how these mechanisms change. MQALA also provides measures of benefit/harm. When obtained from two or more individuals, values of these new measures can be analyzed statistically as described in Appendix A with an example that uses the randomized multiple N-of-1 clinical trial design. In addition, Appendix B demonstrates how values of a measure of hormone interaction can be analyzed statistically.
Given all this, a fundamental part of our leadership challenge is to help the world to distinguish between MQALA and the statistical method and to demonstrate optimal uses for both methods. This challenge involves both establishment issues and technical issues.
5.1.1. The Statistical Establishment
The statistical method is well established. To appreciate this establishment, it helps to distinguish two major components - statisticians themselves and people who defer too much to statisticians. Power is at stake.
Many statisticians are enshrined in departments of statistics and biostatistics as well as informatics programs. Collectively, statisticians have huge intellectual mass. The writings in some of their professional journals are arcane. The mathematical formulations can make ordinary people like me long for their native tongues. Perhaps the brightest minds can elaborate the most complicated “solutions.”
Despite all this intellectual mass, I doubt if any statistician can analyze a randomized multiple N-of-1 clinical trial as proposed in Appendix A with anything as simple as a single-group t-test on the mean - that is, without MQALA. This clinical trial example is important because this design may be the most productive and ethical experimental design to evaluate the benefit/harm of treatments for the management or control of chronic disorders. This design enables the new vision of integrated clinical research and practice. In addition, Demonstration 1 of Appendix B demonstrates use of the single group t-test to analyze values of a measure of hormone interaction from a sample of 6 ewes. And if MQALA makes statistical analyses so simple, how many investigators really need statisticians for this and many other related tasks involving mechanisms and benefit/harm - tasks that often are best addressed by developing data movies with DataSpeaks Interactions®?
Excellent data movies exist for capital markets. Do you see much evidence that statisticians do much better than the rest of us with their investment portfolios?
The demands and limitations of the statistical method itself raise additional important power issues. This can be illustrated in the context of clinical trials. The statistical method generally requires rather large samples of patients and study designs that can not optimize the care of the individual patients who become “subjects.” Such trials often require large organizational support, big budgets and, quite often, government regulation. This fosters big establishments. The statistical method has done little to help empower individuals with N-of-1 clinical trials, which are the gold standard for evidence based medicine (see, for example, http://www.cche.net/usersguides/applying.asp ).
Statisticians largely have a lock on what gets published and funded on matters involving empirical induction. They help make key decisions about approving drugs. These decisions can ultimately benefit us or harm us, kill us or save us. The power of statisticians may have served us well as we started to move from anecdote to science in disciplines such as medicine. Peer review by statisticians can continue to serve us well on issues of describing groups, sampling populations, and making inferences about populations as required to guide public policies. But statisticians should not be allowed to block other methods of empirical induction such as methods that can help explicate mechanisms of complex adaptive systems.
Statisticians have grown accustomed to their prerogatives. I learned that at least one noted statistician resented my patents that put information into the public domain without peer review. Patents primarily just have to be innovative and useful. We should expect some statisticians to resent the fact that I have published this document to the Web.
Although I am not a statistician, statistics has had a formative influence on my life. I taught statistics to undergraduates as a psychology graduate assistant. I worked with a prominent statistician at Dartmouth on a project to evaluate the Surveillance, Epidemiology and End Results (SEER) program of the National Cancer Institute. He threw my farewell party and went on to cofound the journal Statistics in Medicine. I took post-doctoral training in a mental health statistics program at the School of Public Health at the University of North Carolina. More recently, I helped teach quantitative methods to medical school residents. I have known statisticians as colleagues and friends. All this made it more difficult for me to recognize the statistical establishment as the primary competition for DataSpeaks, Inc.
I have presented MQALA to a number of statisticians. They see probabilities and some simple formulae that also are used in statistics. Statisticians conclude that MQALA is not good statistics. The statisticians are right, at least on one of these two counts. MQALA is good. But MQALA is not good statistics. MQALA is not statistics at all.
I have been referred to statistics textbooks. One pharmaceutical industry statistician ran extensive simulations with what came to be MQALA. We gave two presentations and the abstracts were published. These abstracts are cited in Patent 6,317,700. Although the simulations were supportive of the new methodology, the presentations and abstracts did not elicit much response. The statistician left industry and is pursuing other interests. I am grateful for his efforts.
My experience is that statisticians generally are bright and well intentioned people. I suspect that they resist MQALA primarily because it falls outside the lens of their experience. Simple solutions to major problems can elude leading minds for a long time. This has precedent. Here is a classic example.
Surgical death rates in a Viennese hospital apparently were over 50%. Ignaz Semmelweis and others advocated hand washing, a practice of particular value for thought leaders that did autopsies and went from autopsy rooms to surgical and delivery rooms without washing their hands. The simple hand washing solution was resisted, even ridiculed, apparently because it did not make sense. This was before the discovery of germs and the germ theory of disease. The value of hand washing was outside the lens of experience for leaders in power at the time.
Microscopes helped investigators see germs. DataSpeaks Interactions® will help investigators see interactions in a way that interactions never have been seen before. More than a century after the germ theory of disease, investigators still are trying to understand the interactions that describe germs - particularly how they act as agents to cause disease and respond to agents that might cure disease. Measuring and seeing interactions should speed progress in practical scientific understanding. But it will take time for investigators to appreciate what they see even after they see interactions measured.
Appendix B includes three demonstrations that show how interactions are measured with DataSpeaks Interactions®.
Statisticians have assumed the mantle of responsibility for protecting large swaths of the scientific community from error. But they view their responsibility through a narrow lens. As a result and despite providing valuable services, statisticians have become a major impediment to progress.
5.1.1.2. Deference toward Statisticians
The second major component of the statistical establishment is the excessive deference that most people show toward statisticians. The combination of the obscurity of their arcane writings, daunting difficulty of the mathematical formulations and power of statisticians over the lives of scientists and other people seems to inspire awe and deference. Statisticians can seem to get away from addressing their supplicants with quiet inscrutability.
Leaders in various fields have deferred me to statisticians almost immediately as soon as they understand that my work has something to do with analyzing data. At times I feel dismissed because my Ph.D. is not in statistics. I’ve had successful people throw up their hands and say “you are way above me” or “we are not in the same ballpark” almost as soon my technology reminds them of statistics. Such apparent manifestations of deference have happened for a method, MQALA, which is basically as simple as defining discrete events and using 2 x 2 contingency tables to measure interactions or temporal contingencies (See Appendix A). Some experts who work with data seem to have found this simplicity insulting, off-putting, or below them. The ruling presumption seems to be that longstanding problems must have difficult solutions. But simple solutions can be best, even if they require changing professional habits.
Some people seem to assume that the best way to solve longstanding and important problems involving the use of data must be to do more of what statisticians have been doing for decades. This appears to be part of why some life science centers are turning more to higher mathematics. Although higher mathematics often is helpful, the more immediate and simple solution to large and important classes of unsolved problems appears to involve the computation of measures of interaction and temporal contingency before statistical analyses. Appendix A describes this simple solution in the context of measuring benefit/harm in clinical trials. Appendix B demonstrates this simple solution with an interaction between hormones and connectivity in the brain. These appendices also illustrate how measurement of interactions first can simplify mathematical and statistical treatment of data.
Excessive deference toward statisticians has been costly. Great minds that invent microarrays, functional imaging machines, and Web enabled monitoring devices often defer or delegate to statisticians for data processing issues outside of describing groups, sampling individuals, and making inferences from samples to populations. As a result, much of the value of their discoveries and inventions has yet to be realized.
The statistical method became established before MQALA was invented. My impression is that either method could have been established first. If MQALA had been first, statisticians might be trying to break into the MQALA establishment.
This might explain why the statistical method was established first. People have been known to seek certainty and absolute truth. Scientists have been known to seek the immutable laws of nature in accord with deterministic worldviews. Statistics in this context might be viewed as an attempt to separate mathematical truth from error. This might be the basis for much of the lofty position afforded to statisticians. In contrast, MQALA just measures temporal contingencies that describe the mechanisms by which complex systems work, change, and adapt. MQALA also appears to support a worldview that can hold individuals accountable as responsible agents that deserve to be rewarded, punished, and honored.
Perhaps some leaders are tired of deferring and delegating their crown jewels to statisticians. Besides, deference to authority is not very scientific.
5.1.2. MQALA and the Statistical Method - Technical Differentiation
Given apparent difficulties in distinguishing different computational methods of empirical induction and their varied uses, it may help to differentiate methods. Although there are more than two methods that are relevant to empirical induction, this section focuses on MQALA and its primary competition, the statistical method or statistics.
The differentiation between MQALA and the statistical method is not so much a matter of strengths and weaknesses as a matter of different methods for different types of problems and different types of data. I present these comparisons and differences in bold broad strokes. These statements call for additional investigation. Many of these statements make similar points in different ways.
Statistics is about groups and populations. MQALA is about individual systems. Individual systems include populations investigated as entire populations.
Statistics works best for time-invariant systems. MQALA can be used to measure how interactions change as systems change over time.
Being largely timeless, statistics struggles to account for mechanisms of internalfunction, response, agency, and adaptation. By accounting for time, MQALA is well suited to account for the mechanisms of function, response, agency, and adaptation as these were distinguished above.
Statistical measures of correlation were designed for and work best for cross-sectional data. Measures obtained from MQALA were designed for and work best for multiple time series data.
Statistics works best for linear relationships. MQALA appears to work for both linear and nonlinear relationships. MQALA informs users about the form of relationships between measures of interaction and various analysis parameters.
Statistics uses “interaction” to describe non-additive relationships between independent variables in time invariant systems. MQALA uses “interaction” to describe relationships between one or more independent variables and one or more dependent variables in systems that can change over time. I sometimes use “dynamic interaction” to distinguish the way MQALA uses “interaction” from the way that statistics uses “interaction.”
Statistics works best with homogeneous groups - an important reason why clones and inbred strains of organisms are highly prized from a methodological perspective. In addition, MQALA works well for individuals that may be unique and can be investigated over time.
Statistics works best for hypothesis driven science. MQALA facilitates data driven discovery science. MQALA also expands the scope of hypothesis driven science by enabling statistical tests of measured dynamic interactions.
Statistics works best when there are few variables. Thus, there is much emphasis on variable reduction by investigators of complex systems. MQALA works well for many variables, although demand for computational resources can increase rapidly. In addition, MQALA can be used to help accomplish variable reduction. Appendix A describes an example that reduced one independent variable, drug dose, and 20 dependent health variables to one variable that quantified apparent benefit/harm.
Statistics works best for analyses involving systems’ parts. MQALA works well for analyses of systems’ parts and syntheses of how many parts work together to form more or less coordinated systems.
Statistics works best for cross-sectional data. Thus, for example, clinical trials often are analyzed with pre- and post-treatment difference scores in health variables. This practice effectively squeezes out time to form a single timeless number for each patient. MQALA works for longitudinal data with two or more variables, preferably time series data. MQALA was not intended to work at all for cross-sectional data.
Statistics works best with experimental designs for groups. MQALA works with time series experimental designs for individuals. Often it is productive to combine both types of design.
Prevailing inferential statistical methods work best for rejecting null hypotheses. MQALA measures the amount and strength of evidence for positive and negative interactions over time between independent and dependent variables. Although MQALA uses contingency tables and hypergeometric probabilities to compute values of measures, the measurement process by itself does not test hypotheses.
Statistics works best when individuals are randomized to different treatment groups and treatments remain fixed for each individual. In contrast with MQALA, levels of independent variables must change over time for individuals in order to get nonzero values for measures of dynamic interaction or temporal contingency. In other words, the statistical method often works with categorical independent variables while the independent variables for MQALA can be dimensional variables with two or more levels that must change over time to get non-zero interaction scores.
Both MQALA and statistics are most apt to yield valid results with experimental data. Both methods can be used with non-experimental data. MQALA does a better job in evaluating the temporal criterion of causal and other predictive interactions in non-experimental data.
MQALA is a measurement system that yields scores and values of measures. Statistics is a method and system for analyzing the values of measures.
Statistics often works with dimensional variables as dimensional variables. In contrast, MQALA must convert series of values for dimensional variables into sets of series for discrete events that can be either present or absent on most measurement occasions. This has been described before and in Appendix A.
Statistics often uses measures of central tendency and variability that work best for groups. MQALA computes measures of longitudinal association, interaction, or temporal contingency for individuals.
MQALA works well in data mining for interactions in time series data with two or more variables when the data are about individuals. The statistical method works best in mining essentially timeless data about multiple individuals.
The statistical method can be described as being good for developing data snapshots of groups as in a census. MQALA can be described as being good for developing data movies of individuals including entire populations.
In summary, the statistical method as currently used often seeks to separate truth from error in a timeless and changeless world of immutable laws where individuality is not important. This is not the world of complex adaptive systems. MQALA accounts for individuality and time in a real world that changes over time, where temporal contingencies help shape complex adaptive systems, and where individuals can be held accountable as responsible agents.
MQALA and the statistical method are distinct and often complementary. The two methods often should be integrated to help make statistics relevant when science also needs to account for time and individuality in a world that includes complex adaptive systems.
5.2. Stephen Wolfram and Forrest Gump
I will hazard to venture a few comments about another potential competitor or alternative to MQALA in the space occupied by computational methods relevant to understanding nature - namely cellular automata and the Principle of Computational Equivalence as enunciated by Stephen Wolfram.
Stephen Wolfram - genius, entrepreneur, and primary creator of Mathematica, apparently the world’s leading scientific software system for technical computing and symbolic programming - authored A New Kind of Science. The title of this tome, the high status of its author, and the phenomenon generated by publication of this book suggest some recognition of need to advance scientific methodology.
Wolfram’s book includes nearly a thousand original pictures that allow, as described on the book’s dust jacket, “scientists and non-scientists alike to participate in what promises to be a major intellectual revolution.” Many of these pictures show patterns generated in accord with rules that determine whether or not particular cells should be black or white. Similarly, MQALA deals with discrete events that are considered to be either present or absent as described in Appendix A. Perhaps there is some deep connection between MQALA and cellular automata as the latter is enunciated by Wolfram. I invite such explorations.
Many of Wolfram’s pictures show patterns formed when the consequences of rules or programs unfold over steps. Some of these patterns are strikingly similar to patterns observed in systems found in nature. Some patterns reveal randomness. Simple programs can produce complexity.
It appears as if these rules or programs are being presented as alternatives to immutable laws of nature as sought by some more conventional scientists. If so, the scientific quest according to Wolfram would appear to be discovery of rules or programs that account for the unfolding of nature. It appears as if such discoveries would occur primarily in the heads of scientists.
In contrast, MQALA does not start with, depend on, or assume that there are any rules, programs, theories or immutable laws of nature. MQALA starts with time ordered data and asks the data to speak through software that reveals patterns. MQALA uses computation to measure temporal contingencies to facilitate understanding of complex adaptive systems. Perhaps randomness such as that found by Wolfram helps make contingencies interesting because contingencies matter.
Perhaps the essence in the world of complex adaptive systems is less like either rules or immutable laws of nature that merely unfold over time in a determined and sometimes random way and more like Forrest Gump’s saying, “Shit happens.” “Life is like a box of chocolates: You never know what you’re going to get.”
“Shit happens” seems to resonate with many people. A recent Google search on this phrase yielded about 70,100 hits, heavily represented by computer programmers characterizing different religions. DataSpeaks Interactions® can be said to measure how shit happens with some regularity. This rather crude description might be welcomed by those who are tired of deferring to the power of the statistical establishment. It can bring some fresh air to what some people may see as the stuffy world of scientific formalisms.
Shit happened to me as I discovered MQALA step by step. Shit happened when many other scientific discoveries resulted from serendipity. Perhaps I deserve more credit for persistence and self-confidence in pursuing MQALA than for discovering MQALA itself.
Perhaps it is time for us as a people to escape the fate of determinism and the hopelessness of pure randomness by accepting the awesome responsibility of measuring the temporal contingencies of nature and creating the future in accord with this knowledge and a respect for fundamental human values.
Of the two methods, MQALA and cellular automata, MQALA appears to be closer to data as it is being and can be collected now. In contrast to being a new kind of science, MQALA can be described as back-to-basics science where the basics are measurement, data, and experimental control. As such, MQALA shows more promise for getting us out of data swamps.
The sections on scientific worldviews and responsible agency provide more information related to Wolfram.
5.3. Departments of Empirical Induction
Different methods and algorithms for empirical induction work best for different types of problems and with different types of data. It will be increasingly important, given this growing diversity, to make relevant and useful distinctions before acting.
One way to foster intelligent use of different methods for drawing generalized conclusions and making predictions from data might be, as suggested in Patent 6,317,700, to establish departments of empirical induction where experts in MQALA, the statistical method, neural networks, genetic algorithms, cellular automata, chaos theory, etc. can thrash it out, foster new types of study designs, develop intellectual mass, educate leaders, develop mathematical models and theories of real systems, develop artificial systems, and advance science and engineering. Such departments would have the potential to become new “centers of calculation.”
Departments of empirical induction could replace departments of statistics and biostatistics. The new name would be less prejudicial in favor of a particular method that happened to be established first. The new name would suggest openness to new possibilities. The new name would help users recognize that they need to make relevant choices. The new departments would help educate people who could keep up with and lead new and emerging ways of doing science. Establishment of new departments of empirical induction calls for strong leadership as described in the Call for Leadership and discussed in the section on leadership.
The time is right for the advancement of MQALA as embodied in DataSpeaks Interactions® software.
Interactions are widely discussed, even in our most prestigious scientific journals. But they never seem to be effectively measured over time for individuals. There is much surprisingly loose talk and writing about protein-protein interactions, drug interactions, environmental interactions and all sorts of interactions. A fundamental problem is that the concept of interaction is not given a common, clear, specific, and objective scientific meaning, which comes from concrete operational definitions and measurement procedures that can be applied in many disciplines.
MQALA, as embodied in DataSpeaks Interactions®, provides the required operational definitions. DataSpeaks Interactions® actually measures interactions so that they can be investigated scientifically. The same measures can be used to help solve practical many problems.
MQALA is a measurement system. Users can select many options as they develop scoring protocols that are appropriate for specific investigations and problems. No one scoring protocol is appropriate for all data sets and problems. Appendix A presents some of the scoring options. Appendix B illustrates several specific scoring protocols.
The time is right in terms of the “omics” of biological science - genomics, transcriptomics, proteomics, metabolomics, physiomics, etc. DataSpeaks Interactions® can help add time, function, and individuality to the omics by going beyond identification and characterization of substances to describe how substances help form systems that work, change, and adapt.
The time is right in terms of shifting paradigms of science and growing ferment about scientific methods. I have already described an apparent shift from hypothesis driven science to data driven discovery science. I have also commented about Wolfram’s book, A New Kind of Science. MQALA offers hope to bridge the fault lines that are developing among different conceptions of science. MQALA can help build toward a new consensus about what science is and does with respect to investigations of complex adaptive systems.
The time is right in that many leaders perceive great opportunities to be where different disciplines such as biology, medicine, and computer science converge. They are right. Leaders speak of collaboration, interdisciplinary studies, integration, and convergence. MQALA and DataSpeaks Interactions® provide a common methodology that can be applied to investigate how complex systems of many types work, change, and adapt.
In addition, DataSpeaks Interactions® can be used to investigate interactions between and among variables normally considered to be subjects of different disciplines. One example is the neural control of behavior. As such DataSpeaks Interactions® can foster interdisciplinary and collaborative investigations and take advantage of these opportunities.
A common methodology is a great unifier. The market opportunity sections indicate how DataSpeaks Interactions® can be applied to many problems. As such, it can help unify many disciplines. The time is right in terms of institutional responses to the ferment in science and the feeling that science, despite all its breakthroughs, is not living up to its potential to improve human welfare. As examples, the United States National Institutes of Health recently announced the NIH Roadmap for Medical Research (http://www.nih.gov/news/pr/sep2003/od-30.htm). “With this theme, New Pathways to Discovery, the NIH Roadmap addresses the need to understand complex biological systems. Future progress in medicine will require quantitative knowledge about the many interconnected networks of molecules that comprise cells and tissues, along with improved insights into how these networks are regulated and interact with each other. Researchers predict that more precise knowledge of the combination of molecular events that lead to health or disease will help to revolutionize the practice of medicine in the 21st century.” “New Pathways to Discovery also sets out to build a better “toolbox” for today’s biomedical researchers.” DataSpeaks Interactions® has the potential to be the primary breakthrough in such a toolbox. Notably, the NIH announcement does not seem to anticipate actually measuring the interactions that help define biological systems and help describe how patients respond to their environments, including treatments. The Food and Drug Administration (FDA) of the United States also is demonstrating institutional response to changing imperatives by issuing draft guidelines for personalized medicine (http://www.fda.gov/bbs/topics/NEWS/2003/NEW00969.html). These are intended to provide guidance on how “to individualize therapy by predicting which individuals have a greater chance of benefit or risk -- thus helping to maximize the effectiveness and safety of drugs. FDA believes that pharmacogenomic testing can be smoothly integrated into drug development processes.”
“This is FDA’s first step towards integration of this new field into the process of demonstrating that new drugs are safe and effective…” FDA is beginning to recognize the importance of individuality. DataSpeaks Interactions® measures interactions, including the benefit/harm of treatments, for individuals.
The Grand Challenges in Global Health initiative is another response to changing imperatives (http://www.grandchallengesgh.org/ArDisplay.aspx?ID=29&SecID=302). The Foundation for the National Institutes of Health (FNIH) and the Bill & Melinda Gates Foundation sought to identify and fund “proposals for research on these critical scientific and technological problems that, if solved, could lead to important advances against diseases of the developing world.” Given that one of the Founders of Microsoft is primarily responsible for the impetus and sponsorship of this Grand Challenge, it is both ironic and a rare opportunity that the key to health improvement is better software, namely DataSpeaks Interactions®.
My response to the “Call for Ideas” of The Grand Challenges in Global Health initiative was written with knowledge about how to meet this challenging idea.
The time is right when The New York Times recently identified “Does Science Matter?” as number 1 in a list of 25 of the most provocative questions facing science (http://www.nytimes.com/2003/11/11/science/11MATT.html). There are serious challenges to science throughout much of the world. Perhaps it is time for a methodology of science that accounts for the temporal contingencies that describe and shape complex adaptive systems and appears to support a worldview that can hold individuals accountable as responsible agents.
Some leaders question the importance of information technology when they ask, “Does IT matter?” (http://itmatters.weblog.gartner.com/weblog/index.php?blogid=10). At least one major pharmaceutical research and development facility is cutting back on IT. Perhaps IT needs a new big idea to help regain relevance.
MQALA will help make science and IT matter because contingencies matter as mentioned in the Call for Leadership and described more fully in the responsible agency section.
The time is right in terms of political need. The United States and various countries appear to be highly polarized on many great issues of our day including economic policy, environmental policy and the impact of our foreign and military policies at home and abroad. Much of the vitriol that may be tearing us apart can be traced to our failures to understand complex adaptive systems. Science is not keeping up with demands for answers as stakes increase. Political platforms end up taking positions on issues that no one really understands. People get blamed rather than ignorance. Many leaders can be held accountable for not doing more to advance scientific understanding.
New scientific methods that work to help people understand systems and to create systems that work to achieve generally accepted goals could help restore our trust in intelligence, rationality, and scientific evidence.
The time is right for additional reasons that will be presented in the context of eight selected market opportunities.
6. Eight Selected Market Opportunities
Development of DataSpeaks, Inc. as a business appears to be the most efficient, effective, and fastest way to deliver the benefits of DataSpeaks Interactions® to people despite protestations I have experienced from academe.
DataSpeaks Interactions® may be a bigger market opportunity than the Web browser. Creation and application of scientific understanding with a computational method of empirical induction may be a bigger opportunity than browsing in a world that still needs software that accounts for mechanisms, individuality, and time. In addition, DataSpeaks Interactions® will drive collection of data movies as well as adaptation and emergence in a world with an expanding horizon of possibilities.
Here are eight selected and overlapping market opportunities for DataSpeaks Interactions®.
All of these opportunities can be considered as different facets of one large opportunity in the software market.
One aspect of our business development strategy could be to create an avalanche effect through synergy among different markets commensurate with the leadership and resources that can be brought to bear on the advancement of DataSpeaks Interactions®.
In addition to these eight market opportunities, DataSpeaks Interactions® has the potential to drive demand for computing infrastructure and many data collection technologies. Companies with such technologies should help advance DataSpeaks Interactions® to increase demand for their own products and services. Internet2 (http://www.internet2.edu/) also could help enable DataSpeaks Interactions®. DataSpeaks Interactions® is demanding of computational resources.
DataSpeaks Interactions® could be developed as Excel add-ins or Excel add-ons. Such versions probably would have limited functionality but could help seed the market.
All eight specific market opportunities can improve economic productivity and human welfare. Each opportunity has its pros and cons. Early adopters in all markets could achieve huge competitive advantages. Of the eight market opportunities, I recommend that Visualizing How Brains Work, Change, and Adapt should be pursued first.
6.1. Revitalizing the Pharmaceutical Industry
DataSpeaks Interactions® can help revitalize both drug discovery and clinical research.
Productivity of the pharmaceutical industry as measured by approvals of new chemical entities has been declining despite huge and rapidly growing research and development budgets, development of amazing new data collection technologies, combinatorial chemistry, and decoding of various genomes. The decline of productivity in this huge industry is illustrated at http://www.technologyreview.com/articles/hall1003.asp , which also links to my response in the forum.
Patent 6,317,700 identifies over 20 ways that MQALA can help revitalize drug discovery and development. The current emphasis on mergers, acquisitions, partnerships, organization of work groups, design of collaborative work spaces, and blockbuster drugs might not be sufficient for a long term revitalization of the pharmaceutical industry. The pharmaceutical industry requires fundamental innovation and a change in culture.
DataSpeaks Interactions® is an option that can empower those who are affected by the pharmaceutical industry to revitalize the industry and improve the cost-effectiveness of drugs and drug research.
6.1.1. Revitalizing Drug Discovery
DataSpeaks Interactions® has the potential to revitalize drug discovery by actually measuring the interactions that describe mechanisms of health and many functional disorders as well as mechanisms of action. Mechanisms of action include mechanisms of new and established treatment agents as well as mechanisms of agents such as germs and allergens that can cause health problems.
Demonstration 1 of Appendix B is a quantitative description, obtained with MQALA, of the mechanism by which two hormones interact. Claims in Patent 6,516,288 cover ways that MQALA can be used to describe mechanisms in which hundreds or thousands of variables can interact in a more or less coordinated manner over time. Specific disorders of coordination appear to be diagnostic of specific health disorders.
Actual measurement of interactions that describe mechanisms of health and disorder is a fundamental change from current practice in diagnosing many health disorders. The status quo in medical diagnosis largely is restricted to measuring high or low levels of different variables such as blood pressure, cell counts, mental depression, hormone levels, glucose levels, and various lipid fractions.
Diagnosis by measurement of high or low levels of various health variables has been helpful. But such measurement is only a beginning because levels alone say so little about what specific disorder’s are, how patients should be treated, what drugs and other agents do, and what new drugs must do and not do in order to be safe and effective.
DataSpeaks Interactions® helps enable a new strategy for medical diagnosis. The new strategy for medical diagnosis can be described as moving beyond levels of action to measures of interaction that describe mechanisms.
Some problems of diagnosing health disorders in terms of high and low levels of health variables will be illustrated in the context of hypertension. One aspect of the problem is that there probably are about as many different types of hypertension as there are different mechanisms that can produce high blood pressure. Apparently thousands of endogenous and exogenous variables can affect blood pressure. Merely knowing that blood pressure is high says little about what causes it to be high and what should be done about high blood pressure. Diagnoses need to be more specific in order to identify genetic predictors and individualize treatments as well as to target drug discovery and development more effectively.
Another aspect of the problem of diagnosing by levels of health variables can be illustrated with a simple case in which there are only two interactants. Many combinations of levels of two interactants can yield a more or less effective interaction much as many pairs of values (integers and fractions) can yield a particular mathematical product such as 24. This suggests that levels of particular individual interactants at particular times say little about the functional integrity of coordinated biological systems. This fundamental problem of diagnosis by levels becomes worse when there are many interactants.
I will illustrate some aspects of the failure to actually measure interactions that describe mechanisms of health and disorder by reference to aspects of my personal work experience. Many readers might be able to relate to these experiences.
I once worked with a psychiatrist who essentially said that my job was to produce statistically significant results - any results, including significant correlations between values of variables that were measured repeatedly for some subjects. Then he would explain the results in terms of interactions between endogenous substances in nervous systems - interactions that could be up or down regulated by treatments. Then we could publish the results together to advance our careers. This could be described as a publication questing through significance questing scheme.
Failure to actually measure dynamic interactions between endogenous substances and how these interactions may be affected by treatments opens doors to pure speculation in science. The psychiatrist I worked with became infuriated when I resisted his scheme. He diagnosed me as having “flat affect.”
Similarly, dysregulation theories of various disorders were popular, at least in the 1980s when I worked in psychiatry. I tried to convince investigators to actually measure the interactions that could put their theories to the test. Citations in Patent 6,317,700 document some of my unsuccessful efforts.
Statistical review could be expected to help reduce many such abuses. Given the volume of published literature that manifests such abuses and untested theories, my impression is that review has not been very successful. Talk of interactions and their up and down regulation suggests pent up demand for actually measuring interactions. But statisticians are not prepared to help meet this demand or control these abuses because actual measurement of interactions that describe biological mechanisms appears to be outside the lens of their experience. Abuses permitted by failure to measure interactions could be overcome by establishing departments of empirical induction.
DataSpeaks Interactions® can make it easier to identify mechanisms of treatment by actually measuring ordered or disordered mechanisms and how these mechanisms are affected by treatments. Demonstration 1 illustrates the measurement of interactions in the context of reproductive endocrinology. In addition, DataSpeaks Interactions® could be used to measure how interactions change upon administration of drugs that might affect hormones or block or sensitize hormone receptors.
Here is another example of measuring mechanisms of drug action. The section about functional brain image analysis and Demonstration 3 describe how DataSpeaks Interactions® can be used to measure functional connectivity between and among brain regions. Disordered interactions could be diagnostic of many functional brain disorders. In addition, interactions could be measured before and after drug administration to investigate how drugs may affect functional connectivity involving thousands of brain regions simultaneously. This illustrates one potential high throughput method for investigating drug effects, a method that appears to remain essentially untested.
DataSpeaks Interactions® can make drug target discovery and validation more efficient. It can help squeeze much of the mystery and unpredictability out of drug discovery and development.
6.1.2. Re-engineering Clinical Research
Re-engineering the clinical research enterprise is an important part of the NIH Roadmap (http://nihroadmap.nih.gov/clinicalresearch/index.asp). “Clinical research is the linchpin of the nation’s biomedical research enterprise.” The boundaries between NIH clinical research and pharmaceutical industry clinical research appear to be blurring. Both need to be re-engineered.
Efforts to re-engineer clinical research are laudable. However, infrastructure improvements called for in the current NIH plan show little or no evidence of recognizing how fundamental the re-engineering effort needs to be in order to achieve substantial progress.
Clinical research still is rather primitive despite all the trappings to the contrary. Progress is being made. But this progress in evaluating treatment effects and in translating research results into clinical practice is unnecessarily slow and expensive. Progress so far is just an inkling of what we will be able to achieve after clinical investigators start measuring the benefit/harm of treatments over time and across variables for individual patients before any statistical tests. Appendix A, Patent 6,317,700 and one of my reprints illustrate the actual measurement of benefit/harm.
A primary function of clinical trials is to evaluate the safety and efficacy of treatments. This is being done without measuring the benefit/harm of treatments as interactions between treatment and health variables for individual patients. Statistical tests are being performed on health variables, not benefit/harm scores. This critical distinction is not often made.
People who are not experts in clinical research and who defer to the experts often seem surprised to hear that clinical trials do not measure and test the benefit and harm of treatments. In contrast, experts in clinical research seem to find it difficult to believe that there is any alternative to performing statistical tests on health variables broadly defined. These unfortunate facts testify to the awesome power and influence of the statistical establishment.
These points are true despite the fact that benefit/harm scoring has ancient roots. Survival often depends on temporal contingencies. Animals learn from temporal contingencies. People and other organisms have learned to avoid encounters with poisonous plants and dangerous animals most directly from the temporal contingencies of encountering them. People were learning how to take care of themselves and each other long before the first randomized controlled group clinical trial. Clinicians learn and gain insights from the temporal contingencies of providing care to individual patients. Patients sometimes disobey doctors’ orders because of the unpleasant contingencies of treatment. Such learning is described briefly in the behavior modification section.
Benefit/harm scoring measures the health related temporal contingencies of treatment. Evidence from temporal contingencies is distinct from and often complementary to evidence from group comparisons. But evidence based on temporal contingencies, which would account for individuality and time, is barely used in most scientific evaluations of treatments.
One formative influence on what came to be MQALA was my interest in algorithms that used response to drug challenge, de-challenge, and re-challenge to help evaluate evidence for adverse drug reactions. This evidence illustrates temporal contingencies. Occasionally clinicians are held accountable for harming patients or continuing expensive ineffective treatments if clinicians do not take reasonable steps to monitor how individual patients respond to treatments even if the treatments were used in accord with treatment guidelines based on group clinical trials. This suggests some primacy of evidence based on temporal contingencies over evidence based on group comparisons. In addition, N-of-1 clinical trials, which produce evidence essentially based on temporal contingencies, have been identified as the gold standard for evidence based medicine.
Benefit/harm scoring with DataSpeaks Interactions® relies on the same type of evidence as the old drug challenge, de-challenge, and re-challenge algorithms. But it develops this evidence with many additional steps. As examples of additional steps, MQALA evaluates the same type of evidence by computation from data rather than by subjective impressions in peoples’ heads. MQALA measures both harmful and beneficial effects. Treatment does not have to be merely present or absent but can vary in dose over time. MQALA can account for effects on tens or hundreds of health variables simultaneously. In addition, MQALA can account for drug-drug interactions, temporal parameters that affect interactions between treatment and health, and how evidence for benefit/harm varies over time when, as examples, patients adapt to or are sensitized to the effects of treatments.
So what does the discovery of MQALA and the invention of DataSpeaks Interactions® mean in terms of re-engineering clinical research? Here are some initial impressions together with some information about why these changes would be valuable. I present this in the context of using drugs, which can vary in dose over time, for the management or control of chronic disorders. This is an important part of clinical research but not the only part.
One implication of MQALA is that conventional parallel group clinical trials generally will become unethical in addition to being needlessly expensive, hugely inefficient, and largely inconclusive about major issues. The new alternative would be randomized multiple N-of-1 clinical trials with one or more groups. A multiple N-of-1 clinical trial consists of a coordinated set of N-of-1 clinical trials. DataSpeaks Interactions® would be used to measure apparent benefit/harm over time and across one or more health variables. The statistical method would be used to test the null hypothesis of no benefit/harm in either single group or multiple group designs.
Both single and multiple group multiple N-of-1 designs would randomize doses, which could include zero dose as placebo, to time periods. Multiple group designs also could include randomization of patients to different treatment groups. In general, different doses of the same drug would be evaluated both in individual patients and single groups of patients in a manner analogous to the way drugs can be titrated in clinical practice.
DataSpeaks Interactions® can be applied to data from many conventional clinical trials. This would be a good way to become familiar with the new methodology and mine old trials for new understanding and insight. However, the value of reanalyzing conventional clinical trial data is limited because conventional clinical trials are not designed to provide reliable and valid measures of treatment effect or benefit/harm for individual patients. For example, conventional designs do not distinguish placebo responders from true responders to investigational agents. This avoidable failure makes it difficult to target drug development to patients most apt to benefit and away from patients most apt to be harmed. For example, conventional clinical trial designs make it difficult to capitalize on the success of the Human Genome Project. Conventional trial designs make it difficult to find genetic predictors of differential response because conventional trials do not provide reliable, valid, and specific measures of benefit/harm for individual patients.
Conventional clinical trials that test health variables are largely inconclusive with respect to major questions such as whether or not a drug should be approved for marketing. Part of this problem derives from the fact that most treatments affect more than one health variable and that tests on many health variables in single trials create problems involving the management of statistical significance levels. This limitation creates huge problems in selecting health variables for inclusion in trials and selection of variables for primary tests in particular trials that gather data on more than one health variable. This limitation, together with a lack of computational procedures for combining results from many clinical trials that test different health variables, means that drug approval might be more a matter of social consensus than scientific test. Since social consensus is subject to the effects of power and influence, approval or disapproval of a drug with a given benefit/harm profile is not very scientific and predictable.
An alternative approach, based on MQALA, was proposed in Appendix A. Appendix A illustrates how the results of a randomized multiple N-of-1 clinical trial with various doses and 20 health variables could be analyzed with a single group t-test on mean overall benefit/harm scores, one such score from each of the 20 individuals in a group. Similar use of the t-test is shown in Demonstration 1 of Appendix B.
Here is another facet of the re-engineering effort that would be made possible with MQALA. MQALA would allow users to separate determination of treatment effects from how patients and clinicians value these effects. Thus, for example, clinical trial results could be made available on an interactive Web page where patients with their clinicians could decide on how they value the different effects of particular treatments. For example, some drugs cause impotence. A particular patient could decide if this is a beneficial or harmful effect and how important this effect is compared to effects on other health variables. After a patient and clinician selects a set of weights and directions, he could click to see if the drug is apt to be beneficial or harmful to him based on his own preferences and weights. Similarly, two or more drugs could be compared. This also would help personalize or individualize treatment.
Other facets of the re-engineering effort would involve packaging or otherwise providing medications to facilitate randomized N-of-1 clinical trials, preferably with multiple doses, and the collection of the required data movies, preferably over the Web. Information about planned doses could be supplemented with data about actual doses and levels of drug or drug metabolite levels in bodily fluids. Health variables could include information about variables usually collected in laboratories, symptom rating scales for variables often used to collect data about safety and efficacy, computerized measures of physical and mental performance, and quality of life measures. MQALA can be used to relate treatment effects on different levels of drug effect hierarchies.
Clinical research was largely set asunder from clinical practice decades ago with the advent of conventional randomized group clinical trials. Many of the effects were salutary. But this largely created problems of separate budgets for research and practice and of translating the results of clinical research into clinical practice. The solution that needs to be re-engineered is to integrate clinical research with clinical practice. MQALA would help make this possible because best practices for clinical research would be essentially identical to best practices for providing health care. But this also calls for re-engineering health care systems.
6.1.3. Competing Visions for Clinical Research and Practice
Discovery of a computational method to measure apparent benefit/harm of treatments over time and across variables for individual patients might be the primary innovation that needs to be considered while re-engineering much of clinical research and practice. To illustrate, I will use broad strokes and statements made from a point of view to paint the old and the new competing vision about how clinical research can be conducted. Both visions assume that we have promising drug candidates with potential value for the management or control of chronic health disorders. We start with the first clinical trials for men and women.
6.1.3.1. The Old Vision for Clinical Research and Practice
The old vision essentially requires us to start by making some slightly educated guesses about doses and indications. Major decisions have to be made when there is the least amount of clinical data. Generally, relatively narrow groups of subjects are targeted with rather small sets of predetermined doses. Early mistakes mean that the drug is targeted to patients that will not benefit, to patients that will be harmed, and away from some patients that could benefit. In addition, the doses could be wrong. Such mistakes have killed many drug development programs. Since the old vision does not include procedures to collect reliable and valid measures of benefit/harm for any patient, it is difficult to improve targeting of drug development and patient care. Similarly, since the old vision does not include procedures to find optimal doses for individual patients, it is difficult to improve dosing.
Typical clinical trials collect data on more than one health variable and for more than two time points. However, it is best to have one primary statistical test per clinical trial. When the statistical test is performed on a health variable, this means that data about health variables other than the primary health variable is underutilized. Similarly, data on more than two time points often largely goes to waste because change scores generally are based on only two repeated measurements, a baseline and an endpoint. Failure to measure benefit/harm fosters disputes and controversy about selection of health variables and time points.
One type of controversy involves the type of health variables for primary statistical tests. Many clinicians seem to favor objective laboratory measures, which are quite specific and often favor specialties that clinicians happen to practice. Patients, and often employers and governments that often help pay for health services, may have difficulty appreciating the import of specific laboratory measures. Many patients, some clinicians, and some payers might favor more subjective health status measures such as the SF-36 Health Survey that are more comprehensive and can be used to help evaluate the relative value of different treatments, including treatments for different disorders. In addition, prevailing analytic methods make it difficult to evaluate relationships between the laboratory measures and health status assessments.
The old vision often discourages systematic assessment of treatment emergent events for several reasons. Systematic assessment together with conventional methods of data analysis usually target adverse events compared to unanticipated beneficial effects. Systematic assessment can increase reporting rates of adverse events, thereby making drugs look bad compared to drugs evaluated with more haphazard voluntary reports. Haphazard reports are subject to many uncontrolled factors that increase variability of reporting and make it difficult to detect drug effects. Haphazard reporting has been favored by some to avoid learning about adverse effects of drugs. This puts patients at risk and makes financial loses mount when drug development projects are terminated late and when drugs have to be withdrawn from the market.
The old vision amasses information in small increments that are difficult to integrate. Particular trials might help answer particular questions such as whether or not a particular dose of a particular drug is better than placebo with respect to a particular health variable over a particular period of time defined by baseline and endpoint assessments. This specificity makes it difficult to make broad decisions about approval of drugs for marketing and use of drugs for particular patients. Broad decisions need to be based on more comprehensive and realistic evaluations.
Since conventional clinical trials are not designed to optimize treatment of subjects that participate, the trials often have to be paid for with large research budgets that are separated from health care budgets. In addition, the ethics of such trials often are challenged for good reason. Furthermore, the old vision fosters large regulatory agencies that tend to pass judgment on substantive rather than procedural issues and abridge the rights of individual patients that could benefit from new drugs.
The old vision almost guarantees conflict and failure at great expense, currently over $800,000,000 for approval of each new chemical entity.
After approval, new drugs enter the less controlled world of clinical practice where there is considerable chance that new drugs will be recalled, that the responsible companies will be subject to liability, and that the regulatory agencies that may be blamed will respond with fewer approvals and more expensive drug approval requirements.
Approvals for marketing generally unleash expensive marketing blitzes that are subject to abuse, do little to improve disease management, and may fail to collect additional information from actual clinical practice that can expand scientific knowledge about drugs and optimize treatments for individual patients.
The old vision supports a number of large establishments and leaves executives pleading to an angry, disillusioned, and suffering public for patience, understanding, high drug prices, and protection of the status quo. Some recognize that “The patient is waiting.” But the old vision - the standard operating procedure - helps assure that patients will continue to wait.
The status of evaluating safety and efficacy without measuring the benefit/harm of treatments for individual patients is similar to the status of investigations of infectious diseases before the discovery of germs.
6.1.3.2. A New Vision for Clinical Research and Practice
This new vision would make clinical research more scientific, productive, cost-effective, and ethical. The key to this vision is the actual measurement of the benefit/harm of treatments as interactions over time between measures of treatment and measures of health for individual patients. This vision would integrate the best clinical research with the best clinical practice. The measurement of benefit/harm would be accomplished by DataSpeaks Interactions®.
The new vision would make clinical research both more standardized and adaptable. Standardized trials can be more adaptable when the design includes fixed contingencies based on objective measures of benefit/harm. It is fitting for the new vision to be more adaptable than the old vision because the former is based on MQALA, a methodology for investigating how complex systems work, change, and adapt.
MQALA helps enable N-of-1 clinical trials. This includes coordinated sets of N-of-1 clinical trials or multiple N-of-1 clinical trials. I predict that such clinical trials will become the gold standard for much clinical research and practice. Appendix A includes aspects of a multiple N-of-1 clinical trial for high blood pressure.
N-of-1 clinical trials can be designed with adaptive dosing and adaptive data collection. Multiple N-of-1 clinical trials can be designed with adaptive patient selection. Adaptability would be based on contingencies that are based in turn on measured results. All contingencies and measurement procedures would be specified in research protocols that make the procedures objective, transparent, and reproducible.
MQALA makes it easy to process data from N-of-1 clinical trials that include more than two doses for each individual. Accordingly, doses can be randomized to periods using procedures that allow generally escalating doses over successive periods of time. Benefit/harm would be monitored over time by computation as a function of dose for individual patients. Dose escalation could stop after it was determined that higher doses were producing no additional benefit or if higher doses were shifting the balance from increasing benefit to increasing harm.
Since benefit/harm scoring is equally sensitive to both beneficial and harmful effects, evaluation of both safety and efficacy could begin immediately after the second assessment of health and a change in treatment, including the initiation of treatment, for each individual patient. Investigators would have to specify whether higher levels on each health variable are either beneficial or adverse.
Multiple N-of-1 clinical trials with adaptive dosing would make it easier for investigators to identify optimal doses for each individual patient as well as distributions of optimal doses and average optimal doses for entire samples of patients or any subset of patients.
Adaptable dosing calls for programmable dosing and data collection devices that could be designed to help maintain blinding or masking in clinical trials. Such devices should be Web enabled.
The new vision also would allow adaptive collection of data for health variables such as laboratory, symptom, performance, and quality of life or generic health status variables. One data collection strategy would be to follow hits on more general queries with more specific queries. This strategy is illustrated with the SAFTEE instrument (Systematic Assessment of Treatment Emergent Events). More information about adaptive data collection can be found at www.qualitymetric.com.
My interest in SAFTEE was another formative influence on what came to be MQALA. I continue to hold that a major factor that has impeded widespread adoption of more systematic, comprehensive, and adaptive data collection about the effects of treatments has been the lack of a methodology such as MQALA for processing the data. With MQALA, event rates, which tend to be higher with systematic assessment, have little bearing on benefit/harm scores except in extremes when events are either almost always present or almost always absent. The new vision overcomes a major source of resistence to systematic assessment for adverse events.
The new vision also calls for adaptive patient selection. This becomes possible because DataSpeaks Interactions® can be applied to data from multiple N-of-1 clinical trials to provide reliable and valid measures of how individual patients respond to treatments. These measures would make it possible to efficiently identify specific diagnostic and genetic predictors of differential response to treatments. The fact that the same gold standard methodology can be used for both clinical research and clinical practice would help make adaptive patient selection feasible in terms of patient numbers.
Adaptive patient selection would mean that clinical trials could begin with wide varieties of subjects and that targeting could be improved during the course of trials. This would increase the odds that new drugs would be found to be safe and effective for at least some specific groups of patients.
One implication of adaptive multiple N-of-1 clinical trial strategy is that essentially the same clinical trial design could be used for clinical trials for many different drugs for many different types of patients. Once established, the new vision would reduce the need to redesign clinical trials. In addition, it would be easier to combine results from different trials and compare the cost-effectiveness of different treatments.
The new strategy is designed to overcome many of the failures of the old vision.
6.1.4. Opportunities and Challenges
DataSpeaks Interactions® can revitalize the pharmaceutical industry by actually measuring, discovering, analyzing, synthesizing, and visualizing interactions that describe biological and treatment mechanisms as well as the benefit/harm of treatments. This is a big market opportunity. However, the new technology is disruptive because it calls for changes in the way health disorders are diagnosed and for a new vision of clinical research.
The pharmaceutical industry was an early adopter of high throughput data collection technologies. Perhaps no other major industry is so thoroughly controlled by the statistical establishment. Perhaps no other industry is so thoroughly mired in data swamps in a way that so dramatically reduces its potential.
Organizations - pharmaceutical companies, contract research companies, tools companies, universities, institutes - that become early adopters of DataSpeaks Interactions® could have huge competitive advantages even if pivotal trials continue to be conducted for some time with outmoded methods.
Perhaps it is time for at least one major pharmaceutical company to break with the pack and lead with data processing based on a new method that accounts for time and individuality and that actually measures biological mechanisms as well as the benefit/harm of treatments.
Almost everyone recognizes that health care in the United States needs reform. However, fundamental health care reform largely is gridlocked because there are almost as many proposals for reform as there are constituencies and points of view. Furthermore, some constituencies are represented by mammoth organizations with great momentum. Gridlock suggests that no one really knows what to do and that potential reformers have yet to recognize a fundamental underlying problem that can be agreed upon now and addressed in concert. Recognition of this problem and discovery of a technology that can help solve the problem creates a huge market opportunity.
Health care reform has earned a bad name. It has become a political and ideological football. Large bureaucracies would be counterproductive. Grand reorganizations are not sufficient. Cash infusions are not apt to unleash competition in markets or increase efficiency and accountability. Malpractice litigation does not seem to be improving health care. Malpractice reform does not create systems that reduce medical errors. Sensitivity, compassion, and good intentions are not sufficient to overcome ignorance. Current practices compromise ethics.
In addition to these conventional aspects of health care reform, health care systems should be expected to deliver new benefits made possible by developments such as comprehensive genotyping, Web enabled health monitoring devices, and home health care. To a large extent, failure to reform health care means that many breakthroughs in life sciences and technology will not help patients.
Health care reform is stalled for (1) lack of sufficient scientific understanding about how health-related systems work, change, and adapt and (2) lack, until now, of a scientific method to obtain the required scientific understanding in an efficient manner. Perhaps many different health care constituencies could rally around these points and start to achieve health care reform step by step. Health care needs to become a system that manages contingencies more effectively.
Given strong leadership, DataSpeaks Interactions® software has the potential to help unleash health care reform. All other sections of this Web site help support this claim. Perhaps more important than anything else, fundamental health care reform will have to overcome the awesome power of the statistical establishment. Statistical methods rule where it counts in evidence based medicine. But, for example, statistical methods are almost useless for processing the time ordered data in the charts of individual patients. People are left to process such data in their heads, which is overwhelming, odious, costly, unscientific, and prone to error. People can not process so much data in their heads and keep up with the demands of the modern age.
This focus on a technology that can improve scientific understanding of health related systems does not mean that health care reform depends primarily on professional scientists and researchers. To the contrary, access to MQALA, with its capability to account for time and individuality, will empower most health care constituencies, including many patients and potential patients, to improve their practice and behavior in accord with scientific understanding.
It often is said that most treatment episodes are experiments of sorts, especially in the context of chronic disorders. Diagnoses are somewhat like hypotheses. Treatments are interventions. Clinicians diagnose and treat and watch what happens. Depending on what they happen to see, clinicians may change doses or treatments and try again. Currently this whole process is haphazard and subjective, accounting for much avoidable suffering and cost.
MQALA applies to time ordered data for individual systems. This has major implications in terms of strategies for health care reform. It means that individual clinicians, individual practices, individual hospitals, and individual hospital systems can start pursuing health care reform one by one. Those who reform first and fastest and best are apt to dominate their markets.
Technological solutions also are earning a bad name in health care. After all, new diagnostic and treatment technologies have been known to increase the cost of care. DataSpeaks Interactions® is a different type of technology. It has the potential to increase the cost-effectiveness of most other technologies. Savings from reductions in lost and damaged lives, treatments for iatrogenic conditions, wasted treatments and procedures, costly professional time, and legal liability together with productivity improvements that result from increasing professional and patient empowerment have the potential to offset the cost of new software, training, and infrastructure required to support the software.
DataSpeaks Interactions® is not a final solution for reforming health care anymore than it is a drug or diagnostic system. But it is a software tool that can be used to reform health care, just as it is a tool that can be used to help discover and develop new drugs and diagnostic systems. Adoption of this tool will be disruptive. Health care reform based on scientific understanding and methods will require leadership and much hard work. Most of the science has yet to be done.
DataSpeaks Interactions® can help keep health care reform from being a one time event. It can help health systems to become learning systems so that they can improve continuously and adapt to new circumstances including new treatments and new and newly resistant pathogens.
Despite being based on technology, fundamental health care reform of the type envisioned here can make it easier to provide more personal and humane health care in addition to better outcomes for whole populations of people at prices people are willing to pay. In addition, it will make health care a more satisfying experience for millions of dedicated employees and for people who use health care services. Furthermore, DataSpeaks Interactions® can shift some of the burden of health care back to people who will be empowered to be more responsible agents to maintain and improve their own health and that of their loved ones.
The health related systems that we need to understand to reform health care include biological systems. In this respect, the solution to the growing crises in health care is similar to the solution for the growing crises in the pharmaceutical industry. Other health related systems that we need to understand include populations, economies, behavioral and social systems, as well as brains and artificial intelligence systems that are touched upon in other sections of this Web site.
Discussion of health care reform is a tall order. I will offer a few initial ideas about meeting the challenge to help jump start the process of reforming health care from a new scientific and technological perspective.
Health care will be considered to comprise several major interconnected markets - providers, payers, as well as patients and potential patients.
DataSpeaks Interactions® could make all the difference in reforming health care. This makes it worthwhile to consider how DataSpeaks Interactions® can be of value and how it can penetrate the health care market step by step. Section 2.8.2 of Patent 6,317,700 includes ten reasons why MQALA would be valuable from a practical perspective largely in the context of health care for the management or control of chronic disorders.
Health care providers will be considered to include (1) clinicians and all their support personnel including nurses and people who provide diagnostic and therapeutic services and (2) administrators and managers of health care practices, hospitals, and hospital systems.
DataSpeaks Interactions® can help clinicians both to diagnose health disorders and to evaluate treatments for health disorders. Doing so would help clinicians advance their careers, medical science, and patient welfare.
Many functional health disorders involve disordered interactions between and among biologically active substances and other health related variables. I described diagnosis by measurement of disordered interactions and mechanisms in the context of drug discovery.
DataSpeaks Interactions®appears to be the first software package to actually and effectively measure potentially diagnostic interactions using time ordered data for individual patients. This opens doors to many opportunities. Here is a simple example.
Intensive care units often monitor both blood pressure and pulse rate. I suspect that measures of interaction between these variables might provide important diagnostic information about the status of cardiovascular systems that is not revealed by the way these measures currently are used. This opportunity needs to be investigated by cardiologists. The beginning of this process is basically simple - measure the interactions in basically the same way for each of many patients, identify variations in these measures across individuals and over time for individuals, and identify what the variations mean in terms of diagnosis and treatment.
Clinicians in many specialties will be able to find similar opportunities. In this way, many clinicians would become familiar with MQALA and value of measuring interactions. The analogy described some advantages of using data movies, rather than data snapshots, for understanding dynamic processes such as health and disorder. The section on revitalizing the pharmaceutical industry described the value of going beyond levels of variables to measuring interactions in medical diagnosis.
Companies that sell health monitoring and functional imaging devices could increase demand for their products by providing modules of software based on MQALA to process the data.
6.2.1.1.2. Treatment Evaluation
The current development status of evaluating treatment effects in clinical practice is akin to the status of diagnosis before the modern era of laboratory tests and imaging procedures. The benefit/harm of treatments, as it becomes evident over time for individual patients, largely is evaluated subjectively, even when the required data are available in medical records.
Subjective evaluation of treatment effects for individual patients often reminds me of the old practice of clinicians diagnosing diabetes by tasting urine to see if it is sweet. Both practices are subjective, odious, and often inadequate.
DataSpeaks faces a major challenge in trying to get people to use DataSpeaks Interactions® for measuring the benefit/harm of treatments. This challenge is illustrated as follows. I was presenting a lecture at a university bioinformatics seminar. Most of the lecture presented results obtained by applying my prototype software to portions of a yeast, cell cycle control, time series, gene expression, microarray dataset that is publically available from Stanford.
Since most people seem to have trouble understanding what it means to measure interactions and I anticipated that the audience might include clinicians, I used a clinical example. I showed a graph with two time series variables - a measure of treatment and a measure of health. The graph showed an association that suggested substantial evidence for benefit. I explained that the summary score for one measure of this association was 10.76. Since this summary score was one score from a distribution of potential scores with a mean 0 and standard deviation 1, 10.76 could be interpreted as providing substantial evidence for benefit. (The distribution of potential scores was defined by the combination of the data and the scoring protocol as described in Appendix A.)
At this point a physician/researcher asked essentially what good does it do to measure the interaction when one can see that the data suggest benefit. I did not challenge him by asking what good does it do to measure sugar levels in urine when you can taste that the urine is sweet. Perhaps I should have. The lecture effectively closed the door on what appeared to be a promissing opportunity at a university where I had been a faculty member years before.
New measures can be keys to scientific progress and improved clinical practice. But when people have always gotten away without measuring something as important as benefit/harm, people seem to forget that measurement is important. Apparently I am the only one who really knows how to measure benefit/harm over time and across variables for individual patients (see reprint and patents).
DataSpeaks Interactions® is disruptive. DataSpeaks calls for strong leadership to help advance a technology that is critical and disruptive.
Health care needs DataSpeaks Interactions®. It is nearly impossible to form reliable and precise subjective impressions about treatment effects, especially when there are many repeated measurements of many health variables obtained while treatments fluctuate in level over time. Imagine trying to process subjectively all the data that are available for individual patients in intensive care. Imagine trying to form reliable subjective impressions about how treatments affect measures of functional connectivity that can be obtained from functional brain imaging. Measures of functional connectivity have the potential to be objective measures for diagnosing functional brain disorders and monitoring responses to treatment.
My impression is that most data relevant to benefit/harm is just ignored basically for lack of computational methods to process the data and to display results. Objective measures of benefit/harm have real advantages as do objective diagnostic measures.
Most people do not realize that apparent benefit/harm can be measured by computation from the relevant time ordered data for individual patients. I suspect that this reflects the power of the statistical establishment. Measurement of interactions between variables for individuals appears to be outside the lens of experience for people better trained to describe groups and make inferences from samples to populations.
Failure to measure benefit/harm for individual patients is costly in terms of health and money. Adverse drug reactions can continue until disaster. Unexpected benefits are not accounted for and beneficial treatments are terminated. Costly treatments are continued when there is no overall benefit and even when there is overall harm. Treatment is not individualized and optimized scientifically. Opportunities are lost to educate caregivers and patients as well as to capture experience that can improve treatment of other patients.
Forces are converging to begin objective measurement of interactions so that diagnosis and treatment can be improved. The key force is the discovery of a data processing method, MQALA, for measuring interactions over time and across variables for individual patients.
Another force is the completion of the Human Genome Project and the emergence of genetic testing. This puts a premium on the identification of specific disordered interactions and sets of disordered interactions that are diagnostic of functional disorders. These measures of interaction can be used to help identify genetic predictors of disorders. Similarly, reliable, valid, and specific measures of benefit/harm can be used to help identify genetic predictors of differential responses to treatments. Health care involves enough patients and can collect sufficient data to help make identification of genetic predictors feasible without major increases in research budgets. Costs would be controlled when clinical research and clinical practice are integrated.
Another force to begin measurement of interactions by computation derives from the rapidly increasing amounts of data that can be included in medical records for individual patients. Devices are being developed that can monitor physiological functions within bodies, radio the data out of bodies, and connect to the Web. Similarly, twenty minutes of functional brain imaging could increase the amount of data in patient records by orders of magnitude from what is now typical. All such data should be considered to be part of patient records. DataSpeaks Interactions® removes a primary bottleneck for using such data to improve health and health care. Development of data movies such as those provided by these examples can help make collection of the data worthwhile and increase demand for the data collection equipment.
A previous sectiondescribes how the measurement of benefit/harm with DataSpeaks Interactions® can be a guiding force in re-engineering clinical research. The new vision would help integrate clinical research with clinical practice. The pharmaceutical industry could be a leader in helping to bring this integrated vision to fruition, especially if it wants to make money providing disease management services and needs to learn more about its drugs by collecting data in secure but centralized repositories.
DataSpeaks Interactions® can help empower clinicians and put them in charge of their own destinies by providing the tools that they need to help optimize the care of their individual patients and by helping to make more clinicians a creative force in generating bodies of scientific knowledge required to improve health and health care.
6.2.1.2. Health Care Administrators
Administrators are the other key part of health care providers. Currently, health care administrators and managers appear to be in a thankless, difficult, high stakes, no win, monkey in the middle position between clinicians that need to be paid and payers that have been given good reasons to think that they are not getting good value for their money. But administrators also are in excellent positions to read the future and help shape a better future for health care.
One way to approach the problem of health care administrators is to compare two industries, the health care industry and the credit card industry. I will do this in terms of personal experience that many people in developed countries can appreciate. Both industries deal with sensitive confidential data.
I live in Michigan and traveled to Florida. When I got home I realized that my credit card was missing. During one call, I was able review every recent transaction - who was credited, from where, when, and for how much. Fortunately, the only problem seemed to be a lost card. My new card arrived the next day. Credit card companies that could not keep track of transactions efficiently would go out of business.
Compare this to when I go or take my children to a new doctor. I’m often handed a clipboard with some forms that ask about health histories and vaccinations. I find the vaccination questions especially troublesome. I recognize that vaccinations are important. But why are they asking me when doctors administered the vaccinations and have the records? I do not remember the details of our vaccinations accurately, fear for the consequences of not remembering, and suspect that some people have memories and records worse than mine. These sorts of experiences make some people wonder about the quality of service for which money is being paid.
Nothing about this comparison of two industries is new. Administrators and many other people recognize that transaction costs and problems created by poor transactions account for up to about one third of health care costs. Efforts are being made to control such costs. Indeed a number of available information technology services can help with the administrative business of health care - ordering, billing, paying, prescribing, referrals, scheduling, access to patient records, access to the medical literature and treatment guidelines, etc. England is embarking on a $17 billion bet on information technology primarily to improve the administrative business of health care.
DataSpeaks Interactions® is of value here because it offers a new reason and strategy for bringing health care into the information age. Measurement of interactions, including the benefit/harm of treatments, will become vital to providing quality health care. Measurement of interactions depends on the collection and processing of electronic data.
Part of the administrator’s problem has been that information technology has been focused primarily on the administrative business of health care. In this respect, it is the administrator’s problem. Clinicians, who have been known to be rather headstrong when it comes to changing their behavior and having their behavior managed, are left with a good out from embracing information technology to improve health care. Until now, clinicians could accept primary responsibility for providing good health care. Administrators only have to provide good pay and infrastructure for clinician practices.
DataSpeaks Interactions® takes away this clinicians’ out. Information technology becomes vital and integral to providing quality health care.
Similar points can be made by comparing diagnostic practice and treatment practice in health care. If there are important unanswered questions about diagnosis, it is expected that clinicians will order laboratory tests and other diagnostic procedures. In contrast, if there are important unanswered questions about the effects of treatments, it still is acceptable to monitor the levels of a few variables and to rely primarily upon subjective impressions about benefit/harm. There is no expectation that clinicians actually measure the benefit/harm of treatments apparently because no one has known how to measure the benefit/harm of treatments. Now that MQALA, a procedure that can be used to measure benefit/harm, is in the public domain, various health care constituencies will have raising expectations.
Anticipate that objective measurement of benefit/harm might become as integral to cost-effective health care when there are uncertainties about the effects of treatments as objective diagnostic procedures are integral to health care when there are uncertainties about diagnosis. At this early stage in our development of scientific understanding of how biological systems work and change, there is no shortage of uncertainty to drive application of information technology that can reduce uncertainty, improve lives, and control costs.
Clinicians might need good information technology far more than administrators. As information technology is developed to make diagnosis and treatment more individualized, scientific, and responsive to patient needs and preferences, administrative services could be piggy backed in on the central requirement that a specific new type of information technology is required to provide quality health care in the modern age. This also means that major information technology initiatives in health care, such as that in England, should be set up to include capabilities to collect and develop data movies.
Both cost-effectiveness and systems of payment are major issues in health care reform. Payers include insurance companies, governments, patients, and employers that provide health care benefits. Although DataSpeaks Interactions® can be used to help investigate impacts of various payment systems on health, society, jobs, and economic competitiveness when the software is applied to relevant time series data, the focus here is on the cost-effectiveness of health care.
People value good health and many people are willing and able to pay, some more than they are paying now. But people also generally expect good value for their money. Increasing the value of health care that actually is provided might be more important to reform than spending more money. More money is apt to follow reform that provides better value.
There appears to be growing interest in accountability in health care. This is evidenced by the Foundation for Accountability (FACCT, http://www.facct.org/facct/site/facct/facct/home#webmd).
FACCT advocates for and helps enable a “person-centered health care system.” However, FACCT appears to be limited by the prevailing scientific worldview and the statistical establishment. For example, although FACCT provides a number of quality measures, it appears that they have yet to realize that it is possible to measure the benefit/harm of treatments for individual patients as described in the new vision for clinical research, Appendix A, a reprint, and elsewhere on this Web site. Measurement of benefit/harm can advance accountability in health care.
DataSpeaks Interactions® will both make it possible to provide better value and increase expectations to provide better value in health care. Anticipate that concerns about cost-effectiveness will shift from specific treatments, diagnostic procedures, and facilities to more global advances in information technology for health care systems. These advances will be integral to diagnosis and treatment and help create integrated health care systems adaptive to the diverse and changing needs of individual patients. Figuratively, health care needs central nervous systems that can actually process, rather than merely collect and store, time ordered data. Such systems eventually will account for time and individuality.
New technologies can raise expectations for more productivity and better value. DataSpeaks Interactions® could raise these expectations dramatically. The new vision can raise expectations because it describes specific steps that can be taken now to solve fundamental and costly problems in clinical research and practice.
Payers often turn to outcomes research for guidance in setting health care policy. However, outcomes research has been severely constrained by the statistical establishment. Now that MQALA has been discovered, major constraints can be overcome to improve outcomes research in at least two major ways. The first way involves the use of health status measures and measurement of benefit/harm. The second major way to improvement involves individualization, group averages, treatment guidelines, and payment policies.
6.2.2.1. Health Status Measures and Measurement of Benefit/Harm
As described before, users of the statistical method in clinical trials typically perform statistical tests on health variables. As examples, trials for blood pressure drugs usually test measures of blood pressure and trials for clinical depression usually test depression scores. Among other problems, this makes it difficult to obtain comprehensive evaluations of treatments and to compare value with trials that test different treatments and health variables.
Health outcomes researchers have developed generic health measures such as the SF-36 Health Survey. This includes use of computer adaptive test technology to provide versions that are short, precise, and valid. Some versions, especially versions with rather short recall intervals, can be administered repeatedly. For more information, see www.qualitymetric.com.
Such developments in measurement of health status are major achievements. However, the value of these achievements is severely limited when users merely substitute health status scores for values of more conventional health variables while performing primary statistical tests. Mere substitution of variables tends to force avoidable and unproductive disputes about whether primary statistical tests should be performed on generic health status measures, more disease specific health status measures, or more conventional diagnostic measures. Testing health variables also fosters an unnecessary and costly proliferation of clinical trials as when different trials for the same treatment test different variables.
The alternative to testing health variables is to use DataSpeaks Interactions® to measure overall benefit/harm and profile benefit/harm across many health variables before conducting statistical tests. Some advantages of measuring benefit/harm are included in the “new vision” section.
MQALA also would make it easier to investigate and understand how benefit/harm with respect to generic health status measures is coordinated with benefit/harm with respect to other health measures such as diagnostic measures, laboratory tests, symptom surveys, measures of physical and mental performance, role functioning, and health perceptions.
Each type of health measure has its own constituency. Use of MQALA to investigate interactions between and among variables at different levels of health measurement hierarchies could help bridge gaps between clinicians who might favor clinical trials that measure laboratory variables and patients and employers that might be more concerned about how treatment affects performance at home and at work.
6.2.2.2. Individualization, Treatment Guidelines, and Payment Policies
Perhaps the greatest impact of the discovery of MQALA on outcomes research and health care policy is that it provides, largely for the first time, crucial scientific information about diagnosis and the benefit/harm of treatments that helps enable individualization of health care. Knowledge of patients’ genomes together with diagnostic information from snapshots of patients’ conditions seldom are sufficient to individualize and optimize patients’ treatments at this early stage in our development of medical understanding.
Customers and clients in other businesses are coming to expect more individualization in products and services as we move from the industrial age to the information age. Many people want to individualize their appearances, their homes, their cars, their lifestyles, and their financial services. Perhaps patients also will come to expect health care that is both evidence-based and individualized.
To a considerable extent, health care still is bucking the trend toward individualization. Major trends in health care policy are towards treatment guidelines for large groups and towards restrictive formularies and restricted access. These trends are more akin to public health policy than individualized health care. The opposing tends tend to pit clinicians and managers against each other and reduce distinctions between clinicians and public health workers.
Section 2.8.3 of Patent 6,317,700 discusses relationships between the public health and individualized care approaches to medicine from a historical perspective and how some conflicts between the two approaches can be resolved by MQALA. A subsequent sectionof this pagedescribes how MQALA can help improve public health.
Both public health policy and individualized health care can improve health. However, the public health approach alone largely is in conflict with achieving one of the great promises of the Human Genome Project, which is to help enable more individualized or personalized health care. Such conflicts need to be resolved in order to help achieve many health breakthroughs.
The apparent conflict between treating individual patients as individuals or as group averages relates to Paul Meehl’s distinction, dating back to the mid-1950s, between clinical (subjective, impressionistic) and actuarial or statistical (mechanical, algorithmic) prediction. Actions based on actuarial predictions generally yield better outcomes. However, there has been much resistance to implementing actuarial prediction. Perhaps much of this resistance derives from fundamental limitations of statistical prediction as well as from people overestimating their powers of prediction and wanting to feel important.
This illustrates a fundamental limitation of statistical prediction. Suppose that a particular patient with a chronic disorder has been treated with a particular drug in accord with a treatment guideline that is based on a consensus about the results of many randomized group clinical trials. Also suppose that response of the patient to drug challenge, de-challenge, and re-challenge suggests that the drug is causing serious liver toxicity. The prudent and expected choice would be to change or discontinue the recommended treatment. This may be a rather extreme example. But treatment often needs to be individualized or personalized both in terms of biological responses of individual patients and their preferences. In many cases such as this with an adverse drug effect, essentially clinical prediction trumps actuarial prediction.
A number of “algorithms” for assessing adverse drug reactions account for response to drug challenge, de-challenge, and re-challenge. But since such “algorithms” do not actually measure benefit/harm as an interaction between measures of treatment and health, they are largely subjective and impressionistic.
Much resistance to actuarial prediction might be a convenient and motivated overreaction to situations in which essentially clinical prediction trumps actuarial prediction. One approach to overcoming this resistance may be to adopt an additional algorithmic method, MQALA, which uses time ordered data to account for individuality and time.
MQALA also would advance randomized N-of-1 clinical trials, the gold standard for evidence based medicine (see, for example, http://www.cche.net/usersguides/applying.asp). One factor limiting advancement of this gold standard has been the lack a better method for analyzing data from N-of-1 clinical trials. The statistical method is not well suited for analyzing the results of N-of-1 clinical trials, especially when there are more than two doses. Appendix A illustrates how MQALA can be used to analyze N-of-1 clinical trials by measuring the benefit/harm of treatment.
MQALA enables a scientific means for personalized prediction and optimal care for individual patients. MQALA helps enable scientific treatment of individuals as individuals. The statistical method helps enable treatment of individuals as group averages. Both methods often need to be used together since each of us is both an individual and a member of various groups.
Use of MQALA together with the statistical method can help provide the best of both worlds in health and health care - optimal treatment of individuals that is based largely on MQALA and optimal treatment policies for groups that are based largely on the statistical method. Using the terminology of the dialectic process, the thesis is the statistical method, the antithesis is MQALA, and the synthesis toward which we should strive is the complementary use of MQALA and the statistical method.
In practice, I would expect this synthesis to mean that treatment episodes often would begin with provisional diagnoses and provisional treatments in accord with the most relevant treatment guidelines that are available. Then, especially when there is clinically significant uncertainty about outcomes, treatment would be optimized based on actual measures of apparent benefit/harm together with any additional relevant statistical information. Treatment guidelines would be modified if necessary in accord with the accumulation of new experience. This entire process would tend to “close the loop” so that clinical guidelines would continually guide practice and experience gained from practice would continuously update guidelines. This process would become part of future health care systems as clinical research and practice are integrated in accord with the new vision.
If treatment of each individual is optimized with help from MQALA, average health is apt to improve. At the same time, use of the statistical method may be the best way to bring experience gained from other patients to bear on the treatment of each individual.
The development of treatment and health care guidelines is becoming more challenging as diagnoses become more specific almost every day. Many guidelines may need to become genome and history specific. Perhaps the only way to obtain such guidelines is to integrate clinical research and clinical practice. This requires advances in information technology both to create the guidelines and to make the guidelines accessible when needed. Quality health care requires intelligence that goes beyond the experience of individual clinicians. The health care system itself will have to become intelligent.
Health care providers should anticipate that payers may not continue to pay providers that do not actually measure the benefit/harm of treatments when there is important uncertainty about outcomes. In addition, payers could start to expect the use of randomized N-of-1 clinical trials when such trials are cost-effective. Such trials are the gold standard of evidence based medicine and help provide valid measures of benefit/harm. In addition, measurement of benefit/harm could be used to help hold providers accountable for outcomes. This also could help shift payment from payment for treatments and other procedures to payment for measured outcomes.
Many payers are concerned about raising drug prices. Drugs can be one of the most cost-effective means for improving health. But payers can and should act to make drugs more cost-effective.
Supporters of the status quo in the pharmaceutical industry often argue that drugs must be expensive in order to support the huge costs of drug discovery and development. These supporters suggest that cutting drug prices would be like killing the goose that lays the golden eggs. This view supports certain interests of the statistical establishment and its subjects, which includes the pharmaceutical industry.
We must not kill the pharmaceutical industry. The alternative is to make drugs more cost-effective by revitalizing the pharmaceutical industry.
Payers should challenge the status quo in the pharmaceutical industry. More specifically, payers should challenge the pharmaceutical industry to try DataSpeaks Interactions® for measuring the benefit/harm of treatments and the mechanisms of health, disease, and treatment. Such trials of the new methodology probably are the most important steps that we can take now on the way to making drugs more cost-effective by making drug discovery and development more productive and efficient.
The apparently growing expectation that every health problem calls for an expensive professional solution could bankrupt payers, especially if this expectation absolves people from certain responsibilities concerning their own health and that of their loved ones. There might be limits to how much collective payers such as governments, insurance companies, and employers should pay to treat the consequences of over indulgent and self destructive behaviors.
6.2.3. Patients, Potential Patients, and Lay Caregivers
Perhaps one of the most important steps that can be taken now to improve the cost-effectiveness of health care is to empower patients and their families to take more responsibility for their own health. New information technology, scientific understanding, and scientific worldviews may converge to help make patients, potential patients, and lay caregivers more responsible. Leaders have a responsibility to help make the required technology and scientific understanding available to the public to improve health and the general welfare.
MQALA can help empower people to act more responsibly. I will work from the very general to some of the more specific implications of MQALA and DataSpeaks Interactions®.
As described in the responsible agency section, the MQALA scientific worldview apparently supports holding people responsible for their behavior, especially under certain conditions that knowledgeable leaders can help provide. This contrasts with the prevailing scientific worldview, which tends to view people as victims and passive respondents, controlled by their environments.
A worldview that enhances responsibility is fundamentally important to improving the cost-effectiveness of health care. The prevailing view appears to be that someone or something else is primarily responsible for most of our health problems, both for causes and for cures. I suspect that many professionals, including some tort attorneys, as well as some industries such as the diet industry benefit from the prevailing view. But there are points beyond which it is counterproductive to pass off responsibility and pay for others to be responsible.
Everyone is an agent, realizing that even chemicals and germs are agents. Some people are responsible agents. Some people are more responsible than others and some people have more responsibilities. Our leaders have particularly large responsibilities. Agents can be responsible for good and bad.
Health effect monitoring can help make people more responsible. Section 4.2.2.2 of Patent 6,317,700 describes health effect monitoring. Health effect monitoring is fundamentally different from health monitoring in that only the former monitors both independent or treatment and dependent or health variables and attempts to elucidate causal and other predictive interactions for the monitored individual.
MQALA and DataSpeaks Interactions® help enable health effect monitoring at the level of individuals. The independent variables could be measures of prescription drugs, over the counter drugs, alternative and complementary medicines, dietary components, allergens, pollutants, stress, and behaviors - almost anything that can fluctuate in level over time for individuals, can be monitored over time, and can affect health.
The dependent variables for health effect monitoring could be laboratory variables; measures obtained with Web-enabled health monitoring equipment; symptom and health rating scales including those that are Web enabled; as well as mental, physical and work performance - almost anything that can fluctuate in level over time for individuals, can be monitored over time, and can be considered as a measure of health. Users would specify whether higher levels of dependent variables were beneficial or harmful.
It would be best to collect health effect monitoring data under conditions of experimental control such as randomized N-of-1 clinical trials. This would help assure that the resulting scores are valid. However, when experimental control and blinding or masking are not feasible, temporal analysis parameters that are part of DataSpeaks Interactions® can be used to evaluate the temporal criterion of causal and other predictive interactions involving networks of variables.
DataSpeaks Interactions® would be used to measure benefit/harm in health effect monitoring. Processing the data with DataSpeaks Interactions® makes the data useful to advance scientific understanding and improve decision making.
Health effect monitoring can be a vital new tool for disease management programs and evidence based medicine. Some disease management programs already emphasize individuality and responsibility (http://www.healthmedia.com/research/strechers_insights.html). DataSpeaks would take this further.
DataSpeaks Interactions® helps close the loop involving clinical research and practice. Episodes of treatment and other interventions for particular patients or clients would begin and be guided by best available information from other people. If there is concern and uncertainty about optimizing response, patients or clients could enter health effect monitoring programs. The resulting information about benefit/harm could be used both to optimize the care of particular patients and improve treatment and care protocols for other patients or clients. Research and practice would be integrated and the problem of translating research results to clinical practice would be avoided.
Health effect monitoring may have particular power to motivate behavioral change because the results are most directly relevant to the person that the data are about. People often seem to consider themselves immune from consequences of behavior that befall other people. In addition, results from health effect monitoring may help modify behavior. Beneficial effects may reinforce and harmful effects may punish the behaviors that produced them.
6.2.4. Consumer Driven, Market Oriented Health Care Reform
DataSpeaks Interactions® can help enable consumer driven, market oriented health care reform by, for example, measuring biological mechanisms that can become disordered to cause dis-ease, measuring benefit/harm as well as mechanisms of treatment effect, and enabling health effect monitoring as a fundamental new tool for disease management and evidence based medicine.
Perhaps one of the most important implications of this approach is that it enables a bottom up approach to health care reform. With this bottom up approach, most individual stakeholders can start to take specific steps now to reform health care in an atmosphere of experimentation and shared learning. This could be called a Thousand Points of Light approach to health care reform.
I already offered suggestions about what individual clinicians and groups of clinicians could do both in the context of diagnosis and treatment. I made additional suggestions in the context of drug discovery and drug development. I also made suggestions for health care administrators and managers.
Health care providers should offer health effect monitoring services. Although taking personal responsibility for health generally may decrease demand for particular individuals to require professional services, providers that do the best job in gaining recognition for helping to keep patients healthy are apt to attract more patients when they need professional services. Health care will continue to require a lot of professional services. Health care providers could, for example, make health effect monitoring services available through their Web sites.
Payers have considerable leverage for reforming health care. The sections on payers include a number of recommendations. Payers, including charities, foundations and governments, could channel some of their funds from feeding the beast that has grown out of the old vision for clinical research and practice to support pilot programs that nurture the new vision of integrated research and practice.
Patients, potential patients, and lay caregivers also can help drive cost-effective health care reform. They could use health effect monitoring services and outcomes data while favoring providers accordingly. Various patient support groups could organize and use health effect monitoring services both to help optimize the health of their individual members and contribute to bodies of scientific information about their health concerns.
In contrast to this bottom up approach, many health care reform efforts have taken top down approaches in which conflicting interests of massive groups collide like galaxies to create great turmoil but little organized or progressive change. Health care systems appear to require central nervous systems. This requires information technology that includes DataSpeaks Interactions®. Development of information technology systems in health care should proceed with greater urgency as people recognize that these systems may be at least as critical to clinicians’ tasks of providing quality health care as they are to administrators’ tasks of processing transactions.
Most constituencies could start pressuring the pharmaceutical industry, which has the potential to provide some of the most cost-effective treatments, to reinvigorate itself by starting to measure biological mechanisms as well as the benefit/harm of treatments and the mechanisms of treatment effect. Agencies that regulate the pharmaceutical industry also need some encouragement to foster reform.
Data processing infrastructure and services companies such as IBM could help make DataSpeaks Interactions® software accessible to the world with some confidence that their modest but visionary investments will pay off.
Some of the most dramatic advances in health derive from practices that have been applied to whole populations. Such public health practices include providing safe drinking water and sanitary sewage disposal.
The previous market opportunity focused on how MQALA can help reform health care through applying DataSpeaks Interactions® to individual patients. In contrast, this section focuses on how DataSpeaks Interactions® can improve public health by applying DataSpeaks Interactions® to individual populations. Populations in this section are collective entities or composite individuals that are geographically defined.
The market for DataSpeaks Interactions® in epidemiology is not large compared to the other six market opportunities addressed by this document. However, this market could have large impact if policies based on use of DataSpeaks Interactions® improve public health. This could help advance DataSpeaks Interactions® as well as health. In addition, this market provides additional opportunities. One opportunity is to expand the market from epidemiology to clinical epidemiology.
Another opportunity might be a good strategy for advancing DataSpeaks Interactions®. Although the statistical establishment is our primary competition, it will be important for DataSpeaks to gain recognition from the statistical establishment. However, achieving this recognition can be tricky. Much depends both on how statisticians react and how those who often defer too much to statisticians react.
First, I will focus on some strategic considerations of particular relevance to statisticians.
Multiple time series epidemiological data currently appear to be some of the most difficult data for statisticians and epidemiologists to process on the way to scientific understanding. However, over the last decade or so, much time series epidemiological data has become available from devices that monitor environments. Much time series data involves the health effects of ambient air pollution. The Health Effects Institute, which is a partnership of the U.S. Environmental Protection Agency and industry (http://www.healtheffects.org), has been a leader in the use of such data. Many articles based on this work have been published including Fine Particulate Air Pollution and Mortality in 20 U.S. Cities: 1987-1994 that was published in the New England Journal of Medicine (http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmd=Retrieve&db=PubMed&list_uids=11114312&dopt=Abstract). The methods used in such studies appear to represent the gold standard for analyzing multiple time series epidemiological data. In addition, the results of such studies have been used to establish air pollution policies. My understanding is that much of the data is publicly available and could be used to demonstrate new data processing methods such as MQALA.
Section 2.5 of Patent 6,317,700 illustrates how MQALA and the statistical method can be used alone or in combination for epidemiological investigations. This section, which was written as a hypothetical example before I was aware of ongoing work that used multiple time series data in epidemiology, is a distinct alternative to the current gold standard methods.
In brief, MQALA would be used to measure interactions between daily air pollution levels and daily death rates in each of several geographically defined regions. Then a single group t-test on mean interaction scores would be used in an attempt to reject the null hypothesis of no relationship. A similar use of the t-test is shown in Demonstration 1 of Appendix B.
I propose comparing the strengths and weaknesses of methods based on MQALA with the strength and weakness of the gold standard methods that have been published for understanding relationships involving measures of air pollution levels and measures of morbidity and mortality.
One basis for comparing new and gold standard methods would be ease of use and understanding. After measuring interactions with MQALA, the initial statistical challenge reduces to performing single-group t-tests on means (see Demonstration 1). The t-test is one of the simplest, most basic and widely understood tools for statistical inference. As such, MQALA is not apt to increase demand for statisticians. However, MQALA provides values of new measures that often need to be processed in additional ways with the statistical method. This includes development of mathematical models based on the new measures of interaction. This has the potential to restore demand for statisticians. But this could be a major and difficult change from the status quo.
Other bases of comparison would include how well the different methods account for potential confounders such as weather, for the effects of various pollutants that might work together in linear or non-linear ways, for how pollutants might be related to syndromes of events and for the temporal criterion of causal and other predictive interactions. As suggested by Section 2.6 of Patent 6,317,700, MQALA has features to address many such issues. However, this would require substantial additional work.
Second, I will focus on some strategic considerations that could affect the advancement of MQALA from the perspective of those who often defer too much to statisticians.
Policy makers might favor the relative simplicity enabled by MQALA. However, policy makers generally are slow to embrace the results of new methods as a basis for making policies. Perhaps one of the most important considerations here would be to get buy in from statisticians who are experts with the established methods. For such reasons, demonstrations with MQALA should involve experts with established methods whenever possible. In addition, demonstrations with other types of complex adaptive systems, perhaps most especially brains, could help advance DataSpeaks Interactions®.
Although the market for software in epidemiology is small, the market could be expanded by extension into clinical epidemiology. In addition to expanding into health care as discussed above, DataSpeaks Interactions® has many potential applications that would help users understand how individual people are affected by pollutants, allergens, dietary components, and other environmental variables. Web enabled systems could help users to collect and process data in accord with their own concerns.
6.4. Visualizing How Brains Work, Change, and Adapt
Perhaps the most promising market opportunity for DataSpeaks, Inc. to pursue first involves use of DataSpeaks Interactions® to visualize how brains work, change, and adapt. This, by itself, would not be the biggest market opportunity. But it may be the most strategic business opportunity. This opportunity appears to be a good choice because there are a number of specific steps that can be taken now at relatively low cost to help advance this measurement concept, gain leadership, develop this market, and enter additional markets. Eventually, DataSpeaks Interactions® would have to be developed so that it can meet validation requirements of regulatory agencies such as the U.S. Food and Drug Administration.
DataSpeaks Interactions® can be applied to currently available data to actually measure mechanisms of brain function. Functional brain imaging provides some of the best data movies that are available for multiple individuals. Techniques such as functional magnetic resonance imaging (fMRI) and Positron Emission Tomography (PET) are providing data movies with good temporal and spatial resolution. In addition, data are readily available from public sources such as www.fmridc.org.
Additional modalities such as electroencephalography (EEG) and magnetoencephalography (MEG) provide data movies of brains in action. DataSpeaks Interactions® can be used to measure interactions involving measures of rhythmicity.
Perhaps the simplest initial demonstration would involve application of DataSpeaks Interactions® to measure function in individual brains. For this application, all the data would come from the brain scans themselves - that is, both independent and dependent events would be determined from the brain scan data alone as in Demonstration 3.
Measurement of brain function as interaction between and among brain regions goes beyond measurement of action in brain regions. My impression is that modalities such as functional MRI could be better described by example, as visualizing the “brain in action” despite the use of “functional” in describing the imaging modality. DataSpeaks Interactions® would be applied to these data to visualize coordination of action between and among brain regions in accord with Patent 6,516,288, which involves action coordination profiles. Perhaps fMRI would be described more accurately as “action MRI” to contrast it with structure MRI. This has precedent (http://www.newhorizons.org/nhfl/about/cornerstone.html).
Each frame in a brain scan data movie is comprised of a number of pixels. Each pixel shows the level of activity in a volume element or voxel of a brain or some nearby structure. Each pixel or voxel that represents a particular region is a variable. Repeated measurements of each variable would form a time series variable, one variable for each voxel. Each time series variable would have the same number of repeated measurements as there are frames in the data movie.
Mechanisms of brain function would be visualized by measuring the interactions between and among brain regions. In this context, which involves neural circuitry, interactions or temporal contingencies often are described with terms such as connectivity, interconnectivity, functional connectivity, effective connectivity and pathways. These are the connections and pathways that help define our identities as working individuals.
Currently, the relevant literature demonstrates substantial interest in measuring connectivity, presumably in part because disordered or diminished connectivity is thought to underlie many functional mental and neurological disorders. However, to the best of my knowledge, attempts to measure connectivity apply statistical or related mathematical techniques that do not actually measure the interactions themselves as functions of relevant analysis parameters in individual brains. Appendix B shows interactions can be measured as functions of analysis parameters that account for levels of the interactants, the episodic nature of events, delays, and persistencies.
Demonstration 3 of Appendix B includes a small portion of results from a preliminary demonstration for a small patch of motor cortex. These results appear to provide strong evidence for layering. These results would appear to help validate DataSpeaks Interactions®. The striking pattern observed in these results calls for explanation whether or not the pattern represents different levels of neural activity.
The use of DataSpeaks Interactions® to visualize brain function could be done after some modification of the software demonstrated in Appendix B. The software would have to be ported to a higher performance computing environment. Portions of the software would have to be modified to accommodate more scores. More specifically, the current prototype exports scores and values of the analysis parameters that yield each score to Excel. As such, the prototype is limited by the number of rows in one Excel spreadsheet (about 65,000). In addition, it would be necessary to program a new means for displaying measures graphically.
Appendix A and Patent 6,317,700 identify additional features of MQALA, perhaps most especially Boolean events, that could be used to define the presence or absence of an almost endless variety of events involving assemblies of neurons. These features have the potential to allow investigators to capture more of the complexity of how brains work, change, and adapt.
Initially, for the sake of simplicity, I recommend limiting initial demonstrations to two-dimensional cross-sections of brains or to brain slices. I would start by recommending a cross-sectional image of results color coded so that the magnitudes of apparently excitatory interactions would be presented with one spectrum of colors and the magnitudes of apparently inhibitory interactions would be represented with another spectrum of colors. Alternatively, the visualization could use two colors displayed with different degrees of brightness with zero value scores being black. Additionally, I would recommend an interactive display in which a user would use a cursor to identify a particular pixel in a brain slice. After a particular pixel is identified, all other pixels would show how activity levels in the pixel selected as the independent variable interact with activity levels in all other pixels.
This type of display would be one way of visualizing action coordination profiles as described in Patent 6,516,288. In this case the interactions need not be causal because the data were not collected under conditions that experimentally controlled levels of activity in pixels used to define independent events. It may be possible to achieve some degree of experimental control with techniques such as transcranial magnetic stimulation.
The basic interactive display just described would show the summary scores for each interaction. Each summary score probably would summarize thousands of scores for that pixel depending on scoring options selected by investigators. A relatively simple extension of the display that would be based on capabilities of the available prototype software would allow users to examine measures of interaction or connectivity as functions of all levels of analysis parameters that were selected for use in a particular scoring protocol. For example, the analysis parameter called delay could be used to help investigate the temporal criterion of causal and other predictive interactions. The PowerPoint presentation included to supplement materials for Demonstration 1 includes graphs that show average measures of interaction as a function of delay.
One important reason why the functional brain image analysis application is particularly desirable as an initial market opportunity is that the variables are localized in space in a manner that permits the type of visual display just described. Such a display would be much more difficult with, for example, time series microarray data because microarray variables generally are not so clearly organized in space, at least at this time. Good visual displays of interaction scores can help investigators understand what interaction scores mean. This is important for a new technology that measures interactions, as the term is used here, for the first time.
The basic type of demonstration just described for a particular brain would lend itself to additional extensions that could help sell DataSpeaks Interactions® to potential users. For example, I would recommend preliminary comparisons with different types of brains. For example, corresponding slices could be compared for normal brains and brains of people with Alzheimer’s disease. In this example, I would anticipate diminished or disordered connectivity between the hippocampus and the cortex. Further demonstrations of this type would benefit greatly from integrating DataSpeaks Interactions® with software for statistical analyses.
DataSpeaks Interactions® has the potential to be the basis for using available brain imaging technology to provide a new and objective means for diagnosing functional brain disorders. Such an effort probably would involve measuring interactions from thousands of brains together with statistical analyses of the resulting measures. Effective service as a diagnostic tool would be a good entrée to the larger health care market.
Companies that make functional imaging devices would be good potential customers for DataSpeaks Interactions® because our software has the potential to greatly increase demand for their devices. Some of these companies such as General Electric and Siemens are active in the broader medical informatics marketplace. Such companies could benefit from applications of DataSpeaks Interactions® in health care.
Companies such as Kodak that offer Picture Archival Systems (PACS) also are potential customers for DataSpeaks Interactions®. Such companies potentially would have access to functional images from thousands of patients as required to help make functional imaging a gold standard for diagnosing functional brain disorders and monitoring responses to treatments.
Functional brain image analysis with DataSpeaks Interactions® also may be a good entrée into the pharmaceutical industry. Drugs for disorders of the central nervous system are a large part of their market and have much potential for growth. Many such drugs up or down regulate various neurotransmitter systems. DataSpeaks Interactions® has the potential to help elucidate such mechanisms by measuring how connectivity is affected by drugs. Conceptually this would be quite simple - measure connectivity before and after treatment with a drug or potential drug and analyze any changes in measures of apparent connectivity.
Behavior modification apparently involves changes in connectivity. DataSpeaks Interactions®, combined with functional imaging, has potential value for investigating the neural basis of learning and other forms of behavior modification.
DataSpeaks Interactions® also has potential for investigating the neural control of behavior. The most straightforward uses of this application would involve defining independent events with brain imaging data and dependent events with behavioral data.
I anticipate that scientific understanding of brains that can be obtained with DataSpeaks Interactions® will help inspire development of artificial systems that can learn. MQALA, the methodology embodied by DataSpeaks Interactions®, has a type of face validity for this purpose. Brains work largely through discrete events - all or none action potentials. Similarly, MQALA works by measuring temporal contingencies between and among discrete events. This suggests that MQALA might be particularly suitable for developing mathematical models of how brains work, change, and adapt as well as for developing artificial systems that work and adapt more like brains.
The current state of the art in functional brain imaging often involves activation studies that are used to identify specific brain regions that are affected by stimuli and tasks. Activation studies have been enormously helpful in mapping brains. However, activation studies appear to be of limited value for elucidating higher order phenomena such as attention.
It appears unlikely that attention would be localized to specific brain regions. In contrast, attention to stimuli may be more accurately represented by patterns of connectivity present during periods of time around the time when stimuli are presented. Such patterns of connectivity could involve many brain regions simultaneously. DataSpeaks Interactions® has the potential to elucidate higher order phenomena such as attention by measuring underlying patterns of connectivity in space and time. For example, it may be easier to investigate the neural basis of higher order phenomena such as attention by measuring patterns of apparent connectivity with DataSpeaks Interactions®. Activation studies do not measure such patterns.
In a broader context, human brains epitomize complex adaptive systems. Data processing methods that work for brains might command attention for application to other types of complex adaptive systems.
Brains clearly manifest all three types of mechanisms by which complex adaptive systems have been said to work - function, response, and agency. Patent 6,317,700 includes specific claims that address all three types of mechanism in the context of serial functional imaging.
Compared to the statistical method, MQALA is superior in accounting for individuality and time. Scientific methods that work best for average time-invariant systems would appear to exclude much of what is most interesting and valuable about brains. Interesting features for which MQALA would appear to be superior include how brain actions are coordinated and how brains change through development and aging as well as how brains adapt and show substantial neuroplasticity.
DataSpeaks Interactions® would help enable a logical progression in the evolution of the use of imaging in medicine. Initially, the use of imaging was limited primarily to the visualization of structures, which included broken bones and tumors. More recently, so-called functional imaging has been extended to, for example, visualize and map brains in action. The next step, enabled by DataSpeaks Interactions®, would be to visualize how actions in various regions are coordinated to form entire working, changing, and adapting systems. This appears to be analogous to what can be accomplished with other types of complex adaptive systems.
6.4.1. Visible Brain, Visible Human
DataSpeaks Interactions® can develop data movies of brains in action to visualize brain function in terms of interactions between and among brain regions. As such, DataSpeaks Interactions® can help make brains visible, not just as static structures but as dynamic functioning systems.
Brain structure already is visible as part of, for example, the Visible Human Project (http://www.nlm.nih.gov/research/visible/visible_human.html). This project provides detailed 3-dimensional renderings of the human male and the human female. Given that the Project is limited to the three spatial dimensions, the Project could be described more specifically as the Visible Structural Human Project.
Some years ago after discovering MQALA and first hearing of the Visible Human Project, I began to speculate about a potential “Visible Functional Human Project.” Such a project would add a forth dimension, time, and could begin by addressing function. The project also could go on to address additional aspects of work, namely response and agency as well as change, adaptation and individuality. This new project also could go on to address nested hierarchies of systems at various levels of understanding. A new “Visible Working Human Project” could help organize scientific knowledge about what it means to be human.
This would be a research agenda worthy of great nations. MQALA made the conception possible and makes the agenda feasible.
6.5. Improving Prediction of Economies and Capital Markets
Appendix B includes Demonstration 2 of how DataSpeaks Interactions® can be applied to measure interactions between economic time series. Additional demonstrations using these data suggest that DataSpeaks Interactions® can be used to develop predictive indices as partially described in Section 4.4.3.8 of Patent 6,317,700. Extensions of this work will require investment in programming for DataSpeaks Interactions® prototype software. DataSpeaks Interactions® also can be applied to data about capital markets.
Although application of DataSpeaks Interactions® to capital markets may do more to redistribute wealth than create wealth, this application can increase the efficiency of capital markets. But redistribution of wealth our way does have a certain appeal. Additional wealth could be used to help finance other applications of DataSpeaks Interactions®. Being rich is apt to be more fun when it is generally recognized that one deserves to be rich.
Software that improves prediction of economies and capital markets has the potential to be a substantial market, particularly through the improved financial services that it can support. In addition, this application could be developed in a proprietary manner. Here are some reasons why DataSpeaks Interactions® may be advantageous for prediction compared with prevailing practice. These comments offer some rationale for trying DataSpeaks Interactions®.
Chaos appears to be a genuine component of economies, capital markets, and other complex adaptive systems. However, there also is order. DataSpeaks Interactions® can help users capitalize on this order. Small margins of improved predictability can provide a substantial edge in investment performance.
Those who favor the “random walk” hypothesis question whether capital markets are predictable. Predictability may well be a matter of degree. In addition, it may not be wise to conclude that capital markets are not predictable to a greater degree until all reasonable predictive strategies have been evaluated. Although a number of mutual funds started using artificial intelligence a decade or so ago, none to my knowledge were based on measuring interactions as described in Appendix A.
The random walk hypothesis appears to be predicated on a process similar to Brownian motion. Brownian motion describes the motion of an extremely small particle as being essentially random because it is subject to essentially random impingements of other small particles such as molecules.
Many capital market time series may look like random walks primarily because there has been no adequate way to measure the “impingements” of other variables. DataSpeaks Interactions® can be used to measure such “impingements” by measuring interactions. These interactions really are not just impingements but sustained and variable interactions. The random walk hypothesis needs to be challenged by processing appropriate data movies with DataSpeaks Interactions®.
This appears to contrast sharply with much current practice. Some strategies for technical analysis are based on trying to read and project histories of particular series almost as if they were trajectories. (The sectionon responsible agency contrasts the prevailing scientific worldview, which emphasizes trajectories, with the MQALA scientific worldview, which emphasizes interactions.)
There is little reason why histories of particular variables should be very predictive of the behavior of complex adaptive systems. Predictability is more apt to reside in how different variables interact. Science has made progress by investigating interactions between variables even though it has lacked adequate methods to measure interactions for individuals. Until these interactions are measured and the measures combined to make predictions, it is going to be difficult to make good predictions about economies and for investment decisions.
One aspect of efficient markets is that investors act on information and data soon after it becomes publicly available. But much information in publicly available data is not acted upon because no one has had a good way to measure and visualize patterns of interaction in the data movies. The DataSpeaks’ advantage is in using the publicly available data more effectively.
This illustrates the potential value of accounting for potential impingements in the context of analyzing functional brain images obtained over a period of time. Functional brain imaging can yield many potential variables - one for each voxel corresponding to a particular brain region - much as many variables can be used to investigate economies and capital markets. I will hazard to guess that activity in any one voxel might appear to be somewhat like a random walk or perhaps just rhythmic as long as one fails to account for other variables such as stimuli, behaviors, physiological conditions, and the effects of activity in other voxels or brain regions. Yet it seems unlikely that brain activity is essentially random and without some degree of coordination. Brains that did not coordinate would not have much survival value.
DataSpeaks Interactions® can help users account for coordinated action and improve prediction. In many such cases, whether the cases involve brains, economies or capital markets, progress on the problem of prediction appears to involve measuring the interactions or temporal contingencies with DataSpeaks Interactions® so that they can be investigated scientifically.
The current state of the art in predicting economies and capital markets appears to rely heavily on charting software. Such software often can show multiple variables on a common time axis. This helps people visualize interactions. But, as illustrated in the section about the benefits of developing data movies, the task of accounting for many interactions simultaneously in one’s head soon becomes overwhelming and odious as an unreasonable expectation as the number of potential predictors increases.
The next steps in the development of software for prediction are to actually measure interactions and account for many predictors simultaneously in accord with their predictive power. Then people could do a better job of predicting largely without looking at charts - much as clinicians can now do a better job of diagnosing diabetes without tasting urine.
DataSpeaks Interactions® probably will help make capital markets more efficient and less predictable. In the meantime, there is a lot of money to be made by development partners and early adopters.
Temporal contingencies modify behavior. This is a quick summary of classical conditioning, instrumental or operant conditioning, paired associate learning, associative learning, extinction, habituation, sensitization, and other processes that modify behaviors of organisms and other complex adaptive systems. Section 4.2.6 of Patent 6,317,700 provides some additional information.
Behavior modification exemplifies adaptation and illustrates why temporal contingencies matter and need to be measured. In contrast to temporal contingencies, spatial contingencies may not matter much unless there are interactions between and among objects over time, temporal contingencies. It appears as if evolution might involve temporal contingencies working through mechanisms such as natural selection that involve gene pools.
Temporal contingencies that describe how behaviors are organized and how behaviors are modified generally involve stimuli and responses, which are broadly defined here to include various types of events both internal and external to individual systems. These include reinforcements and punishments. In general, it appears as if different behavior modification processes can be explained in terms of one type of contingency involving stimuli and responses having capabilities to modify another type of contingency involving stimuli and responses. For example, a temporal contingency between a conditioned stimulus and an unconditioned response can change a temporal contingency between the conditioned stimulus and a conditioned response. This type of example has been called stimulus substitution.
DataSpeaks Interactions® provides a fundamentally new way to measure and describe temporal contingencies including those that involve stimuli and responses. Appendix A describes some features of this measurement system. As a new measurement system, DataSpeaks Interactions® has potential to advance scientific investigations of behavior organization and modification as well as applications of the resulting knowledge in fields such as education, training, and health care.
Bad, dumb, and uncivilized behavior is a big market opportunity. But my impression is that some other market opportunities should be pursued first.
6.7. Advancing Responsible Agency
Agents have effects on their environments, including people. In addition, there appear to be certain conditions under which individuals can be responsible for themselves and held accountable by others for their behaviors. When these conditions are present, it is reasonable and often productive to reward, honor, and punish responsible agents in accord with their behaviors. Some people have opportunities to exercise responsible agency when they vote for leaders expected to have favored effects.
Although it appears as if DataSpeaks Interactions® can help advance responsible agency and that this could be the basis for a large market, it may not be good strategy to begin by focusing on this market. For one reason, this opportunity appears to be relatively less amenable to specific business development steps that can be taken now. In contrast, other sections of DataSpeaks.com focus on specific concrete recommendations that should be taken now. As examples, DataSpeaks Interactions® can help revitalize the pharmaceutical industry, reform health care and visualize how brains work and adapt.
I will address this opportunity because it might help potential leaders understand what is at stake as they decide whether or not to help advance DataSpeaks Interactions®. In addition, if any of this discussion actually engages people, it is apt to create demand for DataSpeaks Interactions® and improve human welfare.
Responsible agency involves issues that cut to the heart of scientific worldviews or weltanschauungs. These issues often are discussed in the context of determinism and free will.
Science helps shape worldviews. This includes the effects of scientists such as Copernicus, Newton and Einstein together with their laws and theories; discoveries such as germs; inventions and tools such as microscopes and telescopes as well as methods such as the statistical method.
MQALA is a discovery that is being embodied as an invention, DataSpeaks Interactions®, which is a new set of software tools. Furthermore, it appears to be based on a rather distinct scientific worldview, which I will call the MQALA worldview. As such, DataSpeaks Interactions® appears to have important implications for ethics, personal practices, public policies, political platforms, legal liability as well as accountability in medicine and other practices.
Both the prevailing and the MQALA scientific worldviews emphasize measurement, objectivity, and experimental control to elucidate causal relationships. Of the two, the MQALA view is more dependent on data and software. In addition, MQALA has special implications for issues such as responsible agency.
I do not have any final answers about these great issues that continue to challenge humanity. Now I am just trying to advance MQALA, which is a methodology that we - me and you, my dear reader - can use to help advance scientific understanding. Generally convincing answers to a number of great issues do not seem to exist at this time. The real work is just beginning. MQALA can help advance this work as we seek better answers. I do hope to prime people to think and share their thoughts as we seek to understand our role and future in the world.
This, very briefly, is how I am coming to view the history of our world and scientific methodologies as I continue to develop MQALA. This presentation emphasizes implications for responsible agency.
I visualize the history of our world as a vast river of events over time, subject to quantum and relativistic effects that tend to baffle me. Clusters of events that tend to cohere form various types of objects such as atoms, molecules, planets, and stars. Objects follow trajectories and generally behave in accord with the laws of physics and chemistry.
Already some differences between the two worldviews are becoming apparent. The prevailing view tends to focus on objects that have come to be understood quite well thanks to a lot of intellectual heavy lifting, hypothesis driven science and methods that work best for time invariant systems. These methods work especially well either when systems have few parts that simply follow trajectories or when systems have many essentially identical parts that can be adequately investigated statistically as in statistical mechanics and thermodynamics. These methods do not work as well to understand traffic on busy highways when the vehicles largely are controlled by many individual and unique drivers.
In contrast, the MQALA worldview focuses on events in a river of time. The shift from objects to events is nothing new. Quantum mechanics applies to events as when a photon is considered to be a quantum of action rather than a particle or a wave in a medium. I don’t know if there ever will be any connection between quantum events and discrete events as used by MQALA and described in Appendix A. However, both quantum mechanics and MQALA are fundamentally probabilistic.
It appears as if a number of attempts have been made to use quantum mechanical worldviews to explain phenomena such as responsible agency and consciousness. But these attempts do not appear to be successful in explaining both higher order phenomena and advancing research agendas about how complex systems work, change, and adapt. Speculation often appears to outstrip data. The physical sciences do appear to provide some opportunities for the emergence and evolution of complex adaptive systems through processes such as genetic mutation. Extreme sensitivity to initial conditions creates examples of chaotic behavior.
According to the MQALA worldview, somehow over billions of years at least on earth, complex adaptive systems began to emerge and evolve. How all this occurs largely remains a mystery. But it is clear that there are trillions of individual systems of millions of kinds, many of them nested in systems of various degrees of inclusiveness all the way up to our biosphere. Understanding these systems goes beyond understanding their trajectories.
Individuals with identities such as particular organisms, people, species, populations, brains, economies, ecosystems, and other types of systems and subsystems get organized as they come into existence, sustain themselves from fractions of a second to millions of years, work, change, and adapt. Many individuals disappear and some leave legacies. Some individuals adapt so much that new types of individuals emerge, often retaining aspects of their former selves so that yeast, mice, and humans share similarities. Simpler systems can combine to form more complex systems. Hierarchies of systems emerge. People and organizations of people design systems. Some individuals emerge that seek to understand themselves, other individuals, and the world.
Being part of a river of events, individuals intermingle, swirl in eddies, and flow. The flow can become turbulent. Different individuals are brought into contact and may interact. People with different worldviews, scientific and non-scientific, may interact.
The acceleration of history appears to be positively related to the diversity and richness of interactions involving multitudes of different individuals. Inventions such as the printing press and the Internet, organizations such as universities, processes such as thinking and globalization and services such as Google contribute to interactions in ways that appear to accelerate history. We have reached the present time. We recognize that interactions matter but we really have not adopted good methods for investigating mechanisms of interaction over time.
It is here at the point where turbulence creates a great diversity and richness of interactions involving individuals that the MQALA worldview offers some of the clearest distinctions compared with the prevailing scientific worldview. The MQALA worldview says that temporal contingencies matter. Temporal contingencies are another way of describing interactions or longitudinal associations. But the implications of the difference between the two types of description appear to be profound.
The prevailing scientific worldview tends to focus on immutable laws of nature that apply to everything everywhere. According to these laws, objects follow trajectories. Objects are considered to interact as if they were billiard balls. One great advance occurred when it was recognized that objects are subject to relativity.
The Einsteinian worldview, a particular instance of the prevailing scientific worldview, is that everything is determined and predictable. Laws are contrasted with mere contingency. Einstein condemned contingency by saying that nature does not play with dice. Instead of scoffing at temporal contingencies, MQALA measures temporal contingencies to help make them a subject matter for scientific investigation.
One implication of complete predictability from the immutable laws of nature is that the horizon of unknown future possibilities apparently would collapse into a single point as scientific understanding advances. This may be one interpretation of what is illustrated at http://www.singularitywatch.com/index.html. This contrasts with the MQALA worldview of the future as “an expanding horizon of possibilities,” apparently in accord with Steven Hawking in The Universe in a Nutshell.
MQALA appears to support a relatively unpredictable and expanding horizon of possibilities. The two opposing positions about predictable possibilities are reminiscent of the old controversy about whether the universe will continue to expand forever or if it will collapse. Although the MQALA worldview does not offer full predictability, responsible agents are able to create their futures with some regularity but no assurances. Functional relationships and scientific laws based on measures obtained with MQALA would be probabilistic laws because MQALA is fundamentally probabilistic.
I already mentioned Wolfram and A New Kind of Science. Wolfram appears to offer a variation of the prevailing scientific worldview. Instead of a world operating in accord to knowable and determinate laws of nature that would enable complete predictability, Wolfram appears to see the world unfolding in accord with the rules of programs similar to those used by computers.
According to Wolfram’s variation of the prevailing view, the future appears to be determined by programs. But the only way to know the future may be to run the programs. Even simple programs can produce randomness. This randomness is determined by rules. But all randomness need not be determined by rules. Randomness can make temporal contingencies particularly interesting because they appear to be productive in nature as illustrated by how contingencies modify behavior.
Wolfram appears not to see or anticipate that temporal contingencies matter. Neither “contingency” nor “temporal contingency” are indexed in A New Kind of Science, which has an index of about 64 pages in quadruple columns with fine print.
Wolfram does discuss responsibility in the Notes for Chapter 12. For Wolfram, responsibility appears to reduce to issues of “computational irreducibility.” As such, responsibility and related higher order phenomena would not appear to be amenable to programs of mathematical and scientific investigation. Perhaps measurement, as illustrated by MQALA, connects mathematics to the world, including higher order phenomena.
Scientific worldviews have consequences for the conduct of scientific investigations. The prevailing view directs investigators to discover the immutable laws of nature, primarily through hypothesis driven science. The Wolfram variant directs investigators to discover programs that have produced the history and are producing the future of our world. In contrast, the MQALA worldview would direct investigators to collect data movies in accord with the general thrust of data driven discovery science. Then investigators would develop data movies by using DataSpeaks Interactions® to measure patterns of regularity that describe mechanisms.
Perhaps the most pressing practical implication of the MQALA worldview, which recognizes that temporal contingencies matter, is that temporal contingencies need to be measured so that they can be investigated scientifically. This requires the application and development of data processing and communications infrastructures to collect and develop data movies. These infrastructures would help us develop a practical scientific understanding of the world. Development of this understanding would help shape worldviews and improve human welfare.
MQALA has additional consequences for agency and responsible agency. MQALA measures temporal contingencies between independent events and dependent events (Appendix A). I already pointed out the three types of mechanisms by which complex systems have been said to work - function, response, and agency.
Function describes mechanisms in which both the independent and dependent events are internal to systems. As such, function can be said to describe regulatory control or self-control.
Response describes mechanisms in which independent events, including treatments, are external to systems. Dependent events can be defined on conventional variables such as those that measure actions as well as measures of interaction that describe mechanisms. Measures of response that use measures of interaction as dependent variables describe how mechanisms are changed.
Of these three types of mechanism, response is most akin to the prevailing scientific worldview. The prevailing, deterministic scientific worldview generally appears to treat people as passive respondents and makes them victims of circumstance, controlled and subject of fate. People often are subject to control when they are treated by doctors, educated by teachers, and punished by courts. An emphasis on control often appears to be something that many people dislike about behaviorism.
In addition, MQALA explicitly recognizes agency. Agency describes mechanisms in which independent events are internal to systems and dependent events are external. External events include behaviors that are publicly observable including effects on measuring instruments. Agency includes effects on other individuals. Agency also includes effects on other mechanisms as when doctors prescribe drugs that up or down regulate physiological mechanisms in a manner that might benefit or harm health. This increased and explicit emphasis on agency appears to be a significant departure from prevailing scientific worldviews.
There appears to be ample evidence for agency. Consider how the land now occupied by New York City has changed over the last few centuries. Human agency affects environments. The questions are how and how much, not if. It seems fitting to account for agency explicitly.
But can agents be responsible? The MQALA worldview also offers a perspective on this.
There appears to be a difference between agency and responsible agency. Examples of agents include germs, drugs, digestive systems, people, and organizations of people. Of these, people, and organizations of people that have or include educated nervous systems apparently are more capable of being responsible agents. Responsible agency appears to be understandable in terms of hierarchical organizations of systems.
I already pointed out how systems can be investigated at different levels of organization and understanding such as physical, chemical, biological, psychological, social, and cultural. I emphasize this in Patent 6,516,288. I illustrated how interactions can be measured between events defined at different levels of systems organization. For example, I mentioned the neural control of behavior in which action at an apparently lower neural level may interact with action at an apparently higher behavioral level.
It also appears that higher levels may be able to act down on apparently lower levels without violating any laws of nature. People create engines not by violating the laws of nature but by applying the laws of nature. It appears that people accomplish such feats by controlling temporal contingencies, often with the aid of machines and devices. For the engine example, these contingencies can involve physical features such as fuel-air mixtures, compression ratios and ignition times that need to be controlled repeatedly, rapidly, automatically and reliably. As an aside, I suspect that MQALA could be used to make engine controllers more adaptable to variable contingencies such as weather and variations in fuel.
In contrast to engines that control temporal contingencies, some engineers help create structures such as bridges, dams, and buildings that are designed and built to withstand temporal contingencies.
Perhaps of particular relevance for the application of MQALA to investigate responsible agency is the way that MQALA accounts for patterns of dynamic interaction that may be the key to describing the workings of higher order phenomena. For example, I already described measurement and visualization of patterns of connectivity in brains. Such patterns may be central to understanding attention. Similarly, MQALA may be useful in describing the workings of other higher order phenomena such as self awareness, mind, and consciousness, which may be important to understand responsible agency.
Responsible agents can be held accountable by others for their effects. Internalized higher order phenomena that are part of the type of mechanism called function help make people self aware and responsible for themselves.
Some higher order phenomena such as self awareness may be less like objects that can be located in space as with brain activation investigations but more like particular types of perceived patterns of events in space and time. Science itself has been described as finding patterns. DataSpeaks Interactions® measures and visualizes patterns in space and time. Apparently unlike Wolfram’s A New Kind of Science, DataSpeaks Interactions® can enable research programs for investigating certain classes of higher order phenomena. Such research programs can be initiated now. The first steps are to apply DataSpeaks Interactions® to data movies of brains in action as already described.
The section on behavior modification described how MQALA can account for an important form of adaptation. Adaptation that crosses certain thresholds appears to account for emergence of species as well as higher order phenomena. Responsible agency might represent a type of emergence in complex adaptive systems. MQALA does appear to contribute to “an expanding horizon of possibilities” that should be researched.
When I first said that contingencies matter in the Call for Leadership, I referred to the achievement of landing the Spirit and Opportunity rovers on Mars. This achievement provides an opportunity to contrast two scientific worldviews - the prevailing and the MQALA views.
The prevailing scientific worldview works best and is validated primarily for what it can achieve by application of scientific understanding in the physical sciences. This understanding helps us most with physical phenomena such as those that involve trajectories, propulsion systems, and what the physical composition of objects on Mars can tell us about the history of our universe.
But all of this is just part of the story of the great landing achievements. In addition to the successful landings themselves, we need to account for the behavior, motivation, passion, know-how, intelligence of the people as well as the organizations, teams, and the culture that made these achievements possible. The Mars landings are accomplishments of complex adaptive systems at work.
Great achievements can be contrasted with great disasters such as the Columbia shuttle and the World Trade Center disasters. Better scientific understanding of complex adaptive systems can make for more achievements and less disasters.
Nobel Prize winners are honored as if they were responsible. However, the prevailing scientific worldview appears to be inconsistent with responsible agency. Anticipate that MQALA and DataSpeaks Interactions® will enhance the value of honor as the software helps us to understand complex adaptive systems scientifically.
People and humanity have many choices when the world is viewed as “an expanding horizon of possibilities.” Instead of merely seeking to know the world as it is, we have the choice of accepting responsibility for creating the future. Hopefully we will pursue this future in accord with fundamental human values and continue to seed the world with intelligence.
Since DataSpeaks.com is opening with a call for leadership, I will take this opportunity to make a few points about leadership in the context of responsible agency. The first comments apply to leaders who are potential customers and might be interested in certain benefits of DataSpeaks Interactions®.
Leaders who, for example, might want to hold clinicians responsible for the benefit/harm of treatments, the measurement of which is illustrated in Appendix A, would have a responsibility to help provide conditions so that clinicians can measure the benefit/harm of treatments. This includes making DataSpeaks Interactions® available as part of an adequate data collection, processing, and communications infrastructure. Similarly, leaders who want to hold people more responsible for their own health to reduce cost burdens on collective payers have a responsibility to help provide the required infrastructure. In both cases, this would be similar to our general acceptance of responsibility to provide schools and teachers for educating our children (no slight intended) so that children can grow up to be responsible and effective agents.
Punishments of inadequate, dumb, and untoward behaviors are not apt to be sufficient when conditions for the desired alternative behaviors are not available. It can be mean for knowledgeable leaders to hold people responsible without providing the conditions for them to behave more responsibly. Leaders, including politicians, have obligations to help provide suitable conditions. Leaders that execute on plans in accord with their knowledge are apt to be rewarded according.
The remaining comments apply to potential leaders of DataSpeaks, Inc.
DataSpeaks is giving potential leaders opportunities to be responsible agents by leading DataSpeaks, Inc. It is in this context that I will quote James Clerk Maxwell from his essay: Determinism and Free Will (1873). This and related quotes appear on the Web site of the Princeton Plasma Physics Laboratory (http://w3.pppl.gov/~hammett/Maxwell/freewill.html).
Maxwell may be best known for his four partial differential equations of electromagnetism. The following quote suggests that he also anticipated chaos and offered some good advice for potential leaders. I put the last line in italics to emphasize his good advice.
Quoting Maxwell: “For example, the rock loosed by frost and balanced on a singular point of the mountain-side, the little spark which kindles the great forest, the little word which sets the world a fighting, the little scruple which prevents a man from doing his will, the little spore which blights all the potatoes, the little gemmule which makes us philosophers or idiots. Every existence above a certain rank has its singular points: the higher the rank the more of them. At these points, influences whose physical magnitude is too small to be taken account of by a finite being, may produce results of the greatest importance. All great results produced by human endeavor depend on taking advantage of these singular states when they occur.”
The discovery of MQALA appears to be a “singular state” that I trace back to work I did while attempting to create a way to analyze health diary data over 20 years ago. I pursued this effort after reading that no good method existed - an opinion offered by a leading health diary researcher who also had an appointment in a statistics department.
Much like the examples cited by Maxwell, MQALA appears to have unanticipated implications that extend far beyond the triggering event. (For example, I started to write a 3-page document. See what happened on this Web site.) These implications deserve to be pursued. Now impact and success of DataSpeaks Interactions® depends primarily on leadership.
I have encountered many sources of resistance to MQALA and DataSpeaks Interactions®. I have learned from some of them. I will mention some sources of resistance. These points will affect the way I will spend my time and help choose leaders for DataSpeaks, Inc.
I already described my experience with the statistical establishment. I will avoid repeating some of these experiences. Nevertheless, I would be honored to publish with expert statisticians and experts in other methods of empirical induction once they express real interest in MQALA and are willing to share the responsibilities and rewards of leadership.
I used to think that demonstrations of the DataSpeaks Interactions® would help sell the software. Based on this thinking, I spent over $100,000 out of my own pocket to develop the prototype. So far, I have been wrong.
The limited effect of demonstrating DataSpeaks Interactions® has become quite understandable. I already made reference to the discovery of germs. I know little of the history of those who invented, developed, and used microscopes. But I can imagine what these inventors might have gone through. Most people upon seeing germs for the first time, especially around the time that germs were being discovered, probably would say “So what?” and go about their routines. But a few people kept looking, working, and understanding more. It took time for the microscope market to develop.
We have been able to see germs and parasites for a long time, but the work continues as we still seek to develop practical scientific understanding. Understanding developed so far and the products of this understanding have extended millions of lives. Great accomplishments often require persistent hard work. It may take time for the market for DataSpeaks Interactions® to develop. But it might also be possible to create an avalanche effect.
I suspect that most people who look at Appendix B will say “So what?” The physician/researcher who asked why it was important to measure the benefit/harm of treatment was asking the “So what?” question. The answer to the “So what?” question is that measurement enables scientific investigations and practical applications of results to improve economic productivity and human welfare. I’m looking for a few good people with curiosity, guts, know-how and resources to actually work with the software so that they can answer the “So what?” question to their own satisfaction. Such thought leaders will make the market grow.
DataSpeaks Interactions® makes interactions or temporal contingencies visible in ways that they have never been seen before. Compared with germs, interactions are more abstract and potentially more difficult to appreciate when seen. But if a few leaders succeed, many other people are apt to follow. That would help create a market. The first and best leaders will have the biggest advantage.
Development and use of the prototype software did help solidify proof of concept for me so that I could continue working to advance DataSpeaks Interactions® with passion and vigor.
Scientists who investigate complex adaptive systems often speak of interactions. But scientists who really need to measure interactions in order to investigate their subject matter more scientifically and advance their careers generally defer me to statisticians. This behavior has become a reliable predictor of failure.
Statisticians do not measure interactions or temporal contingencies over time and across variables for individuals anymore than they are primary creators of, as examples, laboratory tests in medicine or microarrays. This falls outside the lens of their experience. Statisticians are good at analyzing the values of measures, once variables are measured, to describe groups and to make inferences from representative samples to populations. DataSpeaks Interactions® itself is validated by demonstrations such as those shown in Appendix B.
Potential business leaders that I have met locally appear to be waiting for someone else to lead by expressing demand in a way that would validate DataSpeaks Interactions® as part of a business concept. Business concepts do need to be validated. But leadership to provide validation of business concepts may require guts, curiosity, persistent hard work, and actually working with the software through scientific, technical, and academic thought leaders that business leaders trust.
I have worked hard and know that I need help. Now I am casting a bigger net on the Web with DataSpeaks.com. Many people have suffered and died because of some combination of my ineffectiveness as a leader and the unresponsiveness of my audience. (See my discussion of levels.)
The extensive material on DataSpeaks.com provides many opportunities for me to be found wrong on various particulars. But if DataSpeaks.com includes some modicum of fundamental innovation and original truth, DataSpeaks Interactions® deserves to be tested. I have already experienced people who have tried to nit pick me to death, which generally is understandable but not productive. I would prefer to avoid reliving these experiences.
DataSpeaks Interactions® is quite easy to test. I already described the brain visualization example. Fortunately, DataSpeaks Interactions® is software, not a biopharmaceutical that needs to be manufactured and tested in clinical trials. Lives and people’s health do not need to be put at risk. In addition, DataSpeaks Interactions® can be applied to other people’s data. Data are readily available. All this minimizes business risk. Much additional work needs to be done, including continued development of the intellectual property portfolio. But research and development costs can be controlled.
I do enjoy engaging people on the issues. I want to engage thought leaders with resources and who are personally invested in their own data for projects that can lead promptly to concrete results that will advance DataSpeaks Interactions®. At the same time, I want to engage business leaders who can help make DataSpeaks, Inc. into an outstanding and successful company.
There are some people who will avoid DataSpeaks because I lack sufficient authority on the weighty matters discussed on DataSpeaks.com. To these people I offer a timely reminder. Lord Kelvin said "Heavier-than-airflyingmachinesareimpossible." Other authorities echoed the same opinion. Authorities can be wrong. The Wright Brothers, working from a bicycle shop, helped prove that some authorities were wrong.
6.8. Reinvigorating Machine Learning and Artificial Intelligence
Section 4.2.6.4 of Patent 6,317,700 includes specifications for a demonstration learning robot. These specifications are an extension of material presented and referred to in the behavior modification section. This demonstration would just begin to show what is possible with respect to applying DataSpeaks Interactions® to create artificial learning systems.
It may be relevant to note that I had considerable difficulty with the patent specifications for the demonstration learning robot until I introduced the distinction between long-term and short-term memory. This is an important distinction in real brains.
The Sony AIBO (http://www.us.aibo.com/) might provide one way to pursue this market opportunity. The first generation AIBO was released in 1999, well after the patent specifications were written. AIBO has sensory, motor, and computer processing capabilities that go far beyond those mentioned in my patent specifications. It is possible that a version of DataSpeaks Interactions® could be made available to AIBO enthusiasts to promote both AIBO and DataSpeaks Interactions®. DataSpeaks Interactions® may well have the capability of enhancing AIBO’s ability to modify its behavior in accord with its experience in its environment. AIBO could be a good platform for demonstrating applications of MQALA for machine learning.
DataSpeaks Interactions® embodies MQALA, a computational method of empirical induction. Methods of empirical induction are methods for drawing generalized conclusions and making predictions from data. The ability to draw generalized conclusions and make predictions from experience is part of intelligent behavior.
DataSpeaks Interactions® can make an important contribution to artificial intelligence. I have given a number of examples of how DataSpeaks Interactions® uses computation to do that which people often do in their heads. As examples, I described that people and other organisms often learn from “what follows what,” one way of describing temporal contingencies. I described how clinicians can form subjective impressions about the benefit/harm of treatments from information about how individual patients respond to drug challenge, de-challenge, and re-challenge. These are examples of behavior often considered to be intelligent.
The section on predicting economies and capital markets provided additional examples of how DataSpeaks Interactions® helps with achievements generally considered to require intelligence. For example, DataSpeaks Interactions® can go beyond charting software by measuring potentially predictive interactions that people generally try to judge in their heads. Appendix B provides some examples of measuring interactions involving economic data.
The predictive indices feature of MQALA that is described in Patent 6,317,700 covers various aspects of intelligent behavior. The feature starts by measuring interactions and identifying conditions that provide the most predictive power. It combines information from multiple predictors into a single predictive index. It can measure predictive performance and adapt to improve predictive performance. It has the potential to differentially weight predictors in accord with predictive power and select optimal subsets of predictors. The performance of the whole system adapts automatically as relationships between predictor and predicted variables change as economies respond to external conditions including economic policies. DataSpeaks Interactions® helps make it possible to predict economic variables without knowing economics.
Given various scandals such as those involving security analysts and equity analysts, it often may be valuable to let the data speak in accord with objective, reproducible, transparent operations that can be specified in protocols. This would raise some professional standards of operation to be more like scientific standards of operation.
I already described how DataSpeaks Interactions® can be used to help visualize and understand how brains work, change, and adapt. I anticipate that this understanding can help inform the development of artificial systems for machine learning and artificial intelligence. I also anticipate that development of artificial systems can enhance understanding of real systems much as mathematical models can aid understanding. Similarly, DataSpeaks Interactions® could contribute to a mutually beneficial relationship between systems biology and synthetic biology.
Much of empirical science can be described as learning from experience recorded as data. However, at this time in history, there appears to be a major chasm between the methods of formal science and the methods of natural learning. Most of our brains do not work by performing mental t-tests and analyses of variance. In contrast, we often learn from temporal contingencies. MQALAcan help bridge the chasm between the methods of formal science and the methods of natural learning. Perhaps this, more than anything else, will help get us out of data swamps.
It is not clear how scientists and practitioners will fare after designed systems start doing more of the intellectual heavy lifting that helps keep us busy now. But in the meantime, there are a lot of discoveries to be made and services to be provided that can improve economic productivity and the general welfare.
I will make some key acknowledgements in advance just in case this call for leadership succeeds.
First and foremost, I would like to acknowledge my wife, Cathy Gofrank, and my extended family for putting up with my distractions and being supportive of me over the years. Special thanks to my children, Stephon, Alexander, Curtis, and Christie who are doing great despite my distractions. Recently Christie, age 10, found a “Payday” sticker and put it on a piece of paper to me between “When is” and a question mark.
I acknowledge Farideh Bagne, mother of my two oldest children, who helped fire my ambition but lost patience with my efforts.
I acknowledge Dr. Gordon Guyatt whose publication of an article on randomized N-of-1 clinical trials in the 1986 New England Journal of Medicine gave me hope during a difficult time.
I acknowledge Mickey Katz-Pek of Biotechnology Business Consultants for her support and assistance to me and other entrepreneurs. I acknowledge Robert Palmerton for assisting during the early unsuccessful years of trying to turn this technology into a business.
I acknowledge Jill Goldberg who worked so well with me to engineer the prototype software as a work for hire while she was at Cognitive Bionics.
I acknowledge David H. Brenner and Tom Edwards of IdeaWorks LLC for one long meeting that inadvertently initiated most of the material on this Web site. David chairs the 2003-2004 Great Lakes Entrepreneur’s Quest.
I acknowledge Phillip Covington for helping to make this Web site a reality.
Finally, I acknowledge Google. Browsing creates contingencies that have helped advance my work, reinforcing my conviction that contingencies matter. Often I have been amazed by what I fail to find. Why aren’t people using apt phrases such as “computational measurement software,” “temporal contingency analysis,” “multiple N-of-1 clinical trial design,” “advancing responsible agency” and “educated nervous systems” along with other such phrases that are a prominent part of DataSpeaks.com? Why are there about 33,300 hits for “complex adaptive system” but no hits for “simple time invariant systems,” which appears to be a reasonable contrast? Anyway, thank you, Google!
How to Develop Data Movies - A Primer on How DataSpeaks Interactions® Works
MQALA and DataSpeaks Interactions® work by a basically simple, iterative process. I’ll start by describing the data that are to be processed. I have referred to this process as developing data movies and as temporal contingency analysis.
A data movie consists of a chronological series of data snapshots. Each snapshot corresponds to a set of values obtained from measurements made at particular times or measurement occasions. Each picture element or pixel corresponds to one variable. Ideally, all variables relevant to particular systems of interest and types of operation are measured often and periodically. Generally, if you don’t know if a variable is relevant, try to include it to the extent that data collection and processing resources allow - results obtained with DataSpeaks Interactions® can help users determine if and how variables are relevant.
DataSpeaks Interactions® measures interactions or temporal contingencies. Measurement of interactions requires at least two variables or sets of variables. However, data movies of many systems can easily have hundreds or thousands of variables. DataSpeaks Interactions® can be applied to hundreds or thousands of variables.
DataSpeaks Interactions® works by two major steps. First, it uses the data to determine the presence or absence of discrete events at particular times over periods of time. This contrasts with much conventional software that continues to work with dimensional variables. Discretization is crucial. We need to define events in order to measure temporal contingencies between events that describe how systems work, change, and adapt.
Second, DataSpeaks Interactions® uses probabilities and other simple mathematical operations to measure the temporal contingencies between discrete events for two variables or sets of variables. This process can be repeated millions of times to cover all interactions of interest. It helps that key measures, which quantify the amount and direction of evidence for temporal contingencies, are standardized with respect to all possibilities given the data and a specified scoring protocol. After this, DataSpeaks Interactions® works primarily by summarizing and displaying the results.
Here are a few more details. DataSpeaks Interactions® essentially begins by using simple binning processes to define potentially hundreds or thousands of types of discrete events on each variable or set of variables. For example, with DataSpeaks prototype software, I commonly use up to eight analysis parameters simultaneously to define discrete events. Additional parameters are possible. Such parameters account for levels of the variables, temporal aspects such as delay and persistence of any interaction, and the episodic nature of many events. Discrete events are determined to be either present (1) or absent (0) on most measurement occasions to form a time series of ones and zeros for each type of event.
Our experience in going from analog to digital devices suggests that discretization or digitization of data by forming series of ones and zeros to represent the presence and absence of discrete events is a reasonable option. The process need not result in any loss of information. Although analysis parameters can have many levels to account for all of the information in the data, this often becomes a waste of computational resources after analysis parameters are represented with about 7 to 12 levels. The binning processes used here suggest a new form of digital data processing in the temporal domain.
Here is a simple example of defining treatment and health events starting with the assumption that an investigator has a data movie with only two variables, daily drug dose and daily blood pressure, over 100 days for an individual patient. Assume that the data were collected from a randomized N-of-1 clinical trial with a range of at least several different doses. One type of treatment event is present, for example, if daily dose is 50 or more on at least 5 out of 7 consecutive days. A type of adverse health event is present if systolic blood pressure is 140 or more on at least 2 out of 5 consecutive days.
Users don’t have to know in advance what levels define events that account for the strongest interaction because users can evaluate hundreds or thousands of types of such events simultaneously. Users can select additional analysis parameters to investigate delay and persistence of apparent drug response. If in doubt about including or adding an analysis parameter or increasing the number of levels of any analysis parameter, do so with certain caveats to the extent that computing resources allow. Results will help tell users how all selected analysis parameters and levels might be relevant to any interaction.
After discretization, DataSpeaks Interactions® can measure the interaction between the two variables, drug dose and blood pressure, for all selected types of discrete events. Assume that in this example, the user would set DataSpeaks Interactions® to indicate that higher levels of blood pressure were bad. One step is to cross-classify all the series of ones and zeros for dose with all the series of ones and zeros for blood pressure. This yields an array of 2 x 2 tables. The array would have the same number of dimensions as analysis parameters that the user selected. Such arrays can easily have thousands of entries as illustrated in Appendix B.
Each 2 x 2 table in an array for this example is used to compute values of measures (scores) for each location in the array. Each score measures either the strength or the amount of evidence for an interaction under the conditions specified by the location of the score in the array. Positive scores would indicate apparent benefit. Negative scores would indicate apparent harm.
The primary scores in this example quantify the amount of evidence for benefit and harm. Each of these scores is standardized with respect to all possible 2 x 2 tables given the marginal frequencies of the particular observed 2 x 2 table. DataSpeaks Interactions® currently standardizes scores to have mean 0 and standard deviation 1. Appendix B shows that this procedure can yield large positive and negative scores with low probabilities of occurring by chance alone. In contrast, the strength scores are not standardized and can range in value from -1 to +1.
Arrays of standardized scores are easily summarized by selecting extreme values to identify the conditions that provide the most evidence for interactions. Interactions can be summarized and visualized as functions of any analysis parameter such as dose or any combination of analysis parameters such as dose and delay.
DataSpeaks Interactions® is a data processing program that both analyzes and synthesizes data. Analysis and synthesis become two aspects of the same process. Analysis here involves describing interactions in detail. The arrays of scores, which easily can have thousands or tens of thousands of interaction scores, allow results to be analyzed in great detail. Synthesis involves summarization of arrays of standardized scores to draw generalized conclusions in accord with DataSpeaks Interactions® being a software system based on a method of empirical induction, MQALA, as described in the patents.
Synthesis can be extended across dependent variables. Extend our “drug for blood pressure” example, by assuming that 20 health variables were measured daily to help evaluate safety and efficacy in the same patient. All interaction scores and all summary interaction scores can be differentially weighed in accord with clinical significance or patient preferences and averaged to draw generalized quantitative conclusions about benefit/harm over all of the dependent variables for the patient that was investigated.
Generalization can be extended from individual patients to groups and populations by applying the statistical method to corresponding interaction or benefit/harm scores from two or more individuals. This is illustrated in Appendix B. Thus the method used by DataSpeaks Interactions® and the statistical method often can be complementary methods for drawing generalized conclusions about groups and making inferences from samples to populations. In such cases, measurement with DataSpeaks Interactions® comes before statistical analyses.
To continue with our “drug for blood pressure” example, assume that we have data for a sample of 50 patients - 100 repeated measurements of dose and each of the 20 health variables for each patient. I call this the randomized multiple N-of-1 clinical trial design where “multiple” refers to patients. The results could be evaluated statistically with a single group t-test on mean overall benefit/harm scores. Rejection of the null hypothesis in the positive direction would indicate benefit. Rejection in the negative direction would indicate harm. One reason why this approach can work so well is that it reduces the number of variables that need to be analyzed statistically from 21 (one treatment variable, dose, and 20 health variables) to 1. Of course, benefit/harm can be profiled across all 20 health variables for each patient, for any sub-sample or the entire sample of patients. Similar use of the t-test is shown in Demonstration 1.
DataSpeaks Interactions® has additional features, described in Patent 6,317,700, to account for phenomena such as those involving sets of independent variables acting in concert (e.g., drug interactions, protein complexes). These features work by applying Boolean operators to each of the series of 1s and 0s for one variable with each such series for one or more additional variables in a set. For example, a Boolean independent event can be said to be present when protein A is present and either protein B or protein C is present and protein D is not present. In addition, Boolean events can be defined on sets of dependent variables to investigate phenomena such as syndromes and the effects of master controller proteins in biological networks. Such events can be defined across all levels of the analysis parameters such as those previously described. In this manner, DataSpeaks Interactions® can be used to investigate complex events.
Other features of DataSpeaks Interactions® can measure changes in interactions over time. These changes indicate changes in the amount, strength, and direction of evidence over a period of time. In addition, such changes can indicate development, aging, learning, habituation, sensitization, potentiation, and other time dependent processes.
DataSpeaks Interactions® can be used to investigate interactions involving variables at different levels of understanding (e.g., laboratory values, symptoms, health perceptions, and quality of life; biological, psychological, and social). As such, it can foster interdisciplinary and collaborative investigations.
DataSpeaks Interactions® is demanding of computing resources. An important challenge might be to marshal enough computing power to process data on a large scale. We can turn this problem into an opportunity if we want to get help from companies that sell and service computing infrastructure including grid computing.
The patents provide more information about the operational details of DataSpeaks Interactions®.
Three Proof-of-Concept Demonstrations
Appendix B includes brief descriptions of portions of three MQALA proof-of-concept demonstrations that were performed with DataSpeaks Interactions® prototype software. Appendix A is a brief primer on how MQALA works.
Demonstration 1 involves reproductive endocrinology. Demonstation 2 involves economic time series. Demonstration 3 involves functional brain image analysis.
Procedures similar to those demonstrated here can be applied to measure interactions, temporal contingencies or longitudinal associations to describe, elucidate, and visualize mechanisms that describe how these and many other types of complex systems work, change, and adapt.
Although each of these demonstrations presents results for data movies with only a few variables, MQALA can be applied to investigate hundreds or thousands of variables simultaneously as described for action coordination profiles. As examples, DataSpeaks Interactions® can be applied to time series gene expression microarray data, repeated measurements proteomics data and to functional brain imaging data for whole brain slices and brains.
These demonstrations yield results unlike those that most viewers have ever seen before. This can make it difficult for viewers to appreciate what thay see as discussed before in the context of seeing germs for the first time and in the context of interactions being more abstract than germs. Nevertheless, measurement and visualization are fundamental to practical scientific understanding.
Demonstration 1 - Reproductive Endocrinology
The first demonstration was performed to validate MQALA by showing that MQALA detects known interactions involving reproductive hormones. This was the first major demonstration performed with DataSpeaks Interactions® prototype software. Demonstration 1 is particularly revevant to market opportunities that involve reviatalizing the pharmaceutical industry and reforming health care.
The portion of Demonstration 1 that is presented here focuses on only two variables, namely the levels of two hormones over time. The same methodology can be applied to investigate how many types of action are coordinated in many types of complex systems.
Data for this demonstration are described and reported in two publications. The first publication is: Padmanabhan, V., McFadden, K., Mauger, D.T., Karsch, F.J., and Midgley, A.R. (1997). Neuroendocrine control of follicle-stimulating hormone (FSH) secretion. 1. Direct evidence for separate episodic and basal components of FSH secretion. Endocrinology 138, 424-432 (http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmd=Retrieve&db=PubMed&list_uids=8977432&dopt=Abstract).
The second publication is: Midgley, A.R., McFadden, K., Ghazzi, M., Karsch, F.J., Brown, M.R., Mauger, D.T., and Padmanabhan, V. (1997). Nonclassical secretory dynamics of LH revealed by hypothalamo-hypophyseal portal sampling of sheep. Endocrine 6, 133-143 (http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmd=Retrieve&db=PubMed&list_uids=9225127&dopt=Abstract).
The research for both of these publications was conducted at the University of Michigan. I am grateful to authors that provided access to data described by these publications.
This description of Demonstration 1 presents results for only one interaction, namely the interaction between gonadotropin releasing hormone (GnRH) functioning as the independent variable and portal luteinizing hormone (P-LH) functioning as the dependent variable. Hormone levels were measured every 5 minutes for about 6 to 12 hours for each of 6 ewes. Additional information about this demonstration is available in the form of the original PowerPoint presentation. Supplementary material for this presentation identifies abbreviations and scoring protocols in addition to providing a few more detailed numerical results.
The longitudinal association score that was obtained with DataSpeaks Interactions® and quantifies the amount and direction of evidence for the GnRH to P-LH interaction in one ewe, based on 143 repeated measurements of both hormones, is 76.728. This observed score is one score from a standardized distribution (mean zero, standard deviation 1) of potential scores, derived by applying MQALA to the data. The probability of obtaining this score by chance alone in its distribution of potential scores is .0000000000000000000000000000000000228366. With this probability and the fact that the score is positive, 76.728 is considered to provide substantial evidence for the expected positive interaction between GnRH and P-LH for an individual ewe.
Temporal contingency analysis has been described as "what follows what." The score with value 76.728 means that episodes or pulses of P-LH are very apt to follow episodes or pulses of GnRH.
A two-tailed single group t-test on the mean of the strength of longitudinal association scores was used in an attempt to reject the null hypothesis of no GnRH to P-LH interaction across the 6 ewes. The null hypothesis was rejected with p<.0001 in favor of a positive interaction. Since the number of repeated measurements of hormone levels varied substantially across ewes, this test was performed on a measure that quantifies the strength, rather than the amount, of evidence for longitudinal associations. The value of the strength of longitudinal association measure that corresponds with 76.728 is .938. This illustrates the complementary use of MQALA and the statistical method.
Results both for the individual ewe and the group of 6 ewes help validate MQALA by showing that it is highly sensitive to a known interaction. Similar supportive results were obtained for other interactions.
These results were obtained with summary scores from a scoring protocol that yielded an 8-dimensional array of 86,400 standardized scores per interaction per ewe. Each array is a detailed quantitative description of an interaction, what follows what. Each array dimension corresponds to an analysis parameter.
The array for any particular ewe/interaction combination can be summarized to investigate the interaction as a function of any analysis parameter or any combination of analysis parameters. The PowerPoint presentation includes slides that show average strength of interaction across the 6 ewes as a function of delay for all interactions investigated in Demonstration 1.
The location of 76.728 in the array identifies the conditions, defined by the analysis parameter levels, which provided the most evidence for the interaction. Furthermore, entire arrays or any summary of arrays for the individuals in the group of ewes could be analyzed statistically to describe particular interactions for the group. If the group of ewes is considered to be a representative sample, statistical inferences can be made about the population.
In brief, the number 86,400 was derived from a protocol of user selected scoring options. Data for the independent variable, GnRH level, were used to define 720 types of discrete events. Each type of event was determined to be either present or absent on most of the measurement occasions for each ewe. The ewe that yielded the interaction score of 76.728 had 143 samples drawn, one every 5 minutes for almost 12 hours.
Each type of independent event corresponded to one combination of five analysis parameters used simultaneously. These analysis parameters were independent variable level (12 levels), 10 combinations of episode length and criterion that were used to account for pulsatile secretion, as well as 3 levels of delay and 2 levels of persistence to account for temporal aspects of the interaction (12 x 10 x 3 x 2 = 720). In the context of the "what follows what" description of temporal contingency analysis, each of these 720 types of events is one type of independent event.
Similarly, data for the dependent variable, P-LH level, were used to define the presence or absence of 120 types of discrete events using 12 values for dependent variable level and the same 10 combinations of episode length and episode criterion that were used for GnRH. All types of events were defined on the standardized residuals from linear regression lines of the variables on time or measurement occasion. In the context of the "what follows what" description of temporal contingency analysis, each of these 120 types of events is one type of dependent event.
Each entire array of 86,400 standardized scores for each directional interaction between two hormones in Demonstration 1 describes "what follows what" in considerable detail. These descriptions describe biological mechanisms.
Dichotomous series corresponding to the presence or absence of each of the 720 types of independent event were cross-classified with the dichotomous series for each of the 120 types of dependent events to yield 86,400 2 x 2 tables. Each 2 x 2 table was used to compute values of the various measures of interaction, longitudinal association or temporal contingency. This procedure of going from dimensional variables to dimensional variables (interaction scores) through discrete events appears to have fundamental scientific significance.
Demonstration 2 - Economic Time Series
Demonstration 2 was performed on a data movie comprised of 12 economic time series, namely the 10 variables used to compute values of the Index of Leading Economic Indicators (LEI), Gross Domestic Product (GDP) of the United States, and values of LEI itself. LEI is a well known standard (http://www.tcb-indicators.org/) that is included here to help readers validate MQALA. This demonstration used quarterly data for 166 consecutive quarters beginning in 1959. Other economic, capital market, and social time series data could be processed in similar ways. The market opportunity section for economies and capital markets provides more information.
These economic data were analyzed using two DataSpeaks Interactions® scoring protocols. Scoring Protocol A is the same scoring protocol that was used in Demonstration 1 (extensive protocol) with one exception. Protocol A for the economic data defined events on the residuals from second order polynomial, rather than linear, regression lines. Protocol A used an 8-dimensional array to 86,400 standardized (mean 0, standard deviation 1) longitudinal association scores to describe each interaction in detail.
Use of similar scoring protocols for Demonstrations 1 and 2 helps illustrate the applicability of MQALA and DataSpeaks Interactions® to dynamic systems of many types. Many different scoring protocols could be investigated for both types of data. Users define different scoring protocols by selecting different DataSpeaks Interactions® scoring options.
Scoring Protocol B is the same as Protocol A except default values were used for episode length, episode criterion, and persistence. Default values are described in Patent 6,317,700. Furthermore, 7 values of delay, 0 through 6 inclusive, were scored. Given this selection of analysis parameter levels, only 84 (12 x 7) types of events were defined on predictors and 12 types of events were defined on the predicted variable, GDP. Protocol B in Demonstration 2 yields 1,008 (84 x 12) longitudinal association scores descriptive of each interaction.
Results for both scoring protocols are summarized in the following Table for Demonstration 2.
First, I will present, compare, and discuss results for the LEI to GDP interaction. For Protocol A, the summary longitudinal association score for the LEI to GDP interaction is 65.410. The probability of obtaining this score by chance alone in its distribution of potential scores is .0000000000000000000000000000210206. The strength of this interaction is .698. With this probability and the fact that the measures are positive, 65.410 and .698 are considered to provide substantial evidence for the expected positive LEI to GDP interaction. LEI is predictive of GDP. Episodes of high GDP are apt to follow episodes of high LEI.
For Protocol B, the summary longitudinal association score for the same interaction was 51.013. The probability of obtaining this score by chance alone in its distribution of potential scores is .00000000000000000000000527054. The strength of this interaction is .541.
The results from both scoring protocols help validate MQALA and DataSpeaks Interactions®. Such results help provide proof-of-concept for the technology behind DataSpeaks Interactions®, if not the business concept.
Results for all interactions in the Table are more extreme for Protocol A than for Protocol B. Here are two explanations for these differences. Both explanations appear to be true to some extent.
First, Protocol A yields more extreme results in part because it summarizes more scores (86,400) than Protocol B (1,008). Summarizing more scores can yield more extreme results for a particular individual by chance alone.
The second explaination is that Protocol A yields more extreme results in part because it uses more optional analysis parameter levels to account for systematic variation between the predictors and GDP. Although I do not know how to compute exact probabilities for getting any particular result by chance alone in a manner that accounts for the number of scores summarized with a MQALA scoring protocol, here are three relevant thoughts and strategies about addressing this issue. Many related issues need additional investigation.
First, the results obtained from both protocols are so extreme that Bonferoni type corrections are not apt to nullify conclusions that the interactions are real. Furthermore, results for both the hormone and the economic data are in accord with expectations based on substantial experience.
Second, actual probabilities for getting any particular result by chance alone, which account for the number of scores summarized by a protocol, could be estimated with randomization tests.
Third, summarizing a large number of scores for particular individuals is not apt to be a problem when MQALA is used to measure interactions before the statistical method is used to describe groups and make inferences about populations as illustrated for the six ewes in Demonstration 1. The reason for this not being a problem is that individuals with positive MQALA summary scores tend to balance out individuals with negative MQALA summary scores when there is no interaction between the independent and dependent variables for any individuals in groups. This “balancing out” makes it harder to reject null hypotheses of no interaction with statistical tests. I have demonstrated this with procedures that include use of the prototype software and an earlier version of the software to measure interactions between series of random normal deviates.
The Table includes interesting results for predictors in addition to LEI. The interactions between both unemployment rates and interest rate spreads with GDP are negative and substantial. All other interactions are positive. How does this relate to your expectations?
Of all the predictors, manufacturers’ new orders for consumer goods and materials is the strongest predictor of GDP with both scoring protocols. There is slightly more evidence for this interaction than for the GnRH to P-LH interaction described Demonstration 1 (77.280 compared to 76.728). However, the GnRH to P-LH interaction was somewhat stronger (.938 compared to .827). The main reason why there can be more evidence for the weaker interaction is that analyses of the economic data were based on 166 repeated measurements while the analysis of the hormone data for the ewe that yielded a score of 76.728 was based on only 143 repeated measurements.
The Table for Demonstration 2 includes information about the combinations of levels for most analysis parameters that yielded the summary interaction measures. LEI is most predictive of GDP two quarters in advance with Protocol B. This appears to be in accord with the way LEI was designed to function, predicting GDP about 6 months in advance. The corresponding result for Protocol A is that delay equals 0. The latter result appears to be due to the fact that Protocol A also accounts for other parameters that involve time - episodes and persistence - and that most of these levels are not at default levels of these analysis parameters.
I examined the measures of interaction as functions of delay for Protocol B. Different predictors have strikingly different functions. Note the difference in results for the analysis parameter levels for manufacturers’ new orders, consumer goods (MfCG) and materials compared with manufacturers’ new orders, non-defense capital goods (MfCap). Results for interest rate spread (Rate) are more similar to those for MfCap than for MfCG. How would you interpret these results? Could it be that MfCap and Rate are under more control by decision-makers in business and government?
Compared to the statistical method, MQALA would be a primary computational method of empirical induction for individuals such as the U.S. economy because such unique and inclusive individuals can not be sampled as individuals or described as groups. There is only one U.S. economy.
Summary of results for the proof-of-concept demonstration for the economic data, which uses various predictors of U.S. GDP.
Abbreviations for the columns are: LAS = Longitudinal Association Score; IVEL = Independent Variable Episode Length; IVEC = Independent Variable Episode Criterion; Persist. = Persistence; DVEL = Dependent Variable Episode Length; DVEC = Dependent Variable Episode Criterion.
Abbreviations for the rows are: MfHrs = Average weekly hours, manufacturing; Unemp = Average weekly initial claims for unemployment insurance; MfCG = Manufacturers’ new orders, consumer goods and materials; Vend = Vendor performance, slower deliveries diffusion index; MfCap = Manufacturers’ new orders, non-defense capital goods; Bldg = Building permits, new private housing units; Stock = Stock prices, 500 common stocks; Money = Money supply, M2; Rate = Interest rate spread, 10-year Treasury bonds less federal funds; CsExp = Index of consumer expectations; LEI = Index of Leading Economic Indicators.
Demonstration 3 - Functional Brain Image Analysis
The Tables for Demonstration 3 provide a small representative subset of results obtained when MQALA was applied to functional magnetic resonance imaging (fMRI) data for a 2-row by 24-column patch of voxels from the motor cortex for one subject. The data were constructed as described in the publication at the following link: http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmd=Retrieve&db=PubMed&list_uids=10458943&dopt=Abstract . The data were obtained over the Web from a link that apparently is no longer available.
This is a very preliminary demonstration of an attempt to measure apparent functional connectivity in brains. The intent is to identify thought leaders with resources who will try applying MQALA to functional brain imaging data as described in one of the market opportunity sections.
The MQALA analysis used only a small portion of the data from the cited publication, namely data for the first two rows in the “simple signal” portion of publication Figure 1. These two rows were not affected by any embedded signal shown in Figure 1. This is in accord with the intent to measure apparent functional connectivity as distinct from activation. The data have 384 time-points.
Results from the MQALA analysis illustrated below show striking evidence for patterns in the data. One purpose of these notes is to encourage experts in fMRI imaging and analysis to help determine if these patterns are real. Might the pattern result from some aspect of how the test data set was constructed by authors of the article? Is the pattern biologically meaningful? Is it interesting?
Appendix A provides a brief primer on how DataSpeaks Interactions® works.
The data were analyzed with the “action coordination profile” portion of the DataSpeaks Interactions® prototype software. User specified options were as follows. Under “Transformation Options,” I selected “Linear Regression Residuals.” Under “Dimensional Resolution,” I selected “Fine Z Score (12 levels of resolution).” Under “Scoring Settings” and “Independent Variables,” I selected 2 levels (1 and 2) for “Episode Length” and checked “Analyze episode criterion” (a total of 3 combinations of levels: 1,1; 1,2; and 2,2). I selected 6 levels of “Delay” (0 through 5) and 2 levels of “Persistence” (1 and 2). Under “Scoring Settings” and “Dependent Variables,” I selected 2 values (1 and 2) for “Episode Length” and checked “Analyze episode criterion” (also a total of 3 combinations of levels). This means that I analyzed 15,552 (12x12x3x6x2x3) longitudinal association scores for each pair wise directional combination of one independent variable (voxel) and one dependent variable (voxel). There are 1856 (48x47) such combinations when there are 48 variables.
The analysis, illustrated in part by the Tables for Demonstration 3, took between 36 and 43 hours of to run on a Dell 750 MHz Dell laptop.
It would be far superior to show the results illustrated by the Tables for Demonstration 3 as a set of figures that could be presented interactively as described elsewhere. But we have reached the edge between the past and the future. The future depends largely upon identifying thought leaders with resources who could port DataSpeaks Interactions® to a higher performance computing environment, obtain suitable data for a whole brain slice or, better yet, a whole brain, and doing additional computer programming to convert numerical results such as those illustrated here into an interactive visual display.
Each portion of the Tables shows results for the 2 by 24 patch of motor cortex. In Table 1 and Table 3, the voxel identified by “x” functions as the independent variable and the 47 other voxels function as dependent variables. The letter “x” identifies the voxel selected by the user in the description of the visual display.
For Table 2, “y” identifies the voxel functioning as the dependent variable and the 47 other voxels function as independent variables. Table 2 shows and the corresponding visual display would show how brain activity in all other voxels is associated with brain activity in the selected voxel. In other words, Table 2 shows how activity in voxels other than the voxel identified with “y” is associated with or may affect activity in the voxel marked by “y.” Note that the portion of Table 1 for delay = 0 would have been identical to the corresponding portion of Table 2 if default values had been selected for Independent Variable Length, Dependent Variable Length, and Persistence.
Table 1 shows results for the first 4 voxels functioning as independent variables. Results for the remaining 44 voxels are similar. Table 2 shows results for the first voxel functioning as the dependent variable. Results for the remaining 47 voxels are similar.
Each summary longitudinal association score (LAS) shown in Table 1 and Table 2 is one score from a standardized distribution (mean 0, standard deviation 1) of potential scores that is defined by applying the MQALA algorithm to the data. In this case, each summary score across the 6 levels of delay, shown in the bottom portions of Table 1 and Table 2, is the most extreme positive or negative score in the 8-dimensional array of 15,552 scores for each pair-wise directional combination of two variables. Each summary score for each delay-specific portion of Table 1 and Table 2 summarizes 2,592 (15,552 / 6 levels of delay) standardized LASs.
I tabulated the Tables for Demonstration 3 by hand. The tables are subject to review for typing errors.
The Tables show strong evidence of how the data are patterned. Patterning is the basis of data mining and empirical scientific investigations generally.
The patterning becomes evident in several ways. First and as described above, each score is one score from a distribution of potential scores that has a mean of 0 and a standard deviation of 1. The portion of Table 1 in which voxel 2 functions as the independent variable includes a score of 206.4. Such large standardized scores have a small probability of occurring by chance alone.
All the scores shown in the Tables for Demonstration 3 are positive. Given that each score in Table 1 and Table 2 is from a standardized distribution with a mean of 0, this provides additional evidence for a pattern. In general, high levels of activity at particular times in any region are associated with higher levels of activity at particular times in other regions. The entire analysis did yield 4 negative summary scores at delay = 5.
Table 1 and Table 2 include delay-specific summary scores. Delay is one of the six optional analysis parameters that were used in this analysis to help account for temporal aspects of each measure of apparent connectivity. Of these, delay is most similar to the familiar procedure of lagging variables in relation to each other before doing some type of analysis. However, these delay-specific summary scores are summarized across the five other optional analysis parameters used to investigate temporal aspects of connectivity as well as level of the independent variable and level of the dependent variable. The entire analysis could be rerun much faster by dropping the other five temporal analysis parameters.
Results for delay = 0 show the most evidence for patterning in addition to that described above. In general, these results show two modest scores followed by one larger score. To me, this suggests layering. My understanding is that the motor cortex is layered.
The apparent layering pattern decreases with larger levels of delay. I see little evidence for layering at delay = 5. In additional, the delay = 5 scores tend to be smaller than those for any other level of delay.
Delay, together with reversing independent and dependent variables, can be used to help evaluate the temporal criterion of causal and other predictive interactions. Although these results provide strong evidence for coordinated activity, my first impression is that they provide little evidence that coordination that is causal. I would expect more evidence for causal relations when additional brain regions involved in cascades are included in analyses.
Here is an aside related to delay that may be a source of concern or interest. I understand that there is a time lag of about 1 to three seconds between neural activity and hemodynamic response. I do not think that this is a problem with MQALA. I will illustrate my reason with an analogy. Imagine that you are watching a live interview from China on CNN. The satellite delay can be annoying. But the order of the words is not changed - the whole response is just delayed. Similarly, my impression is that that MQALA can take advantage of high temporal resolution of functional imaging, especially if hemodynamic response time does not vary much across brain regions.
Table 3 shows results for one measure of direction and strength of longitudinal association as contrasted with the other measures that show amount of evidence for longitudinal association. Strength measures are ratios of the amount of evidence for a longitudinal association over the amount of evidence that the data could have provided for a longitudinal association under a particular condition as described in Section 4.1.6 of Patent 6,317,700. Values of the strength measures, which generally are not affected by the number of repeated measurements in time series, can range from -1 to +1.
Tables
for Demonstration 3
A small sample of representative results for functional brain image analysis.
Page last revised 3/2/04
|
|
 |
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| Home | Services | Products | Links | Contact Us |
 |
Copyright © 2003 DataSpeaks®, Inc. |