Susser has analyzed epidemiology's past (1) and finds this discipline currently in transition from an era employing a "Black Box" paradigm to an era of "Eco-Epidemiology" with a new paradigm (2). He admonishes us to choose directions for this paradigm which keep a central focus on Public Health. IIn that spirit, I elaborate on a new paradigm which is compatible with Susser's discussion. I then illustrate its value for the study of infectious diseases.
Epidemiology is in transition from a science which identifies risk factors for disease to one which analyzes the natural systems that generate patterns of disease in populations. The focus of epidemiology is expanding from relationships between exposure and disease variables to the analysis of the systems which give rise to exposures and through which those exposures act to cause disease. Our view of populations is being transformed from a collection of individuals to a set of interactions between individuals -- from an additive heap of risk factor effects to a non-linear system with multiple control mechanisms and leverage points. Dynamic systems models are supplanting fixed mathematical relationships for examining relationships between exposure and disease and predicting the effects of interventions. Issues regarding the conformation of systems are bringing a whole new set of phenomena under the purview of epidemiological investigation. These include contact patterns between individuals as well as the social structures through which individuals affect each other.
This transition is being advanced by the adaptation of methods developed in other disciplines. Eventually it may be further advanced by new developments in the analysis of complex adaptive systems (3) and by addressing questions regarding how disease systems evolve (4).
A possible future for epidemiology is discerned by considering a similar transition in biology. The identification and classification of species was at one time a dominant activity in biology. The role of this activity has been transformed through the formulation and evaluation of theories involving biological and ecological systems. Population biology, evolutionary biology, and ecology have become firmly established traditions as a result of this transition.
In the course of biology's transition, the identification and classification of species was not eliminated as an important activity. Rather it was transformed. Species identification and classification are now undertaken not only in the older context of descriptive biology, but also within a more comprehensive theoretical framework. Rather than only identifying species they encounter in the field, biologists now predict what new species should exist on the basis of evolutionary or ecological theory. Their examination of ecological and evolutionary relationships reinforces their species classifications.
Likewise the transition in epidemiology will transform rather than displace the activity of risk factor identification. Better theoretical structures will point our search for risk factors in more productive directions and decisions on causality will benefit by being put in the context of broader theory. But above all, we will address a new set of questions regarding the nature and behavior of systems.
The transition from risk factor detection to systems analysis is equally important for infectious and non-infectious disease epidemiology. The need for it has been made especially clear by the social epidemiologists. They point out how the old paradigms of epidemiological inquiry focus too much on factors which are identified by examining individuals and they provide rough outlines for new paradigms and methods that will provide more inclusive approaches to social epidemiology. Meanwhile, this transition in infectious disease epidemiology has advanced to the point where it can provide experiences and methods of use to other areas of epidemiology. An introduction to the analysis of infection transmission systems is provided by Roy Anderson and Robert May in their book entitled: Infectious Diseases of Humans: Dynamics and Control (7).
Black box era methods are founded upon an assumption that is inconsistent with the transmission of infection: namely that the outcome of exposure in one individual is independent of outcomes in other individuals (8,9). This inconsistency makes the detection of many risk factors impossible and distorts estimates of the effects of others (8,9). Why then do infectious disease epidemiologists use the black box approach? Two reasons stand out.
First, there are many personal behaviors and environmental contaminations causing infections that can be identified by the black box approach. For a good number of these, mere identification can lead to effective disease control activities. No analyses of how the risk factors act in a larger system and no quantitative predictions of effects from infection control programs are needed. This is the case when the benefits of eliminating a risk factor are clearly greater than the costs or when people will readily change their behavior to avoid a newly identified risk factor. Even though systems analyses methods might detect a greater range of infectious disease risk factors and be of greater value in designing efficient interventions, there are enough easily controlled risk factors that will be identified by black box methods to justify their continued use.
Second, few epidemiologists have acquired the skills needed to analyze transmission systems. Epidemiologists have essential skills which most mathematicians currently engaged in this task lack. They are familiar with the population behavior of infections in a variety of endemic and epidemic situations and they understand the nature and behavior of infectious agents and of immune responses. But they have not developed their ability to judge which epidemiological observations are most important from a systems point of view and to abstract those important elements into a mathematical or a computer model. The skills they need have been expressed by John Holland as follows (3):
"...in building" ... "a model, selection is critical." ... "The model (cartoon) can be more, or less, faithful to the original and, as always, which it is depends on the purpose of the model (cartoon). We may opt for simplicity, or even distorted similarity, at a cost in faithfulness, in order to emphasize some basic element. Newton, in building his models, ignored friction in order to get a more definitive look at momentum. His slightly unfaithful model emphasizes the principle that "bodies in motion persist in that motion, unless perturbed by forces." Aristotle's earlier, more faithful model implicitly included friction, leading him to enunciate the "basic principle" that "all bodies come to rest." Aristotle's model, though closer to everyday observations, clouded studies of the natural world for almost two millennia. Model building is the art of selecting those aspects of a process that are relevant to the question being asked. As with any art, this selection is guided by taste, elegance, and metaphor; it is a matter of induction, rather than deduction. High science depends on this art."
Transmission between individuals is an aspect of reality that should be explicit in the form of models used to study infectious diseases. Black box era models, however, do not employ parameters which reflect either contact between individuals or transmission. Consequently, two crucial determinants are often ignored: 1) the population pattern of who contacts whom, and 2) risk factors that reside in the infected rather than the susceptible individual.
The problem with ignoring contact patterns can be appreciated by imagining two almost identical populations. For both populations, suppose we have information from each individual on the exposures they experience that affect their risk of infection. This would include environmental contaminations, personal behaviors, physiological states, or anything that could be ascertained by examining either the individuals or their environment. Within populations, each individual might differ from each other individual. But suppose each individual in one population is exposed to exactly the same risk factors as a corresponding individual in the other population. That means that traditional epidemiological data sets from these two populations would be identical. Despite this, the two populations can have vastly different levels of infection transmission depending upon who has contact with whom (10,11). One population may have a core group which sustains transmission while the other may not. Black box era analytic methods, which conceptualize risk as an individual based phenomenon, cannot capture risk which is determined by how individuals are connected.
The problem with ignoring risk factors which are characteristics of infected rather than susceptible individuals can be appreciated by considering the inadequacies of black box era approaches to measuring vaccine effects. Vaccines stimulate immune responses that help control infection after the host takes up an agent . A vaccine induced immune response may control some agents before they have any consequences for the individual or for the transmission of infection. On the other hand, it may only decrease the contagiousness of infection without eliminating infection. Such a decrease in contagiousness could provide a crucial element in the control of transmission, especially for HIV infection, where a vaccine which prevents no infection could stop the HIV epidemic merely by reducing contagiousness during primary infection (12-14).
The black box paradigm doesn't recognize the effects of primary HIV infection for two reasons. First it is very difficult to ascertain exposure to someone with primary infection. Second, it is not just the infections directly caused by primary infection that generate its dominance of HIV transmission dynamics. Primary infection can dominate those dynamics even when much more virus is excreted in other stages of infection (12). That is because transmissions during primary infection are more effectively connected into chains which rapidly disseminate infection. The patterns of connection between individuals underlying those chains are ignored by the black box era paradigm. A transmission system analysis, on the other hand, makes the dominance of primary infection stand out clearly (12,16). Estimating transmission probabilities as a function of vaccination status in both the infected and susceptible individual provides a measure of vaccine effects on contagiousness during primary infection that is completely missed by standard methods (14).
How systems analysis captures important determinants missed by the methods of the chronic disease era can be further understood by considering the parameters estimated in an epidemiological analysis and how data is structured to estimate those parameters. Epidemiological data is traditionally structured to estimate parameters which relate exposure to disease in individuals, such as odds ratios or risk ratios. Dependent and independent variables are placed in columns and individuals are placed in rows. Not all data collected from individuals, however, falls into the class of dependent or independent variables. Some data is relevant mainly to the parameters of a transmission system, namely contact rates between different classes of individuals and transmission probabilities during contacts. These parameters reflect how individuals in different rows are connected to each other. The data for a systems analysis should be organized to reflect these relationships.
Data on the social settings in which individuals form sexual partnerships provide a case in point. These data have their greatest value when used to estimate contact rates for a transmission system analysis. When treated as an independent variable in a traditional analysis, a social setting might be have a negative association with infection at one point in an epidemic and a positive association at another (8). That is because the number of infected individuals in a particular social setting can vary dramatically over the course of an epidemic. Sex in social settings where oral sex is predominantly practiced might thus appear to be safe in a traditional analysis conducted early in the HIV epidemic (8). A transmission system analysis, however, might reveal that sex in such settings could later on be crucial in sustaining chains of transmission in a population. The traditional analysis misdirects us away from a crucial control issue because the parameters it estimates, odds ratios for instance, do not reflect the structure and function of a system. They merely reflect transient relationships in the data.
For this new era of systems analysis to emerge, epidemiologists must acquire skills not taught currently in their training programs. These include: 1) facility in abstracting the essence of systems into models, 2) an approach to developing hypotheses about the dynamic systems that generate patterns of disease in populations, 3) an ability to explore the potential behavior of dynamic systems through model analysis or simulation, and 4) a capacity to use model analyses and simulations in the design field studies. At many Universities, epidemiology students might learn these skills in disciplines other than epidemiology. It would be useful to learn them, however, in the context of epidemiological problems.
To this end epidemiologists should collaborate with bio-mathematicians, computer modelers, and systems engineers. This collaboration should encompass both scientific investigation and the development of new courses for epidemiologists in training. Recent technological advances should facilitate the roles of epidemiologists in these collaborations. Programming tools for computer modeling of dynamic and probabilistic systems have been developed to the point where such modeling can be quite feasibly undertaken by almost all epidemiologists. Deterministic models of continuous population fractions can be constructed using a variety of programs like Stella¨ (High Performance Systems, Lyme, NH). Probabilistic models of populations of discrete individuals can be constructed using simulation packages like those under development at the National Micropopulation Simulation Resource at Minnesota (http://www.nmsr.labmed.umn.edu/nmsr/NMSR.html). Instruction in the use of such tools should be an integral part of all doctoral training in epidemiology.
1 Susser MS, Choosing a Future for Epidemiology: Part I: Eras and Paradigms. Amer J Pub Hlth 1996: this issue.
2 Susser MS, Choosing a Future for Epidemiology: Part II: From Black Box to Chinese Boxes and Eco-Epidemiology. Amer J Pub Hlth 1996: this issue.
3 Holland JH. Hidden Order . Helix Books: Addison-Wesley 1995
4 Nesse RM, Williams GC. Why We Get Sick.. Times Books: Random House 1994
5 Krieger N. Epidemiology and the Web of Causation: Has anyone seen the spider?
6 Wing S. Limits of Epidemiology. In Medicine and Global Survival. Wesley RC. Sidel VW. (Eds) 1994;1:74-86
7 Anderson RM, May RM. Infectious Diseases of Humans: Dynamics and Control. Oxford University Press, Oxford, England, 1992
8 Koopman JS, Longini IM, Jacquez JA, Simon CP, Ostrow D, Martin W, Woodcock D. Assessing risk factors for transmission of infection. Am J Epidemiol. 1991;133:1199-209
9 Koopman JS, Longini IM. The Ecological Effects of Individual Exposures and Nonlinear Disease Dynamics in Populations. Amer J Public Health 1994;84:836-42.
10 Jacquez JA, Simon CP, Koopman JS, Sattenspiel L, Perry T. Modeling and Analyzing HIV Transmission: The Effect of Contact Patterns. Mathematical Biosciences, 92: 119-199 (1989).
11 Koopman JS, Simon CP, Jacquez JA, Joseph J, Sattenspiel L, Park T. Sexual Partner Selectiveness Effects on Homosexual HIV Transmission Dynamics. Journal of AIDS 1988; 1:486-504.
12 Koopman JS, Jacquez JA, Simon CP, Foxman B, Pollock S, Barth-Jones D, Adams A, Welch G, Lange K. The Role of Primary HIV Infection in the Spread of HIV through Populations. JAIDS, under review
13 Koopman JS, Simon CP, Jacquez JA. Assessing contagiousness effects of vaccines and risk factors for transmission. In Kaplan E, Brandeau M, eds. Modeling the AIDS epidemic. 1994
14 Koopman JS, Little RJ. Assessing HIV Vaccine effects. Amer J Epidemiol 1995;142:
15 Jacquez JA, Koopman JS, Simon CP, Longini IM. The role of Primary Infection in Epidemics of HIV Infection in Gay Cohorts. J of Acquired Immune Deficiency Syndromes 1994;7:1169-84 Send e-mail to me