Litigation in matters of toxic exposures usually hinges on proof of a causal link between exposure and illness. We have seen a few extreme (and unsustainable) positions taken by both plaintiffs and defendants in litigation, as well as a larger number of cases where degree of causation or level of proof is legitimately debatable. This article is the second in a series outlining approaches useful in analyzing the question. The series began with an introduction to toxicology, proceeds in this issue with an introduction to epidemiology, and will continue with a discussion of exposure and risk assessment.
Epidemiology is the study of health among groups of people. Although epidemiology uses much of the same core knowledge used by physicians to treat individuals, its field of view is much broader and calls for use of quantitative skills not generally taught to or used by clinicians. Environmental epidemiology includes: studies to determine whether unusually high levels of morbidity (illness) or mortality (death) are present in a population; exposure assessments; and analytic studies to find evidence of causal relationships between environmental exposures to hazards and morbidity or mortality.
Determining exposure or performing accurate exposure assessment presents the most common major difficulty in performing environmental epidemiologic studies. An environmental epidemiologist usually tries to determine the association (if any) between a known exposure to something in the environment and an effect on health of the group exposed. Because we cannot deliberately expose humans to potential harm under carefully controlled conditions, we rarely have good measures of the amount of material to which they were exposed. Usually, we use one or more surrogate measures of exposure to represent exposure when direct or indirect measures of exposure are not possible, available, or feasible.
Poor public acceptance of study results creates another common difficulty in performing environmental epidemiologic studies. Typically, observational environmental epidemiologic studies are performed in situations in which the public is angry at the government and the investigators, and in which groups of citizens may be in an adversarial situation of having filed a law suit or intending to file one.
Several different types of study (study designs) help us solve different problems: each study design has advantages and disadvantages that determine the usefulness of the design in investigating a particular problem. The different types of study design allow a systematic, sequential approach to investigating environmental health problems. Study designs fall into two main categories: descriptive and analytical. The following study designs are most commonly used and have proven most useful in environmental epidemiology:
Descriptive studies describe the spatial, temporal, and demographic distribution of cases of an adverse health effect (that is, disease, injury, or cause of death) to develop hypotheses about adverse health effect causation that may be tested with analytic epidemiologic studies.
Unlike most epidemiologic studies, ecologic studies employ information on groups, not on individuals. Conclusions should be evaluated carefully as they can be misleading.
Pilot health effects studies and symptom-and-disease-prevalence studies are often used to address public concerns without the great resource expenditure required for more rigorous studies. Follow-up studies may direct limited resources to those situations where preliminary findings suggest that real problems exist.
Pilot health effects studies are limited, preliminary evaluations to determine whether sufficient evidence of hazardous exposures or adverse health effects is present in a population to warrant a more rigorous study.
Symptom-and-disease-prevalence studies measure self-reported adverse health effects and medical diagnoses. These reports may be validated through available medical records. Apparently high levels of an adverse health effect may be studied in more rigorous fashion.
A cluster investigation, usually responds to what appears to be an unusually large number of cases of a disease, or class of related diseases, in a confined geographic area or time. The investigation determines whether a statistically significant increase in the incidence of an adverse health effect has occurred in a particular population during a specified time period. Surveys of over 800 cluster investigations performed by federal environmental and occupational health agencies over two decades revealed only five studies that yielded statistically significant results.
Steps in a cluster investigation include:
1. Defining the extent of the cluster in terms of location, time period, and number and demographic characteristics of the cases.
2. Establishing a case definition of the adverse health effect in question, finding individual cases, and determining which cases meet that definition.
3. Adverse health effect rates for the exposed and referent (unexposed) population are compared to determine whether the difference between them is statistically significant (unlikely to occur by chance alone.) The population of the whole nation, state, or census tract or of another county may be used as a referent population.
4. Age-, race-, and sex-specific adverse health effect rates, and risk and odds ratios may be analyzed to determine whether any identifiable risk factors or confounding factors explain part or all of observed differences in adverse health effect rates.
5. Potential exposures that might explain differences in adverse health effect rates are assessed. The biologic plausibility of each of these exposures causing the adverse health effect in question is evaluated, comparing the effects with known effects of the chemical or similar chemicals.
6. All of these evaluations are integrated and conclusions made determining whether the cluster represents an unusual occurrence of the adverse health effect not explained by chance occurrence alone or usual risk factors. Conclusions may also be drawn as to whether environmental exposures have been identified that may causally explain the unusual occurrence of the adverse health effect.
Analytic epidemiologic studies like prospective or cohort studies and retrospective or case-control studies determine whether a statistically significantly increased incidence of an adverse health effect has a statistically significant association with a particular exposure or risk factor. A number of statistical tools may be used, depending on the data, but a common theme in many is a comparison of the rate of illness in an exposed population with the rate in an unexposed population.
Prospective (cohort) studies identify an exposed and an unexposed group of people at a point in time and examine what happens to them (with respect to various diseases of interest) as time progresses. If the rates of illness are significantly different, causality is evidenced. A common example would be a study that began by identifying all people who worked at a factory during a specified time period, then watching for 30 years to see whether they die of the same conditions at about the same time as do people in the general population. Retrospective (case-control) studies identify a group of people with an illness, identify a second group of people who are like the first in age, sex, and other characteristics but who do not have the illness, then determine whether the rates of exposure to a possible cause are different between the groups.
Site-specific surveillance and registries are methods to look closely over time at populations at high-risk for environmental health problems to see what problems may occur. Site-specific surveillance is a systematic, prospective monitoring for evidence of exposure or adverse health effects in a population at increased risk of exposure to hazards. A registry is a system for collecting and maintaining, in a structured record, information on people exposed to hazards or having adverse health effects of known environmental cause.
As increasing numbers of epidemiologic studies demonstrate statistically significant associations between a statistically significantly increased incidence of disease, injury, or cause of death and exposure to a particular environmental hazard or risk factor, established criteria (like Hill's modified postulates) are used to determine whether the weight of evidence is sufficient to say that a causal connection exists. The greater the number of Hill's modified postulates met, the greater the likelihood of a causal connection. Hill's modified postulates include:
Consideration of explanations other than an underlying cause-and-effect relationship for the observed association.
Strength of the association. A greater level of statistical significance evidences a stronger association. This means that the numeric probability of observing the association by chance alone was very small, so that the observation most likely represents a true phenomenon, under which lies a true cause-and-effect relationship.
Data collection free from bias. Examples of bias range widely and include errors in instrumentation used to measure exposure and changes in the ways people remember events after intervening time and talk.
Did observed differences result from uncontrolled confounding variables? Were potential confounders accounted for in the study design and analysis? (A confounder is a factor associated with both the exposure and the effect, but at different rates. Cigarette smoking and alcohol consumption are each associated with both cirrhosis of the liver and lung cancer. In considering whether alcohol contributed to lung cancer rates, cigarette smoking was a confounding variable. Studies rarely include all relevant confounders like increased life span, smoking, increased urbanization, and decreased incidence of other life-threatening diseases.)
Consistency of results in various studies (that is, similar observations in different populations at different times in different places). This is perhaps the most important criterion.
Plausible biologic mechanism refers to whether the hypothesis makes sense in the light of what we know about how environmental factors act on the body.
Positive dose-response or exposure-risk relationship refers to whether the frequency or severity of illness increases with increasing dose or exposure.
Specific methodologic issues related to chronic endpoints include:
1. Healthy worker effect. Often workers undergo medical screening on employment and at regular intervals. Thus, workers tend to be healthy adults; and worker populations do not include the elderly and very young children, two groups generally considered to have increased susceptibility to chemical toxicant effects.
2. Misclassification of exposure status owing to lack of validity of exposure measures.
3. Relevant confounding variables that are unknown or difficult to measure. At low relative risk levels, a large study is necessary to preclude uncontrolled confounding factors from generating or contributing to the observed findings.
4. Latency period. Any cohort study should have a sufficiently long follow-up period.
Counting is the basic tool of a statistician: calculating rates and determining ratios of illness among groups with differing exposure run close behind counting in importance. Errors in epidemiological studies usually result from errors in determining who to count and how to classify them. Esoteric statistical calculations are far less important, so studies should be accessible to thoughtful "outsiders," once the usual problems with technical jargon have been surmounted. Having said that, we must backtrack a little and suggest that, even in this day of computers, it is worthwhile making sure the arithmetic reported in a study is correct. A few years ago, for example, we encountered a study in which the totals obtained by adding males and females differed from the total obtained by adding all age groups. With that as a warning, we rechecked ratios and found numerous errors, some with striking effects on the conclusions. Although the errors were particularly egregious in that case, it is not unusual to encounter basic errors even in peer-reviewed journals.
As can be seen from this introduction, epidemiologic evaluations used to determine a causal relationship between adverse health effects and toxic exposures is far from an exact science. Epidemiologists must deal with missing, erroneous, or inexact information in making inferences. They must identify possible confounding variables and must assess the probable effect of uncertainties in measures of both exposure and effect before stating conclusions. P-values and 95% confidence intervals are not end points of themselves, but rather are merely information to be considered in the broader context of the question at hand. Although scientists would like to guard effectively against both: I. acceptance of a false hypothesis and II. rejection of a true hypothesis, the statistical tools deal most effectively with type I errors. In this setting, disagreements between observers readily occur. There is no substitute for thought and experience in evaluating these questions.
* Dr. Hutchinson, is on the faculty of Emory Univ. and a consulting firm CEO, while Dr. Leffingwell is a consultant to govt, industry & insurers and serves as a member of two Nat'l Research Council Committees. Both specialize in toxicology, epidemiology, public health, preventive and occupational & environmental medicine.