Tobacco Documents Online

EPIDEMIOLOGY Epidemiology has its origins in the idea that genetic, environmental and lifestyle factors can influence the occurrence of disease. Unlike the laboratory experiments that one generally thinks of as forming the basis of science, epidemiologic studies involve statistical surveys. Through interviews or self-administered questionnaires, epidemiologists try to identify groups of people who match one another in all respects except for the substance or activity being studied. Having identified two presumably matched groups of people, the epidemiologist compares the disease rate between the two groups. Differences in disease rates may or may not permit the epidemiologist to conclude that there is an • association between the exposure being studied and the disease. Whether that conclusion can be drawn depends on the many factors discussed below. What Epidemiology Can Accomplish Through the years, epidemiology has been most valuable in investigating the possible causes of various infectious diseases. In the 19th century, for example, people who contracted rabies were found to have one exposure in common -- being bitten by an animal, most often a dog. Researchers therefore deduced that rabies might be transmitted through animal saliva. Laboratory experiments later confirmed that epidemiologic hypothesis. What Epidemiology Cannot Accomplish In recent years, epidemiology has become increasingly popular -- almost faddish. Hardly a day goes by without a news report claiming that one factor or another has been associated with disease, from the individual foods people eat to an almost endless list of other lifestyle factors. As a result, people have been barraged with health-related advice encouraging them to change many aspects of their daily lives. In reality, the most common causes of death are chronic, degenerative diseases -- such as heart disease and cancer -- that develop over a period of many years and appear to have a wide variety of causes. These disease can be studied only with the greatest difficulty using epidemiologic methods because of, among other things, problems in: • Matching populations groups in all respects except for the exposure in question. • Selecting study subjects that are reasonably representative of some larger population. Epidemiology March, 96 Page 1

To the first point, because they involve human subjects, epidemiologic studies do not and cannot involve the kinds of controlled conditions that tend to characterise laboratory experiments. Humans differ in many ways, both known and unknown. This introduces a potentially large source of uncertainty concerning what factors are actually being compared between study groups. For example, for heart disease alone, more than 250 factors have been identified that may put people at risk. All of these factors would have to be considered in comparing two groups. Some of the factors or characteristics that need to be matched can be verified with relative ease (e.g., age, sex, marital status, etc.). But others (e.g., diet, stress, the amount and location of ETS exposure) are difficult to verify because they depend upon people's memory of events occurring over a rather long period (30 to 40 years is not unusual). • To the second point, epidemiologists generally agree that one cannot apply the results of any study to a larger population not represented by the study population. For example, a study focusing on cancer risks that used only male subjects might not accurately represent the risk in the total population. Risk Ratios The results of epidemiologic studies are commonly expressed as "relative risks" or RRs. A relative risk is the ratio of the disease rate in the exposed group to the disease rate in the unexposed group. • A relative risk of 1.0 indicates that the disease rate in the exposed group is the same as in the unexposed group -- that is, there is neither an increase nor a decrease in risk. • A relative risk above 1.0, if statistically significant, indicates that the disease rate is greater in the exposed group than in the unexposed group. • And a relative risk below 1.0, again if statistically significant, indicates that the disease rate is smaller in the exposed group than in the unexposed group. Weak Associations Epidemiologists have long recognised that relative risks below 3.0 (and certainly below 2.0) imply a "weak association" between the disease being studied and the agent in question. In general, the weaker the association the more likely the results can be explained by "chance" or by any of a number of methodological flaws, including "bias" and uncontrolled "confounding." Epideiniology March, 96 Fage 2

Gary Taubes elaborated on this point in a recent article in Science magazine: "[M]ost epidemiologists interviewed by Science said they would not take seriously a single study reporting a new potential cause of cancer unless it reported that exposure to the agent in question increased a person's risk by at least a factor of 3 -- ivhich is to say it carries a risk ratio of 3. Even then, they say, skepticism is in order unless the study was very large and extremely well done and biological data support the link. " Computing Relative Risk Assume that the one group of people (the "test" population) has been exposed to diesel exhaust and a second group (the "control" population) has not been exposed to diesel exhaust. Also assume that in the test population one observes a lung cancer incidence of 50 cases per 100,000 subjects, and in the control • population one observes 10 cases of lung cancer per 100,000 subjects. In this example, the relative risk or risk ratio is 50/10, or 5.0. In other words, the population exposed to diesel exhaust is 5 times as likely to have lung cancer as the unexposed population. Bias Epidemiological studies are always subject to bias. Bias refers to any factor (in the design, collection, analysis, interpretation or publication of statistical data) that distorts the true nature of the relationship being studied. The most common forms of bias include: • • Publication bias -- the tendency to publish only studies that report an increased risk. • Selection bias -- including or excluding certain individuals from a study, either intentionally or otherwise. • Interviewer bias -- errors stemming from the questions that are asked or the way they are asked. • Interviewee bias or recall bias -- errors resulting from faulty memory on the part of person being questioned. • Misclassification bias -- misclassifying people in terms of their exposure (e.g. classifying current smokers or former smokers as never smokers). Epidemiology March, 96 Page 3

Statistical Significance Even after investigators have attempted to take bias into account, study samples may be unrepresentative just by chance. Epidemiologists routinely measure the impact of chance by testing their results for "statistical significance." Achieving statistical significance at the 95% confidence level (the standard used in epidemiology) means that there is a 95% probability that the result did not occur by chance alone. It is important to bear in mind that a result may be statistically significant, either positively or negatively, and still be quite meaningless. Tests of statistical significance are designed to control only for random errors associated with sample size. They cannot make a good study out of a poor study or compensate for problems unrelated to sample size. 0 Confounding Factors In a laboratory experiment, the investigator has a lot of control in matching animal populations so that the only factor that differs is exposure to the agent being studied. Thus, if there is a difference in health outcome between the groups being compared, the investigator often can say with a good deal of confidence that the difference has been caused by the agent being studied. Working with human populations is not so simple. No two people are precisely alike (in terms of genetics, diet, exercise and a host of other factors), much less two populations of people. If such differences affect the incidence of the disease being studied, they may operate as a "confounding factor." A confounding factor (or confounder) is a factor that is associated with both the disease and the exposure being studied. Consequently, they can be responsible in whole or in part for any associations that are observed. • Researchers therefore must account and control for potential confounders in their investigations. They do so by asking subjects about potential confounders and then making sure that the groups are matched in those respects. For example, a high-fat diet has been associated with an increase in lung cancer, and smokers tend to eat higher-fat diets than nonsmokers. So fat intake, among other potential confounders, would have to be controlled in any study of smoking and lung cancer, to be sure that the results were not due to differences in fat intake. Cause and Effect As noted, it is very difficult to control for every way in which humans differ and to obtain accurate and precise information about human populations. As a result, epidemiology can reveal associations without being able to prove that the agent actually caused the effect. Epiciemiology March, 96 Page 4

Any association identified through epidemiology must be validated as a cause in some other way. Ideally, this is done through a convergence of data that includes carefully controlled laboratory experiments. 0 Plausibility In addition to testing for statistical significance and ensuring that the results are not due to obvious methodological flaws, it is important to consider the question of plausibility. For example, even though a strong correlation could be demonstrated between the decline of the stork population in Europe and declining birth rates, it is implausible that the former causes the latter. Dose-Response In assessing causality, scientists look for a dose-response relationship -- that is, evidence that the rate of disease increases as exposure to the agent increases. The absence of a strong and consistent dose-response relationship tends to indicate that the reported association is not validly attributable to the agent being investigated. At the same time, the presence of a dose-response relationship -- even if strong and consistent -- is not alone sufficient to prove cause and effect. Among other possible explanations, a confounding factor could be responsible for the dose- response relationship. Consistency Finding the same association in several different studies provides some • assurance that the association is not an artifact based on the way one study was carried out or on the special characteristics of an unusual group of study subjects. In this sense, consistency across studies is reassuring -- but it does not prove that the reported association is one of cause and effect. Conversely, inconsistency among studies needs to be explained before one even begins to consider the question of cause and effect. Typically, inconsistent results indicate the absence of a cause and effect relationship. Epidemiology and ETS Epidemiology can be a valuable tool in assessing disease patterns in human populations. Just as clearly, however, epidemiologic methods can rather easily be misinterpreted or misused -- to exaggerate the perception of risks that do exist and/or to create perceptions of risk where none exist. Epid.emiology March, 96 Page 5

A number of scientists have noted that the epidemiologic studies of ETS suffer from most of the weaknesses that can occur in studies of this type. Such weaknesses include the following: • Matching -- ETS researchers have done an extraordinarily poor job of matching the test and control populations regarding factors that might influence the outcome. • Representativeness -- Minimal effort has been made to ensure that the test populations are representative of any larger group of people. In fact, the populations in almost all of the studies are clearly unrepresentative. • Statistical Significance -- Most of the heart disease and lung cancer studies failed to achieve statistical significance. • Weak Associations -- The few studies that did report a statistically significant result generally reported only "weak" associations -- levels far • below those that would be considered meaningful by most epidemiologists. • Bias -- Neither individual ETS studies nor the reviews of such studies have adequately accounted for the many forms of bias that have been shown to exist in studies focusing on chronic degenerative diseases. • Confounders -- No ETS study has even attempted to account adequately for the many factors that could confound or provide a non-ETS explanation for the results being reported. • Plausibility -- Many scientists have pointed out that the claims most consistently made about ETS lack biological plausibility because of the low levels of ETS to which people are exposed. • Dose-Response -- The ETS studies show no clear dose-response. • Consistency -- If anything, the lung cancer and heart disease studies • consistently fail to achieve statistical significance. In addition, there is marked inconsistency in the results of subgroups within the few studies that did achieve statistical significance. ### Epidemiology March, 96 Page 6