Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
CONCEPTS AND STRATEGIES IN PLANNING EPIDEMIOLOGIC STUDIES ON AIR 127 POLLUTION 4 CONCEPTS AND STRATEGIES IN PLANNING EPIDEMIOLOGIC STUDIES ON AIR POLLUTION INTRODUCTION This chapter addresses the principles that underlie the epidemiologic approach to air pollution research. Because ambient air pollution is generally less severe than in previous years, it is harder than it once was to detect most adverse health effects of pollution with a high degree of certainty. Research strategies that used to be demonstrably successful need further development and refinement if they are to be equally successful in solving the major research problems that remain. Initial success in protecting the public from the effects of air pollution has led to a situation in which the research questions must be more precise and the strategies chosen to address them more focused. The need for precision makes it necessary to seek out appropriately sensitive research tools and methods that will be capable of linking today's exposure with today's and tomorrow's health effects. This discussion of research strategies for the epidemiologic study of air pollution is guided by several principles. First, the development of research strategies and the specification of research questions must precede the creation of specific study protocols. Second, epidemiology should be part of a larger framework in which epidemiologic, toxicologic, and clinical studies inform, complement, and reinforce each other. Third, epidemiologic studies of air pollution should be highly sensitive to small risks of disease, because large numbers of people are exposed; to provide credible estimates of those risks, they will require careful attention to all potential sources of error at every
CONCEPTS AND STRATEGIES IN PLANNING EPIDEMIOLOGIC STUDIES ON AIR 128 POLLUTION phase of their design and conduct. Fourth, the multifaceted nature of most air pollution problems requires that planning in epidemiology be interdisciplinary, involving collaboration between different scientific disciplines and between research sponsors and investigators. Fifth, the link between epidemiology and regulation will be a two-way street, with research planning illuminated by a clear understanding of the practical issues and constraints involved in regulation, and the regulatory process influenced by the product and capabilities of research. COMMUNICATION OF EPIDEMIOLOGY WITH OTHER RESEARCH DISCIPLINES Optimal use of the epidemiologic approach requires that it communicate with the parallel disciplines of clinical research and animal toxicology. This is distinct from the development of multidisciplinary research teams. Communication between scientists in the separate disciplines is essential and should not be limited to scientists who are perceived to be working on the same research questions. This communication can always benefit from administrative sponsorship and encouragement, including attention to such mundane matters as attendance at conferences, the locations of offices and laboratories, and participation in strategic planning of research and development of individual research projects. Interdisciplinary contact will improve any epidemiologic strategy in air pollution in at least three ways. First, results of experiments in the laboratory and clinic will generate useful hypotheses for epidemiologic studies. Examples are the observation of cellular, whole-animal, or human adaptational responses to irritants as a basis for studying the responses of various groups to short-term pollution episodes and the observation of a synergistic effect of two pollutants as a basis for selecting communities potentially at high risk. Second, clinical research and toxicology are needed to provide better epidemiologic research tools. The development and validation of short-term biologic markers that reflect long-term respiratory damage constitute a good example. Laboratory investigators concentrating on disease mechanisms or therapy are often unaware of the potential applications of their work in population studies. For
CONCEPTS AND STRATEGIES IN PLANNING EPIDEMIOLOGIC STUDIES ON AIR 129 POLLUTION instance, a diagnostic test that is considered to be superfluous or too inaccurate for clinical work might be highly appropriate for testing large populations, because of its low cost and noninvasive nature. Research and development activity that leads to new research tools for epidemiology should be distinguished from empirical research on disease mechanisms and perhaps require separate administration and funding. Third, as discussed below, evidence from clinical research and toxicology is used by epidemiologists to strengthen the inference of causality in observed relationships between air pollution and health effects. Epidemiologic activity also contributes to clinical research and toxicology. It can be an excellent source of hypotheses for laboratory studies and, more important, provide an outlook on disease that is critical for setting priorities. Apart from its ability to produce data and answer specific research questions, epidemiology offers a perspective and a set of concepts that can be used to guide efforts in disease prevention and research design. CONSTRUCTING APPROPRIATE RESEARCH QUESTIONS âThe epidemiological method is the only way of asking some questions in medicine, one way of asking others, and no way at all to ask many.â16 Research planners can reduce the chances of investing in inappropriate or inconclusive studies by clearly identifying research questions in advance. We refer here to a research question as a broad statement of a problem, rather than a specific epidemiologic hypothesis for an individual study. Major air pollution research questions seldom emerge ab initio from the intellects of creative scientists. Most are public health questions that are developed through interplay of the interests and capabilities of the scientific community with extrascientific forces, such as the availability of research funds and the regulatory process. Once a question has been articulated and is determined to be important from the standpoint of public health, research planners have the goal and responsibility of finding the most productive investigative strategy or determining that epidemiologic studies are not feasible. In practice, an iterative process is used both to develop and refine the formulation
CONCEPTS AND STRATEGIES IN PLANNING EPIDEMIOLOGIC STUDIES ON AIR 130 POLLUTION of research questions and to match each formulation with the optimal strategy. The cycle of marshaling facts to make a question explicit, devising a strategy, and then refining the question with new facts is parallel to the cycle of hypothesis- test-hypothesis that takes place at the level of individual studies. In this type of applied research, the development and refinement of the question and of the strategy are inseparable, and the investigator contributes equally to both. The importance of specifying research questions cannot be overemphasized. As in all scientific disciplines, epidemiologic research questions begin in a rather general form and are pruned to testable hypotheses that form the basis for design of individual studies. Every epidemiologic study of the adverse health effects of air pollution has to be crafted around the pollutant in question and the adverse effect under consideration. The characteristics of the population at risk, the time between exposure and effect, and the amounts of exposure needed to produce an effect of a given magnitude are specified to some degree in the formulation of a study hypothesis. These elements of a hypothesis usually cannot be specified to the last detail; but, for a hypothesis to be useful, some specification of each element must at least be implied. Each study must conform to various investigative constraints. These constraints, which must be appreciated early in planning, are imposed by such factors as: â¢ The frequency and natural history of the disease. â¢ The rate and the extent to which exposure to the pollutant in question changes over time. â¢ The availability of biologic markers of exposure or early effect. â¢ Other known or suspected causes of the health effects in question. The overall research strategy has to be chosen with the best possible understanding of these four separate dimensions, and the detailed research protocol has to be designed to take them fully into account. The more we recognize that exposures to pollutants change and that given diseases have various causes, the more challenging
CONCEPTS AND STRATEGIES IN PLANNING EPIDEMIOLOGIC STUDIES ON AIR 131 POLLUTION it becomes to design a penetrating and efficient epidemiologic research protocol. Protocols are most penetrating and rewarding when the detailed research plan focuses on a single answerable hypothesis. The few exceptions to this important general rule include broad epidemiologic surveys designed explicitly for surveillance and the development of new hypotheses. Epidemiology can be used to produce several different kinds of information about air pollution and health. The kind of information sought depends on the goals of the researcher. The remainder of this section discusses six types of broad research questions that can be addressed epidemiologicily and briefly describes the capabilities of epidemiology to answer each type (for a more complete discussion of these issues, see, for example, Morris's Uses of Epidemiology16): 1. How does the state of health of a particular community with respect to disease Y compare with that of other communities? With itself over time? 2. What is the natural history of disease Y? Its clinical spectrum? How does it progress? 3. Is there any association between exposure X and disease Y? Is the association causal? 4. How much does the risk of disease increase as exposure increases (dose- response relation)? Is there a magnitude of exposure for which the disease risk is zero (a threshold)? 5. What are the risks of disease among various groups in the general population? 6. How much of disease Y is due to exposure X? How much will be prevented if exposure is reduced? Question 1, regarding the monitoring of the health of a population, can be addressed only with epidemiologic methods. The relation of this question to air pollution concerns is indirect, however, except for its role in identifying study populations on the basis of contrasting disease patterns. Question 2, about the pathogenesis and natural history of a disease, can be answered through clinical observation and through epidemiology; but this
CONCEPTS AND STRATEGIES IN PLANNING EPIDEMIOLOGIC STUDIES ON AIR 132 POLLUTION type of question is rarely at the center of interest in air pollution studies, because there appear to be no unique âair pollution diseases.â Moreover, although epidemiology can undoubtedly contribute to the understanding of how an exposure results in disease, detailed information of this kind is normally best provided by intensive clinical and laboratory studies. Most applications of epidemiology in air pollution research have been to answer questions about correlations and causal relationships between air pollutants and various human health effects--Questions 3, 4, and 5. To address them, the epidemiologist normally uses the measure known as relative risk, the ratio of the incidence or prevalence of an adverse effect in one population to the incidence or prevalence in another population with a different degree of exposure. Epidemiology can be very effective in answering Question 3, about the existence of any association between air pollutants and disease, particularly if the relative risk involved is large (sometimes arbitrarily defined as 3.0 or greater).14 In such a situation, it is usually most sensible to compare populations at highest exposure with those whose exposures are low or nil. If the relative risk is extremely large and the disease is relatively rare in nonexposed persons, the association will often be noticed first by clinical observation, as in the case of the relationship between vinyl chloride and angiosarcoma of the liver or between asbestos and mesothelioma. It is reasonable to presume that, in most populations without extreme exposure, relative risks of overt clinical disease associated with air pollution are of low to moderate magnitude (approximately 1.2-3.0). Detection and accurate quantification of effects of this magnitude present the greatest methodologic challenges to contemporary epidemiology. Nevertheless, because it is unethical to conduct studies involving deliberate long-term exposures or acute exposures that might produce irreversible effects, epidemiologic studies will remain the only way to detect pollution-related forms of common chronic or serious acute diseases in humans. Epidemiologic studies of air pollution have much greater limits in providing finely tuned information about dose-response relationships or about thresholds or
CONCEPTS AND STRATEGIES IN PLANNING EPIDEMIOLOGIC STUDIES ON AIR 133 POLLUTION âsafe levelsâ (Question 4), particularly when small relative risks of chronic diseases are produced by years of low-dosage exposure. In those situations, it is extremely difficult to discriminate, with the necessary degree of precision, among populations with relatively minor differences in exposure, for instance, long-term exposure to ozone at 0.08 ppm versus 0.12 ppm. In addition, if the amounts of individual pollutants are highly correlated and tend to rise and fall together from time to time and place to place, the role of each pollutant can be difficult to isolate. Larger exposures can be studied to construct models that simulate the behavior of pollutants at lower dosages, but the reliability of mathematical extrapolation is limited by its dependence on unavoidable and untestable assumptions about the dose-response relationships. Future studies will gradually become better at providing quantitative information about the relationships between air pollutants and health, if the techniques for assessing exposure and effect mentioned in this report are developed and if research planning is appropriate. Epidemiology, the basic science of preventive medicine and public health, views disease at the population level, but it can be used to predict the risks of disease in individuals or groups with various characteristics of interest (Question 5). By virtue of the mandate of the Clean Air Act (if for no other reason), it is essential to understand the range of susceptibility to air-pollution-related disease in the general population. Except in a few instances of controlled-exposure studies of volunteers with defined characteristics, such as pre-existing illness, epidemiology is necessary to identify high-risk groups. But precise determination of dose-response relations for each sensitive group is generally not feasible today. The epidemiologic approach is uniquely suited to answer Question 6, although it seems to have been asked only rarely in relation to air pollution. This type of question, which deals with attributable risk rather than relative risk, asks how much of the overall burden of disease in a population is attributable to air pollution over the entire range of exposure. The answer has obvious and important public health implications, because it indicates the total amount of disease that is potentially preventable and the amount that would not
CONCEPTS AND STRATEGIES IN PLANNING EPIDEMIOLOGIC STUDIES ON AIR 134 POLLUTION occur if air pollution were reduced. Results of research to answer this kind of question are useful in making decisions about reducing exposures across the board, regardless of local or individual variations in susceptibility or risk and regardless of the relative importance of direct effects and synergistic effects, which are more difficult to know ahead of time. Epidemiology offers a direct means for predicting and later assessing the public health impact of changes in the environment, whether positive or negative, deliberate or unintentional. In summary, Questions 1 and 2 are fundamentally epidemiologic, but are likely to be of limited interest in EPA's air pollution activities; epidemiology has made and can continue to make substantial contributions toward answering Questions 3 and 5 and, to a smaller extent, Question 4; and the epidemiologic approach has scarcely been applied to Question 6, although that application might be valuable. CONSIDERATIONS IN STUDY DESIGN, ANALYSIS, AND INTERPRETATION Epidemiologic projects concerning air pollution are among the most difficult to design. One reason is that ambient air pollution adds small increments to the respiratory morbidity risks of large masses of people, rather than large increments to the risks of a few. The dynamic and complex nature of air pollution makes it particularly difficult to measure, and its very nature surrounds it with confounding factors, such as temperature and humidity. Study design thus affords little margin--yet frequent opportunities--for error. Furthermore, epidemiologic studies, unlike laboratory experiments, are not easily repeated, so elimination of potential errors during study design is of paramount importance. Evaluation of all potential errors (such as that caused by the combination of very low exposure and low risk of effect) might lead study planners to conclude that epidemiologic studies are not feasible for providing the particular information desired. The design of each study is a multistage process that begins with the extraction of a specific hypothesis from a general research question and ends with detailed budget estimates. In between, the designers must decide what
CONCEPTS AND STRATEGIES IN PLANNING EPIDEMIOLOGIC STUDIES ON AIR 135 POLLUTION exposure and what outcome are most pertinent and which measurements of each to use. They must then identify potential confounding variables, choose a study method and study populations that accommodate the hypothesis, and plan for the eventual analysis and interpretation of the data. This by no means involves a rigid sequence; the order of the steps varies, and often a decision about one step contains or implies decisions about others. Nonetheless, each step plays a distinct role in the study's integrity and the certainty that can be attached to its results. The overall goal is to design studies so as to increase their validity in detecting and estimating the magnitude of the relationships between air pollution and health. Validity has two components: sensitivity and specificity. Sensitive studies have a low probability of labeling a harmful situation as harmless, and specific studies have a low probability of labeling a harmless situation as harmful. Ideal studies are sensitive enough to detect small but widespread effects, yet specific enough to discriminate real from spurious effects. The complexity of the relationships among all the variables that could affect results of air pollution studies has two important implications: â¢ Study design and planning require expertise regarding the concepts of epidemiologic research design. â¢ The success or failure of future studies will depend increasingly on how well exposure, effect, and confounding variables have been characterized during study planning. Study designers must have as much prior information as possible concerning the distribution of the factors responsible for nonrandom errors and confounding in the various populations being considered for study. The information can be inferred from previous studies of similar populations or obtained directly from the study populations themselves either before or during the study in question. If the information is obtained during planning, it can be used to alter the design of the study, change sample sizes, or influence the selection of study populations. Random errors (and nonrandom errors that behave as random ones in populations) can be
CONCEPTS AND STRATEGIES IN PLANNING EPIDEMIOLOGIC STUDIES ON AIR 136 POLLUTION mitigated by increasing sample size. Thus, epidemiology can sometimes take advantage of such data as those on death certificates, which individually are unreliable, but which yield reliable estimates of group characteristics when used in large numbers. In contrast, errors due to confounding or uneven biases cannot be corrected by increasing sample size, and other strategies must be used. It is not within the scope of this report to discuss all the complex design considerations involved in weighing and managing potential sources of error. Once data have been collected, errors in analysis or interpretation can contribute to the total burden of error (and hence validity) of an epidemiologic study of air pollution. A few pitfalls are particularly likely in air pollution studies, but they can be avoided if anticipated during planning. This subject is at the heart of epidemiologic training and is treated in detail in textbooks and courses on research methodology. A few relevant concepts are introduced in Appendix C to point out the particular challenges for air pollution studies. AVAILABLE METHODS FOR EPIDEMIOLOGIC STUDIES Epidemiology depends on a small number of distinct study methods, any of which can be applied to air pollution questions. DESCRIPTIVE EPIDEMIOLOGY OR âUNIVARIATEâ STUDIES The simplest form of epidemiologic study merely examines the frequency or severity of a disease or of a risk factor by personal characteristics, time, or place. This kind of study is referred to as descriptive or univariate, because it covers either exposure or effect, but not both. Descriptive epidemiology can be very helpful in identifying nonrandom variations in the occurrence of disease or exposure that can lead to study hypotheses. Data on the geographic distribution of exposure to air pollution or of putative effects can help in the planning of research strategies. Maps of nationwide emission of various pollutants exist, and similar maps of air monitoring-station results should also be made available,
CONCEPTS AND STRATEGIES IN PLANNING EPIDEMIOLOGIC STUDIES ON AIR 137 POLLUTION with recognition of the limits of the data. The National Cancer Institute has published mortality maps by standard economic areas in the United States for nonmalignant respiratory diseases.13 Data on geographic variations in respiratory morbidity, which would be even more valuable, are not available in the United States, and their development is hampered by lack of consistency in reporting or coding of morbidity events. In some instances, it might be helpful to know geographic variations in hypersusceptibility factors, such as asthma and Î±1- antiprotease deficiency. ECOLOGIC STUDIES In ecologic studies, the group, rather than the individual, is the unit of comparison. Aggregate information is used, instead of individual data, on both exposure and effect. Studies that use aggregate data on exposure and individual data on health effects can be considered âsemiecologicâ studies. Inferences based on ecologic studies tend to be much weaker, and there is a much greater loss of information than when individual data are used. Ecologic studies can involve comparison of exposure and disease rates over time in the same population (temporal studies) or comparisons of exposure and disease rates in various geographic areas at the same time (geographic studies). Ecologic studies are often attractive because of their relatively low cost, straightforward and rapid completion, and relative insensitivity to demands for data. They have some serious weaknesses, however. Confounding can be a severe problem, owing to the difficulty of obtaining comparable and reliable data for whole populations on such factors as smoking. There are also problems with respect to a wide variety of data elements, including exposure magnitude, disease rates, population characteristics, and migration in and out of the areas studied. Conclusions regarding individual risk must be drawn cautiously, because data on individual behavior that can influence risk are not collected. Recent articles provide a more complete discussion of the problems associated with this method.15 When the required data are of adequate quality, ecologic studies have a role in an air pollution research
CONCEPTS AND STRATEGIES IN PLANNING EPIDEMIOLOGIC STUDIES ON AIR 138 POLLUTION program. Comparisons of disease rates between populations help us to understand the potentially preventable fraction of total disease burden, to target high-risk populations for study, and to avoid studies in populations whose exposure or disease rate is too uniform. COHORT METHOD The cohort, or âprospective,â method can be based on records of exposure and effect that are concurrent or historical (in which case the cohorts are defined by historical information and followup time has already elapsed). This design is usually not suitable for rare diseases, because the sample size must be too large, but it can be excellent for common diseases. The exposed cohort is compared with a cohort or cohorts with smaller or no exposure, and disease or other adverse-effect rates are ascertained during a followup period. The differences are presumed due to the exposures, if confounding can be handled and the populations appear comparable in respects other than the exposure of interest. This method is useful for investigating the effects of long-term exposures and for diseases that have long latent periods. Its advantages include the possibility of looking at many different diseases in the cohorts simultaneously with little increase in cost and the ability to estimate incidence rates, as well as relative risks. Historical cohort studies, commonly used to study occupational exposures, have been used only rarely to study exposures to air pollution. They require identification of subjects on the basis of existing records, reconstruction of individual exposure histories, and tracing of subjects toward the present for key health events. Although workplaces have sometimes routinely collected data that allow reconstruction of exposure, few such data exist for community exposure to air pollutants. Techniques for reconstructing pollution exposure, perhaps based on residential histories, remain to be developed. Such techniques would also be of use in case-control studies. The accuracy of such historical estimates is likely to be relatively poor. Truly prospective longitudinal studies, based on individual measurements of exposure and effect, are likely to provide the strongest basis for epidemiologic
CONCEPTS AND STRATEGIES IN PLANNING EPIDEMIOLOGIC STUDIES ON AIR 139 POLLUTION inference on air pollution, because they give the investigator the greatest degree of control over random and nonrandom errors. But they are sometimes prohibitively expensive. Moreover, it might take years of observation for the outcome of interest to occur in a statistically significant number of subjects. Where responsibility for public health is involved, avoidable delays in collecting pertinent information about air pollution cannot be justified. Apart from ethics, short-term studies (5 years or less) have practical advantages over long-term studies, including more complete followup of subjects, better control of selection bias among subjects who remain in the study, and greater stability of research teams. Longitudinal studies often require a tradeoff between sample size and duration of followup. An investigator might achieve the same certainty about an association between an air pollutant and an effect by studying, for example, 50 people for 8 years and by studying 1,000 people for 2 years. The statistical-power relationships involved in longitudinal measurements of lung function in groups of various sizes for various periods have been analyzed recently by Berry (see Table 1, in Chapter 5 ).2 The cost and information value of these options must be explored in the planning of longitudinal studies. Shorter durations of study tend to mitigate perhaps the largest problem of prospective studies of chronic air pollution effects--loss of subjects to followup because they change residences. In the UCLA study of chronic respiratory effects in two areas of Los Angeles, 54% and 45% of the subjects from the lower- and higher-exposure areas, respectively, were lost to followup after 5 years, owing largely to residential mobility.24 The serious adverse impact of mobility on the power of epidemiologic studies of chronic disease (regardless of the method used) has been discussed elsewhere.20 Another kind of tradeoff--between precision of exposure assessment and statistical precision or power--is related directly to the design of cohort studies. Small cohorts allow more opportunities to obtain precise exposure and effect data; large cohorts, although they usually involve less precise data, provide greater statistical sensitivity and are easier to generalize to unstudied populations. The small-cohort and large-cohort approaches are therefore
CONCEPTS AND STRATEGIES IN PLANNING EPIDEMIOLOGIC STUDIES ON AIR 140 POLLUTION distinct and can be used to address different kinds of questions. Their use in strategies pertaining to current questions is illustrated in Chapter 5. Many chronic disease epidemiologists are working to develop research strategies and tools for the short-term study of the causes and modifiers of long- term conditions. Some of their efforts have focused on short-term biologic markers and the identification of susceptible groups. Valid measurement of early disease might allow studies of chronic lung disease to be completed, for example, in 3-5 years, rather than 15-20 years. Chapter 2 pointed to rapid decline in lung function as a useful physiologic indicator that permits compression of longitudinal studies. Early disease markers might also provide outcome measures that are more frequent than overt disease and thus magnify relative risk and increase the statistical power of cohort studies. The success of the Framingham and Tecumseh studies has led to widespread appreciation of the value of longitudinal cohort studies in identifying biologic and lifestyle measurements that predict disease in groups. Several epidemiologists are using data from these and similar longitudinal studies to identify the conditions under which short-term cohort studies might provide the same information as long-term ones. NESTED CASE-CONTROL METHOD A combination of the cohort and case-control approaches, known as the ânested case-controlâ or âcase-control-within-a-cohortâ study, has recently come into wider use, because it can lead to lower costs and more reliable results. In this method, a large cohort is followed, historically or prospectively, and outcomes are determined for all (e.g., whether a subject died of cancer), but some exposure data (such as results of tests on stored blood samples) are measured or coded only for a subset of subjects who develop a health effect of interest (cases) and a random subset of subjects who do not (controls). Because the statistical value of having more than about three controls per case is usually negligible, and because most chronic diseases have relatively low incidences, even in high-risk groups, savings can be large. The nested case-control method can capitalize on
CONCEPTS AND STRATEGIES IN PLANNING EPIDEMIOLOGIC STUDIES ON AIR 141 POLLUTION situations in which blood or other biologic samples have been obtained and set aside for storage as part of a baseline or periodic evaluation of a cohort.27 Therefore, this approach should permit the wider use of biologic markers in population studies when the cost of each marker assay is a barrier. CROSS-SECTIONAL STUDIES Cross-sectional studies can be considered to make up a subtype of the cohort study, in that subjects are enrolled according to extent of exposure. In cross-sectional studies, exposure and effect data are collected concurrently, and that permits more rapid completion of studies and lower total costs. The chief disadvantages are that information about past exposure is often unreliable or unavailable and migration into and out of the geographic areas in the study (particularly of especially healthy or especially unhealthy people) might bias the sample. In addition, because exposures and effects are measured concurrently, it is difficult to ensure that exposures preceded the effects. Some studies have suggested that one-time cross-sectional evaluations of adverse health effects, such as lung function deficits and respiratory symptoms, are less sensitive than longitudinal evaluations of changes over time. Cross- sectional study of adults in Berlin, New Hampshire, in 1962 showed no greater prevalence of symptoms in those residing in more polluted areas; however, a followup study in 1967 indicated that improvements in air quality had led to detectable improvements in symptom prevalence.6 Studies by Van Der Lende in the Netherlands have suggested that longitudinal measurement of lung function is a more sensitive indicator of air pollution effects than cross-sectional measurement.25 In fact, there are puzzling discrepancies in the normal relationship between age and lung function, depending on whether curves are obtained by cross-sectional or by longitudinal methods.7 The choice between cross-sectional and longitudinal approaches is affected to some degree by the difference between the fluctuation in exposures or effects from person to person and the fluctuation from time to time for the same person. When data on individuals are unstable, compared with inherent variability between persons, there might be little advantage in studying the same person repeatedly.
CONCEPTS AND STRATEGIES IN PLANNING EPIDEMIOLOGIC STUDIES ON AIR 142 POLLUTION Longitudinal (cohort and case-control) studies have substantial advantages in drawing inferences of cause and effect; but, if an investigator is sensitive to the vulnerable aspects of the cross-sectional method and plans accordingly, sound studies are possible for some research questions, particularly for short-term exposures and acute effects. The migration problem can be addressed by restricting the study to long-term community residents or at least analyzing their results separately. This solution is not perfect, because the long-term residents might differ from others in responses to air pollutants. Furthermore, potential errors in assessing long-term past exposure seriously limit the ability of the cross-sectional approach to provide precise information on dose-response relations for chronic effects. However, even for chronic effects, the cross- sectional method can be used to infer magnitude of risk when cause-effect relations are already known, to develop important new hypotheses for further study by other means, and to check and confirm, in a range of settings, a relationship found in a longitudinal study. CASE-CONTROL METHOD In the case-control method, one starts by defining persons who do (cases) or do not (controls) have the outcome (usually a disease) of interest and looks into their history for evidence of exposures. The problems lie in defining the disease, measuring past exposures reliably, selecting the control group, and handling confounders in the analysis. The prevalence of exposure to a suspected agent is the important determinant of sample size and hence study sensitivity, but rarity of the disease is usually not a hindrance in this approach if enough cases can be found, even if from multiple sources. In fact, a case-control study is sometimes the only feasible kind for investigating rare diseases. It permits a variety of exposures to be examined at relatively low cost--a clear advantage over cohort methods. In contrast with cohort approaches, one cannot ordinarily study more than one disease at a time without substantial increases in total sample size to ensure that each outcome is adequately represented. Case-control studies on the effects of ambient air pollution have been infrequent, perhaps because of the
CONCEPTS AND STRATEGIES IN PLANNING EPIDEMIOLOGIC STUDIES ON AIR 143 POLLUTION difficulty of reconstructing past exposures with acceptable precision. Also, cases and controls can differ in their recall of types and magnitudes of air pollution; data can sometimes be strengthened by direct validation of past exposure, and biologic markers of exposure might become valuable in this regard. In contrast, effect or outcome data can be scrupulously defined, with hospital or pathology records if necessary. Because the chief instrument in a case-control study is a questionnaire, it is comparatively easy to obtain information about the wide range of potential confounders that plague air pollution research. Important summary measures, such as attributable risk, can be estimated with this method. Case-control studies have more to offer air pollution research than has been recognized. But uncertainties involved in retrospective exposure assessment still limit their use when precise information on dose-response relations or thresholds is needed. INTERVENTION STUDIES In intervention (experimental) studies, an investigator adds to or reduces exposures of a cohort and then follows the cohort, assessing the impact of the intervention on disease rates in comparison with rates in a control population. When a pool of suitable subjects is randomly divided between exposure and control groups, results are solidly based. In medical intervention, this approach (the randomized clinical trial) is widely recognized as the âgold standardâ for strength of inference and proof of causality. Obviously, air pollution exposures cannot be deliberately imposed on a population, nor is the study of populations placed by chance at potential risk without ethical complications. The study of populations with reduced exposures is the most acceptable alternative. This strategy was used by investigators in England after the Clean Air Act in 1956 caused a great reduction in air pollution in most urban areas. The investigators tracked disease rates and compared them over time with the changing degree of air pollution and found that areas with greater reductions in smoke pollution experienced greater declines in lung cancer death rates.14
CONCEPTS AND STRATEGIES IN PLANNING EPIDEMIOLOGIC STUDIES ON AIR 144 POLLUTION SPECIAL OPPORTUNITIES FOR AIR POLLUTION STUDIES Occasional situations provide opportunities to expand and deepen the inferential base on a particular research question, by virtue of their peculiarities. These opportunities, which are somewhat indirect approaches to the question, can also be unusually cost-effective. The direct, deliberate design and organization of traditional epidemiologic studies are only part of what is required for an effective air pollution research strategy. The other part is the recognition and use of these special, and perhaps fleeting, situations for study. Several types of special opportunities for research in air pollution are discussed in the following pages. Some pertain to rapid response or surveillance, some are settings for the development of protocols, and some are ways of asking new questions by building on current and continuing surveys or studies. Many of these are valuable because they allow epidemiologists to deal with major methodologic constraints. To some extent, the pursuit of information about the association of air pollution and illness in the United States is limited by the number of situations in which observational epidemiology can be conducted. Five important types of situations can be identified: extrapolation from unusually high or low exposures, dramatic increases or decreases in air pollution, the discovery of populations that avoid confounding variables, separation of ordinarily correlated exposures, and exposure of populations to pollutants of concern for the future. OUTBREAK DETECTION AND RESPONSE Daily or weekly changes in air pollution constitute opportunities to learn more about acute effects. Although the most sensitive studies rely on the proper placement of specially designed air monitoring and health evaluation systems, much can be learned by surveillance over wide areas with relatively crude exposure and effect data. Once an outbreak of pollution-related disease has been detected, more detailed studies can be set in motion. However, as mentioned in Chapter 2, better descriptive data on respiratory morbidity will be needed. Coupled with data from the national air monitoring systems, these data could form a surveillance network capable of
CONCEPTS AND STRATEGIES IN PLANNING EPIDEMIOLOGIC STUDIES ON AIR 145 POLLUTION detecting much smaller events than the fog disasters of previous decades. Relatively crude outcome data are being collected in several systems. The Centers for Disease Control collects and compiles year-round information on total excess mortality and pneumonia and influenza deaths for 132 cities. Information on daily mortality, hospital censuses, and school and work absences is readily available in many areas. Although the limitations of these data are obvious, time-series analyses of specific populations can provide useful exploratory information. OCCUPATIONAL STUDIES Occupational cohorts are relatively easy to assemble for studies, and exposure measurement, at times based on personal monitoring, is often available. Although it is hard to find occupational groups exposed only to the major air pollutants, each one found is likely to have major value. For carbon monoxide, radon, and lead, occupational studies have formed a large part of the risk assessment model for community-wide exposure. Use of occupational studies for air pollution research is further complicated by the greater intensity and different temporal patterns of the exposure involved and by the âhealthy-workerâ characteristics of the population at risk. But cohorts appropriate for studies can be identified. Examples include flight crews and attendants exposed to ozone, corn refinery plant workers exposed to sulfur dioxide,8 and professional cooks exposed to gas-stove emission, including nitrogen dioxide. These occupational panels provide few opportunities to detect thresholds, but experiences can be particularly useful in constructing other parts of the dose- response curve and in validating new techniques for exposure or effect assessment, including biologic markers. Kilburn and co-workers, after a study designed to examine the effects of para-occupational exposure to asbestos on wives and daughters of shipyard workers, reported the serendipitous finding of an effect on pulmonary function that might have been due to ambient air pollution.11
CONCEPTS AND STRATEGIES IN PLANNING EPIDEMIOLOGIC STUDIES ON AIR 146 POLLUTION MIGRANT STUDIES Studies of migrants have been useful in disentangling the effects of environment and heredity on disease incidence. In the field of air pollution, epidemiologists have used such studies to establish what is known as a âBritish effect,â referring to residual excess risk of lung cancer and chronic bronchitis in British-born migrants, compared with native populations in the countries to which they migrated.22 Thus, hereditary factors aside, migrants are interesting subjects for study because of the sharp changes in exposure that they might have experienced after emigration. Clues as to the impact of early exposure on the ultimate risk of chronic disease might also be revealed in these groups. Residential mobility within the United States, which complicates local studies, can be turned to advantage by identifying groups that have moved between areas with sharply constrasting pollution conditions. Examples include the migration of rural people to cities, such as Los Angeles, and the movement of city residents to retirement communities in the Southwest. By examining the health changes that accompany these moves, we can in effect ask a series of âwhat ifâ questions regarding the potential impact of changes in air quality. Questions concerning the influence of past exposure on current response (adaptation, in toxicologic terms) might also be addressed with this approach. PRISTINE POPULATIONS We accept some air pollution as an inevitable accompaniment of contemporary industrial society, but we have only scant knowledge about the general health status that we might observe if we had little or none at all. In particular, the shapes of growth and decay curves for lung function in the absence of substantial air pollution are not known. Research in Busselton, Australia, is attempting to describe lung growth and decay curves measured longitudinally in a large, nonexposed population of European descent (personal communication, A.C. Woolcock, University of Sidney, New South Wales, Australia). Until this kind of information is available, it will be hard to rule out a subtle but pervasive effect of air pollution on respiratory function. The range of
CONCEPTS AND STRATEGIES IN PLANNING EPIDEMIOLOGIC STUDIES ON AIR 147 POLLUTION function considered normal might in fact be abnormal compared with that in pristine populations. For example, normal serum cholesterol differs between the United States and Japan, and normal blood lead content differs between a Himalayan population and the general U.S. population.19 As Rose has recently stated, the most difficult risk factor to detect epidemiologically is one that is present in every member of the population.23 POINT-SOURCE POLLUTION In communities near single point sources of pollution, collection of precise data on emission and exposure modeling of the population are often feasible. Moreover, the population at risk is often well-defined and might be contained within administrative units conducive to the organization of an epidemiologic study. Of primary interest, however, is the possibility that communities near point sources have exposure that is relatively uncomplicated by the presence of multiple pollutants. Such conditions can be found in small towns with little industry or automobile traffic. Towns near paper mills (sulfur dioxide), chemical plants (various volatile hydrocarbons), and smelters (particles, sulfur oxides, and heavy metals) are some examples. The selection of a small community for study could constrain sample size and, hence, statistical power; such problems should be considered during study planning. STUDIES OF POPULATIONS IN FOREIGN COUNTRIES Some overseas populations have had long-term exposure to air pollution at magnitudes no longer seen in the United States. Many of these populations are in semi-industrialized cities with crowding and substantial automobile traffic. Because of the exaggerated contrasts in pollution, these cities might provide ideal sites for studies involving questions of the âany associationâ type. Downward extrapolation along dose-response curves could also prove feasible. Figure 6 shows results in selected cities from a global monitoring survey of annual average sulfur dioxide concentrations. Some cities in other countries have experienced rapid industrialization and urbanization with concurrent swift
CONCEPTS AND STRATEGIES IN PLANNING EPIDEMIOLOGIC STUDIES ON AIR 148 POLLUTION FIGURE 6 SO2 in cities of the GEMS monitoring network, 1976-1980. Range of annual averages at individual sites and composite average for each city. Reprinted with permission from Bennett.1
CONCEPTS AND STRATEGIES IN PLANNING EPIDEMIOLOGIC STUDIES ON AIR 149 POLLUTION increases in pollution. In Sarajevo, Yugoslavia, for example, atmospheric emission of nitrogen oxides increased by 277% between 1970 and 1980.3 Health in areas with a longer history of modernization might have deteriorated so gradually as to defy detection. Dramatic decreases in exposure to a pollutant or mixture can also occur in cities undergoing rapid conversion from one fuel source to another. Elimination of a confounding variable, such as smoking or occcupation, through selection of a study population is sometimes more effective than using matching, stratification, or multivariate adjustment techniques in the study design or analysis. For example, one could minimize the impact of indoor versus outdoor exposure by studying groups that live in tropical or subtropical climates, where building techniques allow ample ventilation and people spend greater proportions of their time outside. Situations outside the United States should be sought in which it is possible to separate the effects of exposures that are normally coexistent. For example, sulfate concentration might be high in the presence of low concentrations of particles or ozone, or nitrogen oxide concentration might be high in the presence of low concentrations of other photochemical oxidants. It is important to maximize the benefit derived from expensive longitudinal studies and to disseminate their methods and results as widely as possible. The investment of U.S. effort and funds in studies already going on in other countries could be more prudent, in some instances, than starting similar studies here. For example, U.S. expertise in monitoring and health effects technology can be used to complement and strengthen some studies. Longitudinal studies of note are being conducted in Canada (Hamilton),18 the Netherlands,25 and Brazil (personal communication, H. P. Ribiero, Department of Surgery, School of Medical Sciences of Santa Casa, Sao Paulo, Brazil). The existence of centralized and universal health insurance coverage in other countries (for example, Canada and Sweden) might provide data on which powerful and cost-effective studies could be based. Many nations have responded to the 1973 world oil crisis by reviving the use of traditional fuels and
CONCEPTS AND STRATEGIES IN PLANNING EPIDEMIOLOGIC STUDIES ON AIR 150 POLLUTION exploring alternative technologies. Several fuels--such as coal, wood, and gasohol--are being considered as alternatives to oil in the United States, but there is little information on their potential adverse health effects. Learning from the evolving experience in other countries might prevent costly and unhealthful mistakes or inspire innovative technologies. For example, 90% of the cars manufactured in Brazil are designed to run on fuel that contains a large proportion of ethanol; this will result in a massive shift in the pattern of air pollution in some Brazilian cities, and it is already raising questions about the potential health effects of combustion products, such as aldehydes (personal communication, R. Mendes, Pan American Health Organization). ADAPTATION OF EXISTING COHORT STUDIES The remaining special opportunities concern the design of highly powerful and cost-effective epidemiologic studies through adaptation or âpiggy-backingâ of air pollution hypotheses onto longitudinal studies or surveys. In some instances, cohorts established for other purposes may be adapted to address air pollution questions. The wealth of data from the Framingham and Tecumseh studies was mentioned earlier. Other longitudinal studies of respiratory health have been conducted in Tucson,12 Baltimore,4 and Boston.26 Statistical models designed to predict the risk of COPD have been applied to each cohort.10 These models confirm that age, sex, smoking habits, and initial FEV1 are the most important factors in predicting development of COPD within 10 years of initial evaluation. Air pollution exposure has not yet been thoroughly tested in all the cohorts. It is unlikely that air pollution exposure will match the above variables in significance or substantially improve the predictive power of the models for general populations, but it is conceivable that some indexes of air pollution exposure would prove to be predictors of COPD in a particular cohort. The adaptation of existing data or the incorporation of new data on exposure in these cohorts should be explored. Many data sets covering the health status of numerous people are assembled or updated each year by branches of the federal government, by state and local governments,
CONCEPTS AND STRATEGIES IN PLANNING EPIDEMIOLOGIC STUDIES ON AIR 151 POLLUTION and by private companies. Air pollution researchers should be aware of these epidemiologic opportunities, so that they can work with data assemblers to shape the collected data. Through piggybacking on existing data-collection efforts, useful hypothesis-generating studies might be performed, with substantial economies of scale. Perhaps the best examples of such opportunities are several surveys conducted annually (or in some cases less frequently) by the National Center for Health Statistics (NCHS). For instance, each year since the early 1970s, NCHS has randomly surveyed (in the Health Interview Survey, HIS) about 120,000 people, constituting about 40,000 households in all parts of the United States, to determine the health status of the entire population. Data are available in the HIS on many of the variables for which one would want to control in an epidemiologic study. The health component of the HIS includes questions on acute illnesses during a 2-week period before the interview and on chronic illnesses. Although acute respiratory diseases have recently been studied in this manner, the utility of the HIS data for studies of the relation between air pollution and chronic disease remains to be explored.21 Another survey conducted by NCHS, the National Health and Nutrition Examination Survey (NHANES), is in some respects a potentially more valuable epidemiologic tool. Like the HIS, the NHANES elicits information on important socioeconomic and other characteristics of the respondents. However, in addition to respondent-provided data on acute and chronic illness, the NHANES contains data from physician examinations and medical tests involving each respondent. The latter include data on pulmonary function, blood lead content, carboxyhemoglobin content, concentrations of toxic substances in blood or urine samples, and other potentially valuable items. The data might be most profitably merged with air pollution exposure data for generating hypotheses. Both the HIS and the NHANES could benefit from modifications to make them better epidemiologic instruments without detracting from their primary purpose. For instance, neither regularly provides as much information on lifetime smoking history as would be desirable (although excellent supplements are occasionally added to each). Similarly, it would be desirable to have more
CONCEPTS AND STRATEGIES IN PLANNING EPIDEMIOLOGIC STUDIES ON AIR 152 POLLUTION complete data on individual residential and occupational histories, exercise and other personal habits, and perhaps even preventive and curative medical expenditures. In an effort to uncover other data sets of potentially high value but relatively low cost, research planners should investigate the possibility of working with insurance companies, providers of prepaid health care, and agencies that collect DRG records on hospital admissions. In some cases, such groups might have mortality and morbidity data of high quality coupled with data on residential histories. These latter data are important if links between air pollution and chronic illness are to be investigated. By the same token, some international data bases might have great utility in epidemiologic analyses. Data from countries that have national health-insurance programs might permit not only cross-sectional analysis, but also time-series or longitudinal studies. These data, if available, will be most useful if they originate in countries where ambient air pollution data are of high quality. COST AND INFORMATION The feasibility and limits of epidemiologic strategies are determined by their costs, as well as by their benefits. Estimating the productivity and cost- effectiveness of various ways to answer specific questions has become an integral part of epidemiologic research planning. For this purpose, benefits are usually conceived of as information value--the degree to which the study's results bridge the gap between the question and its definitive answer or provide spinoff knowledge for other studies. The benefits of a particular study are a function of its pertinence and of the quality and amount of information it delivers. Comparisons of the relative productivity of optional research strategies and studies might contribute more than any other planning effort to the establishment of research priorities and the ultimate refinement of question and design. Cost considerations can enter at two levels of planning. At the individual study level, the investigator strives to reduce costs as much as possible without compromising the information value of the study. At the strategic level, the planner balances information
CONCEPTS AND STRATEGIES IN PLANNING EPIDEMIOLOGIC STUDIES ON AIR 153 POLLUTION needs versus costs and distributes many kinds of resources over a collection of studies. Planning at these two levels is interdependent, because the relative costs of possible studies are needed as input in the development of the research strategy, and the evolution of the strategy determines which study designs receive the most thorough cost estimation. Cost and productivity analysis must be viewed as an inherent part of the process of planning overall strategy, and not simply in terms of proposals for individual, isolated studies. The range of epidemiologic study costs is particularly wide, and the relationship between cost and information value is remarkably unpredictable. Some very inexpensive studies provide information of enormous value, and some large and expensive ones contribute little. The inferential value of a study often has more to do with its timing and context than with its cost. In most air pollution studies, it is easy to spend large amounts of money by directing resources at the wrong potential sources of error. For example, greater precision in exposure measurements might be unprofitable if sample sizes are inadequate or study populations are poorly characterized. A research program that aims to produce the best possible epidemiologic studies will have to include funding for complementary nonepidemiologic research, including methodologic and technical development and exposure characterization. Technologic research and development, designed to provide better tools for exposure and effect assessment, must be closely woven into epidemiologic research strategies. Some of the barriers to progress are technologic, just as others stem from ignorance of the biology of disease or the limitations of study method. From the viewpoint of the strategic research planner, investment in critical technology is cost-saving--when appropriate technology is lacking, it might be impossible to answer contemporary questions within reasonable costs. The removal of technologic constraints can often be hastened if scientists clearly specify their needs for research and development. An obvious example is the lack of a personal monitoring badge for nitrogen dioxide that can measure short-term exposures. The limits of epidemiologic research depend in part on the changing state of the art of measurement in several disciplines, so they
CONCEPTS AND STRATEGIES IN PLANNING EPIDEMIOLOGIC STUDIES ON AIR 154 POLLUTION differ from one problem to another and from one time to another. THE MILIEU OF EPIDEMIOLOGIC RESEARCH The development of productive epidemiologic strategies in a difficult field like air pollution research will require time, organization, and patience. The creation of the necessary milieu for such research depends on several factors: â¢ Stable administrative and financial support. â¢ Adequate amounts of overall funding. â¢ An appropriate range of contact with sponsoring and regulatory agency personnel. â¢ Provisions for training and career development of new scientists. â¢ Opportunities for assembling multidisciplinary research groups consisting of epidemiologists, atmospheric scientists, statisticians, and health effects scientists. â¢ Opportunities for interaction among epidemiologists, clinicians, and toxicologists. Some potential adverse health effects of current air pollution can be detected only through serial observations over time. Such studies often entail followup of population groups for several years by well-organized research teams led with an exceptional degree of administrative skill. Successful examples of long-term studies include the 25-year British study of a cohort of children from birth to young adulthood and the Harvard Air Pollution Health Study of the growth and decay of lung function.5 26 The need for administrative and financial stability also applies to a sequence of closely linked studies. The Environmental Protection Agency's record of support for epidemiologic studies since 1970 is one of cataclysmic variation, from strong support to nearly complete withdrawal of funds, in spite of heavy reliance on epidemiologic research for risk assessment and standard-setting. A minimal long-term base budget for epidemiologic studies should be considered. With it, the research community
CONCEPTS AND STRATEGIES IN PLANNING EPIDEMIOLOGIC STUDIES ON AIR 155 POLLUTION could develop confidence in the stability of resources necessary for long-term studies of exposed populations. The conduct of long-term studies or of a sequence of studies in a research strategy relies on some degree of isolation from short-term changes in federal research managers and agency priorities. However, a portion of the epidemiologic research program should have the flexibility to respond quickly to short-lived opportunities or demands and in so doing work closely with research sponsors and regulators. The expertise for designing and conducting epidemiologic studies of air pollution is a rare resource that, once established, needs to be cultivated and maintained. Mechanisms to encourage the training of qualified epidemiologists and to assist in their career development are needed. Continued failure to provide the appropriate milieu for studies will have a downward-spiraling effect: many young and qualified investigators will avoid air pollution problems--a situation that will lead to a further decline in the performance of epidemiologic studies, which in turn impedes the recruitment of new trainees. The successful completion of epidemiologic studies on air pollution requires a continuing collaboration among members of various disciplines, including epidemiologists, atmospheric scientists, health effects scientists, and statisticians. It is essential that all members of a research team be active in the initial planning of a study to optimize the use of pertinent information and tools available from different fields. Monitoring specialists must be involved in population selection on the basis of characterization data and must help determine which types of exposure data need to be collected to evaluate the biologic model of concern. It is equally important for health effects scientists to be involved in planning, to ensure that the biologic markers or indexes of specified health effects are matched to the pollutants being monitored. The sensitivity and specificity of techniques, the appropriate time for sampling, and the doses of material received by target organs need to be considered. Statisticians are essential in planning, to help to ensure that data of the appropriate type and amount will be collected and that problems in data reduction, modeling, and analysis are anticipated.
CONCEPTS AND STRATEGIES IN PLANNING EPIDEMIOLOGIC STUDIES ON AIR 156 POLLUTION The difficulty and necessity of establishing multidisciplinary groups in the framework of existing academic and government research institutions must be recognized. And it is necessary for the epidemiology units to interact with clinicians and animal toxicologists in the same or other institutions. Five-year grants, specialized centers of research, and multiyear cooperative agreements are some of the mechanisms possible for creating stable and productive research groups. It is important to incorporate appropriate peer review of these centers and long-term projects into the funding process, to help to ensure quality. EPIDEMIOLOGIC DATA AND THE REGULATORY PROCESS At the heart of the regulatory process for ambient air pollutants in the United States is the determination of approximate âsafeâ exposures to specific pollutants. In some instances, for some aspects of the standards, epidemiologic data can provide the direct basis for such determination. In other instances, epidemiologic studies cannot yet provide the type of quantitative answer required for control of a particular pollutant. Even in the latter case, the determination of a standard will hardly rely on a single definitive study, and standard-setting inevitably involves the careful weighing of many pieces of evidence of various kinds. It is often not recognized that it is the cumulative force of different studies (each of which could be faulted in one respect or other) that gives strength to the data. Each piece of evidence can contribute to regulatory decisions; in the process of making regulatory decisions, the nature and extent of the uncertainty attached to each piece must be carefully evaluated. Although the proper domains of âscienceâ and âregulation,â or risk assessment and risk management, are becoming well understood, problems of communication remain. Those charged with the responsibility of making decisions might find it difficult to understand why straightforward answers to simple questions are not forthcoming. Attempts (praiseworthy in themselves) to avoid overinterpretation of data can be interpreted as disbelief in the significance of any data; efforts to reduce what must remain subjective to a formal quasi- mathematical process, which the data base might be too
CONCEPTS AND STRATEGIES IN PLANNING EPIDEMIOLOGIC STUDIES ON AIR 157 POLLUTION fragile to support, might mislead those who are unfamiliar with the details of studies into believing that the data are stronger than they really are. Such difficulties cannot be resolved by refinement of data, although that is always desirable. What is needed is the opportunity for those who analyze the data in detail or participate in collecting them to exchange ideas with those who are involved with policy, until each side is sure that the other fully grasps the strengths and weaknesses of the data being relied on. One of the objectives of this chapter has been to describe how uncertainty in epidemiologic studies of air pollution might be characterized, so that it can be captured and conveyed properly to policy-makers. This uncertainty must be weighed against the uncertainty that has been built into the process for setting standards and margins of safety. Epidemiology has many uses in protecting the public from air pollution, beyond the setting of specific standards. Epidemiologic studies can point out areas or major pollution sources that need regulation or some form of intervention and can monitor the health of communities, to assess the results of favorable or unfavorable changes in pollution. Research that is directly applicable to standard-setting should not always take priority over research that is needed for the understanding of broader health issues. Finally, evidence of safety is not the converse of evidence of risk; safety is often much harder to demonstrate. That is the critical difference between the interpretation of studies that find an effect and studies that do not. There are several reasons for the difference: statistical power of studies might be adequate to show large effects, if they are present, but not to exclude smaller effects that are of regulatory interest; nonrandom errors might be more easily dealt with in positive studies; clear demonstration of an effect in one population segment might create a presumption of effects elsewhere that is not balanced by similarly clear demonstration of the lack of effects in other segments; and control measures might be required if a small part of the population is affected, but not rendered superfluous if a large part is shown to be unaffected. Thus, good large studies with negative results can often have less meaning
CONCEPTS AND STRATEGIES IN PLANNING EPIDEMIOLOGIC STUDIES ON AIR 158 POLLUTION than smaller poorer studies that are positive. Fairness is not the issue; negative studies, no matter how large and well conducted, can rarely be used to exclude important health effects in other populations and other circumstances. REFERENCES 1. Bennet, B.G., J.G. Kretzschmar, G.G. Akkland, and H.W. de Koning. Urban air pollution worldwide. Environ. Sci. Technol. 19:298-304, 1985. 2. Berry, G. Longitudinal observations: Their usefulness and limitations with special reference to the forced expiratory volume. Bull. Physio-pathol. Respir. 10:643-655, 1974. 3. Cerkez, R., and A. Knezevic. Study of Pollution and Management of the Environment in the Urban Area of Sarajevo. Sarajevo, Yugoslavia: Institute of Hygiene and Environmental Protection and Radiological Protection, 1983. 60 pp. 4. Cohen, B.H., W.C. Ball, Jr., W.B. Bias, S. Brashears, G.A. Chase, E.L. Diamond, S.H. Hsu, P. Kreiss, D.A. Levy, H.A. Menkes, S. Permutt, and M.S. Tockman. A genetic-epidemiologic study of chronic obstructive pulmonary disease: 1. Study design and preliminary observations. Johns Hopkins Med. J. 137:95-104, 1975. 5. Colley, J.R.T., and D.D. Reid. Urban and social origins of childhood bronchitis in England and Wales. Br. Med. J. 2:213, 1970. 6. Ferris, B.G., Jr., F.E. Speizer, J. Worcester, and H.Y. Chen. Adult mortality in Berlin, N.H., from 1961 to 1967. Arch. Environ. Health 23:434-439, 1971. 7. Glindmeyer, H.W., J.E. Diem, R.M. Jones, and H. Weill. Non-comparability of longitudinally and cross-sectionally determined annual change in spirometry. Am. Rev. Respir. Dis. 125:544-548, 1982.
CONCEPTS AND STRATEGIES IN PLANNING EPIDEMIOLOGIC STUDIES ON AIR 159 POLLUTION 8. Greaves, I., B.G. Ferris, Jr., W.A. Burgess, and D. Essex. Respiratory effects of sulfur dioxide (SO2) among corn refinery workers. Am. Rev. Respir. Dis. 129:157, 1984. 9. Higgins, I.T.T. Trends in respiratory cancer mortality: In the United States and in England and Wales. Arch. Environ. Health 28:121-129, 1974. 10. Higgins, M.W., J.B. Keller, J.R. Landis, T.H. Beaty, B. Burrows, D. Demets, J.E. Diem, I.T. Higgins, E. Lakatos, M.D. Lebowitz, H. Menkes, F.E. Speizer, I.B. Tager, and H. Weill. Risk of chronic obstructive pulmonary disease: Collaborative assessment of the validity of the Tecumseh index of risk. Am. Rev. Respir. Dis. 130:380-385, 1984. 11. Kilburn, K.H., R. Warshaw, and J.C. Thornton. Pulmonary functional impairment and symptoms in women in the Los Angeles Harbor Area. Am. J. Med., 1985. (in press) 12. Lebowitz, M.D., R.J. Knudson, and B. Burrows. Tucson epidemiologic study of obstructive lung diseases: I. Methodology and prevalence of disease. Am. J. Epidemiol. 102:137-152, 1975. 13. Mason, T.J., J.F. Fraumeni, Jr., R. Hoover, and W.J. Blot. An Atlas of Mortality from Selected Diseases. NIH Publication No. 81-2397. Bethesda, Md.: U.S. Department of Health and Human Services, National Cancer Institute, 1981. 309 pp. 14. Monson, R.R. Occupational Epidemiology. Boca Raton, Fla.: CRC Press, 1980. 219 pp. 15. Morgenstern, H. Uses of ecologic analysis in epidemiologic research. Am. J. Pub. Health 72:1336-1344, 1983. 16. Morris, J. Uses of Epidemiology. New York: Churchill Livingstone, 1975. 262 pp. 17. Pan American Health Organization. Red Panamericana de Muestro de la Contaminacion del Aire, Informe Final 1967-1980. (Pan American Network for the Sampling of Contaminated Air, Final Report 1967-1980). Serie Technica No. 23, 1982. 65 pp.
CONCEPTS AND STRATEGIES IN PLANNING EPIDEMIOLOGIC STUDIES ON AIR 160 POLLUTION 18. Pengelly, L.D., A.T. Kerigan, C.H. Goldsmith, and E.M. Inman. The Hamilton study: Distribution of factors confounding the relationship between air quality and respiratory health. J. Air Pollut. Control Assoc. 34:1039-1043, 1984. 19. Piomelli, S., L. Corash, M.B. Corash, C. Seaman, P. Mushak, B. Glover, and R. Padgett. Blood lead concentrations in a remote Himalayan population. Science 210:1135-1137, 1980. 20. Polissar, L. The effect of migration on comparison of disease rates in geographic studies in the United States. Am. J. Epidemiol. 111:175-182, 1980. 21. Portney, P.R., and J. Mullahy. Urban air quality and acute respiratory illness. J. Urban Econ. (in press) 22. Reid, D.D., J. Cornfield, R.E. Markush, S. Siegel, E. Penderson, and W. Heintzel. Studies of disease among migrants and native populations in Great Britain, Norway, and the United States: III. Prevalence of cardiorespiratory symptoms among migrants and native-born in the United States. Nat. Cancer Inst. Monogr. 19:321-346, 1966. 23. Rose, G. Sick individuals and sick populations. Int. J. Epidemiol. 14:32-38, 1985. 24. Tashkin, D.P., V.A. Clark, A.H. Coulson, M. Simmons, L.B. Bourque, C. Reems, R. Detels, J.W. Sayre, and S.N. Rokaw. The UCLA population studies of chronic obstructive respiratory disease: VIII. Effects of smoking cessation on lung function: A prospective study of a free- living population. Am. Rev. Respir. Dis. 130:707-715, 1984. 25. Van der Lende, R., T. Kok, R. Peset, Ph.H. Quanjer, J.P. Schouten, and N.G.M. Orie. Longterm Exposure to air pollution and decline in VC and FEV1: Recent results from a longitudinal epidemiologic study in the Netherlands. Chest 80(Suppl.):23S-26S, 1981.
CONCEPTS AND STRATEGIES IN PLANNING EPIDEMIOLOGIC STUDIES ON AIR 161 POLLUTION 26. Ware, J.H., Thibodeau, L.A., F.E. Speizer, S. Colome, and B.G. Ferris, Jr. Assessment of health effects of atmospheric sulfur oxides and particulate matter: Evidence from observational studies. Environ. Health Perspect. 41:255-276, 1981. 27. Willett, W.C., B.F. Polk, B.A. Underwood, and C.G. Hames. Hypertension detection and follow- up program study of serum retinol, retinol-binding protein, total carotenoids, and cancer risk: A summary. J. Nat. Cancer Inst. 73:1459-1462, 1984.
CONCEPTS AND STRATEGIES IN PLANNING EPIDEMIOLOGIC STUDIES ON AIR 162 POLLUTION
163 Chapter 5 The Application of Epidemiology to Selected Research Questions