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.
Basic Epidemiologic Issues The diseases resulting from exposure to ionizing radiation are usually indis- tinguishable from the diseases that occur in the general population, Leukemia looks the same in an atomic bomb survivor as it does in a person who was not exposed to an atomic bomb blast. Only the increased frequency in an exposed group indicates that radiation exposure may have played a role. However, in- creased frequency alone is usually not enough to establish a causal relationship between an exposure and a health effect. To help evaluate causality it is also necessary to consider the following questions (Hill, 1965~: · Does the frequency of the disease increase as the dose there a dose-response relationship)? increases (that is, is · Can the findings be duplicated by other investigators? · Is a similar effect seen in experimentally exposed animals (some species may not be susceptible) and other laboratory studies? · Can alternative explanations, such as cigarette smoking and heredity be excluded? · Is the finding biologically plausible? · Does the effect cease to occur when the cause is removed? The history of the discovery of an intrauterine effect of radiation is illustra- tive. In the 1920s physicians began to note that mothers of infants with small 13
14 AD VERSE REPRODUCTI HE OUTCOMES head size and mental retardation had often received radiotherapy early in pregnancy (Murphy, 1928). Within a few years these case reports led to a U.S. mail survey of these occurrences, well documented in hospitals, that doubled the known number of cases to 30 (Goldstein and Murphy, 1929~. Soon after, the same effect was induced experimen- tally in rats, and extensive mouse studies demonstrated a close relationship between de- velopmental stage at exposure and effect observed. This finding was confirmed by the study of Japanese atomic bomb survivors exposed in utero (Plummer, 1952; Miller, 1956; Yamazaki and Schull, 1990), in whom the incidence of the effect could be deter- mined. They were from a defined population exposed to a variety of doses and could be studied as a cohort, that is a group exposed together, in this instance, at a single moment in time, for the full spectrum of diseases that occur over the entire life span. The results confirmed those in the earlier case series. In addition, a dose-response relationship that could not be attributed to other variables such as malnutrition was found, the effect was biologically plausible, and the incidence returned to normal in children born subse- quently to those who had been exposed in utero. Small head size is a teratogenic effect (induced by exposure of the developing fetus to the teratogen in contrast to a genetic effect, which is passed from parent to child through a defective gene. Suspicion that radiation was responsible for this effect began with a number of case reports. Similar disease cluster investigations have led to much of what is known about environmental causes of cancer or birth defects (Miller, 19781. Clusters are generally defined as aggregations of events in space and time. They pose a challenge because they are often based on small numbers, are epidemiologically com- plex since a multitude of potential risk factors need to be considered, require compari- sons with background rates that are often difficult to obtain, and involve statistical and mapping techniques that are controversial (Rothenburg et al., 19901. For example, the study of more than 100 cancer clusters by experts at the U.S. Public Health Service's Centers for Disease Control from 1961 to 1982 revealed no environmental causes (Caldwell, 1990) for these clusters. The reason is that almost all small clusters occur by chance even for such rare diseases as childhood leukemia. When a health professional, a parent, or anyone else notices a group of cases in a small area, there is the tendency to draw boundaries of time and geography tightly around the cases. As a result, the occur- rence of disease appears to be unusually high in this defined area. Cancer clusters occur by chance continuously throughout the United States (Neutra et al., 1990~. When it seems an environmental exposure might be responsible, experienced investigators may conduct a case-control study to determine what in the environment is suspect. In such a study, the histories of affected individuals ("cases") are compared with those of similar unaffected persons ("controls") to determine if the exposure in question was significantly more frequent among cases than among controls. However, the elements necessary to help establish causality outlined above must also be evaluated. If the environmental exposure has not been found to cause the cancer or some other disease after heavy expo- sures, as in industry, after accidental exposures, or after receiving medications, it is not plausible to expect an effect at low doses.
BASIC EPIDEMIOLOGIC ISSUES 15 Family clusters of disease are often genetic in origin, but for noninfectious diseases, family clusters of disease are rarely due to a shared environment. Important heritable cancers, such as retinoblastoma in children (Knudson, 1988) and diverse forms of can- cers that occur before 45 years of age in individuals with Li-Fraumeni syndrome (Li et al., 1988), have been identified through the study of such family clusters Clinical iden- tification of these rare disorders has led to new understanding of the genesis of many common cancers (Levine, 19951. The study of unusual family cancer clusters due to genetics has been especially rewarding. Because cancer occurs so often in the popula- tion at large, family clusters most commonly occur by chance. Subtle exposures such as diet are often suspected, but they are difficult to establish. The interpretation of clusters of adverse pregnancy outcomes are particularly prob- lematic to public health officials, epidemiologists, and biostatisticians. The same diffi- culties described above would apply to an even greater extent when assessing multiple adverse reproductive events within a family. Guidelines for cluster investigations have recently been set forth by the Centers for Disease Control (19901. Any study that focuses on a single potential risk factor for disease needs to consider the established and probable risk factors as potentially confounding variables. Possible interactions among risk factors are often difficult to interpret because they may occur by chance in studies that consider a large number of factors. Selection bias can occur if, for example, the participants in the study differ from the control group in aspects other than the factor under study. Such bias can affect the generalizability of the findings and, more importantly, measures of association (Kelsey et al., 19861. The foregoing illustrates the epidemiologic approach to investigating the causers) of disease. One function of epidemiology is to measure disease frequency, which can be expressed as prevalence (the number of affected people in the population at a given time) or incidence (the number of new cases in a given time interval). These studies, when properly done, provide numerators (the number of people affected) and denominators (the number of people at risk), which are the bases of estimating risk. One of the most important considerations in interpreting epidemiologic data is the strength or magnitude of an association between exposure and disease. A number of terms are used to express strength of association or risk, and it is often very difficult to compare one epidemiologic study with another because of the different terms and types of analysis used. The most commonly used term for cohort studies is relative risk (RR) or excess relative risk (ERR). They both are an expression of the risk of the exposed group relative to that of some nonexposed group. A relative risk of 1.0 means that the exposed group has the same risk as the control group, and implies an excess relative risk of 0. If there is a relative risk of 2, then the exposed population would be twice as likely as a nonexposed control group to develop a condition as a result of exposure. This equates to a doubling of the risk. The control group must be essentially the same age, sex, and so forth, as the exposed group or these variables should be considered in the analysis. An increased relative risk does not necessarily mean that there has been an effect of a given exposure. For example, if two populations are being compared and the control
16 ADVERSE REPRODUCTIVE OUTCOMES group for some reason has a lower disease incidence or mortality rate than would nor- mally be expected, the relative risk will be greater than 1.0 even though there is no ex- cess incidence of disease or mortality in the exposed population. For this reason an in- creased relative risk or excess relative risk should not be viewed as establishing causality until such factors have been clarified. Typically, the estimation of risk is accompanied by the assessment of random variation. This is accomplished through the use of statistical significance testing. Using an appropriate test for the type of data at hand, the investigator will derive the p-value, which is the probability that an effect as extreme as that observed could have occurred by chance, given that there is, in fact, no relationship or association between exposure and disease (the null hypothesis). By convention, if the p-value is less than or equal to 0.05, then the association between exposure and disease is considered to be statistically significant. This means that there is no more than a 5 percent or a 1-in-20 probability of observing a result as extreme as that observed because of chance alone if there is, in fact, no association. If the p-value is greater than 0.05, then the effect is considered to be not statistically significant. The current recommended practice in medical and epidemi- ologic research is also to report an informative measure the confidence interval. The confidence interval represents the range of possible values for the parameter of interest (e.g., the relative risk) that is consistent with the observed data within specified limits. The width of the confidence interval reflects the sample size, in that the narrower the interval, the less variability in the estimate of the effect measured; likewise, the greater the width, the greater the variability. A potential problem in epidemiologic studies that sometimes goes unrecognized is bias of reporting or in selecting study participants. A mailed questionnaire may draw a disproportionately higher number of responses from exposed individuals who think that they have been injured as a result of their exposure than from unexposed individuals or those who do not believe that they have been injured as a result of their exposure. Simi- larly, investigators may publish positive results but may not publish negative results, and scientific journals generally favor reports of positive results. Consequently, the public is often left with a distorted view of scientific reality. It warrants noting, too, that case re- ports and case series commonly result in false interpretations rather than new insights into etiology.