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.
7 Data on Humans: Clinical and Epidemiotogical Studies The implications for human health posed by exposure to environmental contaminants, including those in drinking water, can be derived from studies in laboratory animals or in in vitro cell and tissue systems, from reports of clinical observations of isolated exposed individuals or exper- imental exposures and intervention studies in humans, or from direct observations of exposed human populations. To determine the effects on human health, the latter approaches are clearly preferable, but they are constrained by ethical considerations limiting experimentation to certain types of studies designed to prevent or treat disease. Thus, the data on humans generally fall into one of two categories: clinical data, which describe the effects of specific agents on certain individuals, and epide- miological data, which reveal patterns of disease or death in groups of humans exposed to single agents or to a variety of substances. CLINICAL OBSERVATIONS: CASE REPORTS Much of the knowledge concerning the human health effects of toxic materials comes from clinical information amassed after exposure to high doses, such as poisonings and industrial accidents. Such reports typically identify acute effects but can at times also describe chronic, delayed-onset diseases. For instance, Letz et al. (1984) reported two fatalities among workers following an acute inhalation exposure to ethylene dibromide that also produced liver and kidney damage. Such case reports are valuable for describing clinical and pathological effects in humans and suggesting underlying cause-and-effect relationships. When exposure levels can be 226
Data on Humans 227 documented, they may help in defining dose-response relationships in humans. More commonly, broad comparisons are made, as in a study comparing elevated immunoglobulin levels in controls with levels in work- ers who had inhaled pentachlorophenol (Zober et al., 19811. However, it is usually not possible to link a specific exposure with specific illnesses, other than in illnesses with short latency periods (days or weeks), because the clinical features are not unique for many exposures. Thus the neu- rotoxicity of acrylamide, n-hexane, and methyl n-butyl ketone and the hepatotoxicity of 1,2-dichloropropane, pentachlorophenol, and ethylene dibromide (see Chapter 9) have been identified in humans largely because illness or adverse health effects, such as reduced reproductive capacity, occurred shortly after exposure. In general, one can attribute delayed health effects to specific exposures only when there is a unique clinical manifestation. Toxic agents that might cause syndromes with clinical patterns that are not unique are unlikely to be associated with those syndromes. Occasionally, e.g., for Kepone- induced neuropathy (Cannon et al., 1978), a distinctive clinical pattern of illness as well as short latency reinforces a suspected exposure-response relationship. Other illnesses or health effects sufficiently unusual or dis- tinctive to link disease and exposure include sterility in men (Whorton et al., 1977), a reduced proportion of male offspring following exposure to dibromochloropropane (DBCP) (Potashnik et al., 1984) (see Chapter 9), hepatic angiosarcoma following exposure to vinyl chloride monomer (Creech and Johnson, 1974), and the appearance of pleural mesotheliomas many years after exposure to asbestos (Wagner et al., 1960~. Observations of similar illnesses in laboratory animals experimentally exposed to DBCP, 1,3-dichloropropene, bis~chloromethyl) ether, 4-aminobiphenyl, and vinyl chloride supplement the findings in humans. For several of these materials, the evidence for carcinogenicity was found first in test animals and later confirmed in humans (Davis and Mandula, 19851. When there is both delayed onset of effects and a lack of distinctive clinical characteristics, risk assessments must be based on data from an- alytical epidemiological studies, clinical or experimental case reports, and extrapolation from laboratory experiments. Because of the long latency periods, difficulties encountered in making quantitative estimates of ex- posure, and problems in identifying and conducting follow-up studies of exposed populations, epidemiological data are unlikely to be available for most carcinogens and other chronic-disease agents found in drinking water until many years after first exposures. Clinical observations are thus im- portant for suggesting potential exposure-disease relationships. They in- volve studying individual cases of illness or investigating cases with similar clinical features that appear to cluster in time and place or in relation to certain personal or exposure characteristics. A description of two cases
228 DRINKING WATER AND H"LTH of lymphoma that occurred 6 years after accidental exposure to 1 ,3-dichlo- ropropene (Markovitz and Crosby, 1984) illustrates the potential value of such hypothesis-generating clinical reports (see Chapter 91. EPI DEM IOLOG ICAL OBSERVATIONS Most epidemiological studies conducted to date have focused on the evaluation of accidental or workplace exposures. There have been rela- tively few studies of humans exposed to contaminants in drinking water. As an observational science, epidemiology attempts to assemble data answering the basic investigational questions: Who? What? Where? When? How? Who have been adversely affected by exposures? What did they do, or what were the substances to which they were exposed? Where did these events take place? When did exposure occur? How did the exposure come about? And how can the exposure be linked to the observed illness? For each of these questions, the student of human disease must consider whether the observed events could be explained in some way other than through exposure to the substance under consideration. In other words: Who else? What else? Where else? At what other times? In what other ways? In laboratory studies in animals and in in vitro systems, strictly observed protocols can ensure that there are no confounding variables to cloud interpretation of the results of the specific exposure of interest. In contrast, many populations include people who smoke cigarettes, some people whose jobs involve exposure to some chemicals, some who live in rural areas, and others with different life-styles and, thus, exposure opportun- ities. The constraints people place on themselves (or that are placed on them) or the choices they make create a multitude of separate subpopu- lations that usually cannot be compared in simple ways. Underlying many epidemiological studies is the assumption that ex- posure, however estimated, provides an adequate surrogate for the dose of a pollutant delivered in laboratory studies. Recent advances in molecular and biochemical epidemiology have clarified the definition of exposure along with the meanings of other terms used in the field of epidemiology. Exposure commonly refers to levels of pollutants measured externally in the physical environment. Internal dose indicates the quantity of the sub- stance or its metabolites that reaches body tissues. Biologically elective dose designates the quantity of a pollutant or its metabolites that interacts with a particular target tissue or a surrogate for that target tissue (Perera and Weinstein, 19821. In this report, unless otherwise specified, the term exposure refers generically to all aspects of exposure, from levels in the surrounding environment to doses at a target tissue.
Data on Humans 229 Early Studies The beginning of modern epidemiology can be traced back to Snow's analysis of the patterns of cholera mortality in London during 1853 and 1854, which linked health risks to consumption of polluted water (Snow, 1855~. Using available mortality data, Snow was able to pinpoint a source of the disease and recommend a means to neutralize it (removing the handle of the water pump) 28 years before Koch identified Vibrio cholerae as the cholera bacillus in 1883 (Black, 19801. Snow's study had two major parts: detailed observations concerning an explosive local outbreak attrib- uted to infected water that appeared to come from a single neighborhood pump and a comparison of death rates in two intermingled populations that were distinguished only by their use of two different sources of drinking water one relatively polluted and the other not. Snow's work underscores a basic feature of epidemiological evidence: although it de- pends upon observations of individuals, such data can only be interpreted epidemiologically in the context of the exposed population as a whole. Descriptive Epidemiology In looking at the occurrence of disease in a population, epidemiologists often begin by describing the specific disease and the conditions that previous knowledge suggests may be related to it. Attempts are made to discover if the disease incidence has increased over time and whether specific segments of a population are especially affected. By comparing increases in disease incidence with changes in possibly related exposures, hypotheses are generated for testing in more clearly focused studies. Geo- graphical and demographic characteristics (age, sex, race) are also con- sidered to sharpen the hypotheses to be examined. For example, some cancer mortality rates were higher in Louisiana parishes that obtained nearly all their drinking water from the lower Mississippi River basin in contrast to parishes that consumed little such water. This finding supported the need for further testing of an exposure-disease hypothesis (Page et al., 19761. Among the major contaminants found in the river water were the trihalomethanes. Reviewing 14 epidemiological studies relating cancer mortality to drinking water containing trihalomethanes, Williamson (1981) found some mixed evidence of increased risks of bladder, colon, and rectal cancer (see also NRC, 1980, pp. 6-7~. Epidemiological research on arsenic contamination of drinking water has also produced differing results. For example, in a geographic area of Taiwan with a stable population, many people have been exposed to drinking water with high arsenic levels for 40 years (Tseng, 1977; Tseng et al 19681 A similar long-term exposure was documented in a popu-
230 DRINKING WATER AND HEALTH ration living in Antofagasta, Chile (Borgono et al., 19771. People in both of these populations experienced endemic chronic arsenic poisoning, which was associated with elevations in skin cancer. After the installation of a water treatment plant in Chile, disease patterns changed, further strength- ening the link between cause and effect. However, smaller arsenic-exposed populations observed for shorter periods in Alaska (Harnngton et al., 1978) and Oregon (Morton et al., 1976) had no significantly elevated risk for skin cancer. These different findings may reflect differences in the re- spective study designs. Two major limitations in descriptive studies are common to other study designs as well: the lack of specific information over periods of time on drinking water contamination and the inability to control for potential confounding variables, such as cigarette smoking, workplace or other indoor exposures, and nutrition. Hence, the multiple regression analyses used in descriptive studies are best suited to generating hypotheses for additional, controlled investigation. Evidence from such studies must be weighed against the presence of many other suspected or known causes of cancer (DeRouen and Diem, 1975) and against the criteria for consis- tency and plausibility elaborated below. In the meantime, the hypothesized relationship between exposure to carcinogens in drinking water and cancer patterns is undergoing serious examination. Exposure assessment is a pivotal part of environmental epidemiology. For example, the National Human Monitoring Program of the U.S. En- vironmental Protection Agency assessed exposure of 21,000 subjects to selected pesticides through analysis of blood and urine specimens (Murphy and Harvey, 19851. These data were collected as part of the Second National Health and Nutrition Examination Survey (NHANES II) a 4- year study conducted by the National Center for Health Statistics to collect dietary intake data as well as clinical, biochemical, and anthropometric data to determine the health and nutritional status of the U.S. population. Statistically weighted results of the blood serum analyses indicated that the general population is being exposed to some of these pesticides. For example, NHANES II data indicated that 31% of the 12- to 74-year-old subjects living in the Northeast, Midwest, and South had been exposed to p,p'-dichlorodiphenyltrichloroethane (DDT) (median level 3.3 ppb); 99% had been exposed to p,p'-dichlorodiphenyldichloroethylene (DDE), a metabolite of DDT (median level 11.8 ppb); and 13.9% had been exposed to ,B-benzene hexachloride (median level 1.7 ppb) (see Figures 7- 1 through 7-31. Analytical Epidemiology Follow-up epidemiological studies are generally grouped under He heading analytical epidemiology, although all epidemiological studies include anal-
Data on Humans 23 ~ 60 50 - ~n o an LU Cal UJ 40 30 20 10 o 50.9 31 .0 a 3.3 ppb ~4 . 12-74 28.4 2.7 ppb 1 2-24 25-44 . ~ . .. . A. 45-74 AGE GROUP(years) Median Exposure FIGURE 7-1 Percent positive and median p,p'-DDT levels in blood serum for positives by age. Limit of detectability, 2 ppb. Data for Northeast, Midwest, and South only. Data from Murphy and Harvey, 1985. ysis. Analytical epidemiology studies include case-companson studies, cohort studies, some mixtures of the two, and experimental (or interven- tion) studies. Recognizing that descriptive epidemiological studies do not provide proof of cause-and-effect relationships, Hill (1953) proposed sev- eral criteria for examining He resultant data and strengthening the con- clusions derived from them. Did the putative cause precede the observed effect in time? Was the observed effect pervasive, and did it occur in a large segment of the population? Was it seen at more than one time and in more than one place? After intervention, was a reduction in exposure 100 99 3 99.3 > 80 - cn o 60 40 20 o 99.3 99.2 11.8 ppb 5.9 ppb 12-74 1 2-24 12.0 ppb . 8.3 ppb: ;..,..;. ;. ;.;.., I.;;.. ,., ...; ...; 1 25-44 45-74 AGE GROUP (years) Median Exposure FIGURE 7-2 Percent positive and median p,p'-DDE levels in blood serum for positives by age. Limit of detectability, 1 ppb. Data for Northeast, Midwest, and South only. Data from Murphy and Harvey, 1985.
232 DRINKING WATER AND HEALTH 50 40 en o Cal Z 20 30 10 To ~_1 .4 ppb 12-74 12-24 25-44 45-74 26.8 Median Exposure AGE GROUP (years) FIGURE 7-3 Percent positive and median '-benzene hexachlor~de levels in blood serum for positives by age. Limit of detectability, 1 ppb. Data for Northeast, Midwest, and South only. Data from Murphy and Harvey, 1985. followed by a reduction in disease? Is there a reasonable biological basis for linking the presumed cause and effect? Is the effect consistent among studies? Is it not explainable by other known or suspected causes? Is the effect specific to this cause, and does this presumed cause lead only to this effect? These last considerations do not apply to diseases that may have multiple causes as, for example, canceror to single exposures that can lead to several end results (e.g., cigarette smoking, which can lead to different kinds of cancer, low birth weight of infants, and heart and circulatory illnesses). Case-Comparison and Cohort Studies Epidemiological studies can attempt to discover the etiology of a disease by starting from either the exposure or the biological effect. Thus, tests of hypotheses suggested by descriptive studies can start with illnesses and evaluate different exposure histories of the sick and the well (case- comparison studies), or they can first observe the patterns of exposure in populations and then note whether the more extensively exposed develop more (or more serious) illnesses than the unexposed or less exposed (cohort studies). When the nature of exposure and disease permit, both kinds of observations may be made concurrently (cross-sectional studies), as with sperm counts measured in DBCP production workers (Whorton et al., 19791. Case-comparison studies are useful in that they can be performed in small groups of subjects over relatively short periods. For instance, in a case-comparison study conducted in rural South Australia on 218 matched
Data on Humans 233 case-control pairs, an association was found between maternal prenatal groundwater consumption and congenital malformations of the central nenous system and musculoskeletal system (Dorsch et al., 19841. Women who bore a malformed child were three times more likely to have consumed groundwater during pregnancy rather than rainwater. Further analysis showed an increased risk associated with nitrate concentrations in drinking water. Women who drank water containing 5 to 15 ppm nitrate had a nearly threefold increase in risk, and those who drank water containing concen- trations higher than 15 ppm had a risk four times greater than that of controls. Case-comparison studies deal only with a specific illness or cause of death. Thus, they are unable to determine whether the exposure under study is capable of increasing or decreasing the incidence of other illnesses or other causes of death. These studies also move backward in time; i.e., attempts are made to associate changes in morbidity or mortality rates with some previous exposure alteration. In contrast, cohort studies move forward in time. However, such studies are more difficult to use when considering rare disorders such as some cancers, since very large sample sizes are often needed to detect even large excesses in risk. In some situations, morbidity or mortality data assembled in a cohort study can provide the basis for case-comparison studies within that study ("nested" case-comparison studies), thereby enabling investigators to make more detailed assessments of risk (Marsh, 1983~. Experimental or Intervention Studies When a well-defined cause-and-effect relationship has apparently been demonstrated, it is sometimes possible to test the presumed relationship by eliminating or reducing the cause or causes. The elimination or re- duction can be part of a well-designed randomized trial as in the MRFIT (Multiple Risk Factor Intervention Trial) studies of the National Heart, Lung, and Blood Institute (Tillotson and Hulley, 1985) or in studies of the effect of fluoride-treated drinking water on health. A variety of epi- demiological studies have documented a number of health effects of fluo- ridation. Low levels of fluoridation prevent tooth decay (Arnold et al., 1953; Carlos et al., 1962; Dean et al., 1950) and osteoporosis (Bernstein et al., 1966) while increasing bone density (Leone et al., 1955, 19604. However, excessive levels of fluoride can produce fluorosis (Geever et al., l958a,b; Leone et al., 19541. Evaluations of the effects of intervention are usually based on time- trend or time-and-place considerations, such as whether the rate has de- clined following the intervention, allowing enough lag time for effects to
234 DRINKING WATER AND HEATH develop. Thus, the epidemiological process comes full circle, starting with descriptive epidemiology suggesting possible cause-and-effect relation- ships, then to uncovering these relationships through case-comparison and cohort studies, to specific interventions (both planned and unplanned), to the evaluation of these interventions through descriptive epidemiological techniques to verify the effect in a controlled trial. PR I NCI PLES, PROBLEMS, AN D Ll M ITATIONS OF EPI DEM IOLOGY To conduct a satisfactory epidemiological study whose results would be of material use in risk estimation, consideration must be given to four . . . mayor criteria: · The characteristics of the study population must be suitable for the goal of the study. · The studies must be designed and conducted in a manner that ensures unbiased answers. · The studies should be large enough to have the power needed to detect small but important effects. · The exposure data should be of sufficient quality to permit estimation of dose-response relationships. Several aspects of epidemiological studies affect the above criteria. In an effective study, the study population must be large enough to have a high probability of detecting a true effect as "statistically significant," if it is present. Ideally there should be exposures of consequence only to the substance in question, and no other elements capable of affecting the frequency of the same disease should be present. A control population- ideally comparable in all respects except for exposure to the suspect material should be available for comparison. In reviewing the epidemiological evidence on 75 substances used in active commerce that had been found to be carcinogens in laboratory animals, Karstadt et al. (1981) found that fewer than one-third had been or were being subjected to effective epidemiological research. In the in- dustrial settings, they found only small groups of exposed workers and workers who had been exposed to more than one agent. Often, active use of the material had been discontinued, or it had been used so recently that not enough time had lapsed to expect that a meaningful number of cancers would have developed (see also Karstadt and Bobal, 1982~. In reviewing the possibilities for conducting epidemiological studies on di(2-ethylhexyl) phthalate, the National Institute for Occupational Safety and Health found similar conditions and did not undertake such a study (Roberts, 19831.
Data on Humans 235 TABLE 7-1 Possible Study Outcomes Empirical Finding Actual Conditions Positive No Effect Positive True positive False negative (Type II error: p) No effect False positive True negative (Type I error: ~x) As noted earlier, for valid inferences to be made regarding risk of illness in relation to specific exposure, the populations being examined should ideally be comparable in all respects except for exposure to the particular characteristic under study. In his study of cholera, Snow (1855) achieved comparability by the fortuitous design of the natural experiment he ob- served: households receiving either polluted or unpolluted water were side by side in the same community and presumably indistinguishable in all other respects. Thus, factors such as differences in socioeconomic status and place of residence were not present to distort the association between etiological agent and biological effect and thus lead to mistaken conclu- sions. Some statistical techniques permit adjustment when strict compa- rability is not achieved, but they are not always successful. Statistical Significance and Power There are four possible outcomes to any epidemiological study, as indicated in Table 7-1. The study can report an effect (i.e., be positive) or report no effect (the so-called negative study). In reporting an effect or the absence of one, a study can be accurate, i.e., give a true-positive or a true-negative result. If the study produces erroneous findings, it gives a false-positive or a false-negative result. A false-positive result asserts that increased risk exists when in fact it does not (known as a Type I error). A false-negative result asserts that increased risk does not exist when in fact it does (a Type II error). The probability that a study will make a false-positive error is the same as the level of statistical significance orp-value given in most reports as 0.05 or less when single comparisons are being made. It is the probability that the event noted could be due to chance alone and not to any cause-and-effect relationship. The probability of detecting as statistically significant an effect if it is truly present is the power of a study. As power increases, the chances of producing a false-negative error decrease. Power is determined jointly by the level of relative risk being assessed, i.e., the magnitude of the effect, and by the number of cases that would be expected in the exposed pop-
236 DRINKING WATER AND H EALTH TABLE 7-2 Power for Detecting Different Levels of Increased Relative Risk in Relation to Numbers of Expected Casesa Number of Relative Riskb Cases 1.2 1.5 2.0 3.0 5.0 1 0.073 0.116 0.208 0.428 0.796 5 O.111 0.262 0.582 0.948 1.000 10 0.149 0.411 0.836 0.999 1.000 25 0.245 0.726 0.994 1.000 50 0.384 0.937 1.000 100 0.605 0.998 200 0.854 aDavis et al., 1985. bCalculations made using one-sided test and significance level of 0.05. ulation if there were no increased risk. This expected number depends on the sample size and expected disease frequencies in some appropriate comparison populations. Studies of small populations with common dis- eases can have the same power as studies of rare diseases in large pop- ulations. Values of power for different combinations of expected numbers of cases and relative risk values are shown in Table 7-2 (Davis et al., 19851. Not surprisingly, studies of larger samples, and therefore more expected cases, have sufficient power to detect smaller increases in risk. Conversely, studies of small samples (and few expected cases) will have power sufficient to detect only large increases in relative risk. For example, a small study, in which only one case may be expected, will have approximately 0.80 power to identify correctly a relative risk of 5.0 (a fivefold, or 400%, increase in risk), but only about 0.07 power to identify a relative risk of 1.2 (a 20% increase in risk) at a significance level of 0.05. A study with a sample size large enough to expect 200 cases, however, will have about 0.85 power to detect a 20% increase in risk. In designing experimental studies, power should be set at about 0.90 to 0.95. When it is likely that a study will be conducted only once, or at best a few times, as is sometimes the case for epidemiological studies, higher power should be sought in the original design. (For further discussion on determining power see Cohen, 1977.) The finding of an association between hepatic angiosarcoma and ex- posures of vinyl chloride polymerization workers illustrates that even small epidemiological studies (in essence, clinical observations of a few people) may have sufficient power to detect a very large relative risk (Creech and Johnson 1974; Heath et al., 19751. In that instance, four cases of hepatic angiosarcoma were epidemiologically sufficient, since the rate of that rare
Data on Humans 237 tumor was several hundred times greater than expected. With this excep- tional relative risk, these studies had powers close to 1.0. The biological importance of the finding was greatly enhanced by the earlier independent demonstration that vinyl chloride was carcinogenic in laboratory animals (Maltoni and Lefemine, 19741. In contrast is the study by Ott et al. (1980), who examined risk of cancer in workers exposed to ethylene dibromide (see Chapter 91. Here, failure to detect an increased risk may be related to the limited sample size (161 workers studied, 5.8 cancer cases expected) and thus, limited power. The study had 0.29 power to detect a relative risk of 1.5that is, a 50% increase in risk (see Table 7-2 and Apfeldorf and Infante, 19811. The concept of power is especially important in interpreting so-called negative studies. When a specific health effect has been examined in several different epidemiological studies, as they were for studies of health effects from exposure to arsenic in drinking water, results may often seem to be in conflict, some studies yielding positive results (i.e., statistically significant increases in observed risk) and others showing no association or negative results (i.e., nonsignificant variations in levels of observed risk). In this situation, it is important not only to assess the studies for features that might compromise validity or distort analyses but also to examine the power of negative studies to determine if their sample sizes were too small to detect increased risk at the levels medically or biolog- ically worth discovering or observed in positive studies. If power is in- sufficient, such negative studies cannot be said either to contradict or to support the conclusion that increased risk does in fact exist. Confidence limits based on the excess risk estimates of these small studies can some- times be regarded as upper limits unlikely to be exceeded in a study of sufficient size and quality. Conversely, of course, studies yielding statis- tically significant positive results may be found to be false positive. As examples, short-term studies of skin cancer patterns in small populations exposed to elevated arsenic levels are not likely to have sufficient power to detect an effect (i.e., Harrington et al., 1978~. In contrast, the National Cancer Institute is currently analyzing the results of a large case-comparison study on 3,000 cases and 6,000 controls for bladder cancer (Cantor et al., 19851. This study includes data on drink- ing water. However, if the relative effect of drinking water is rather small, then even this large study will have a low power of detecting any such effect. Study-ReIated Limitations Additional limitations associated with epidemiological studies derive from the intrinsic characteristics of diseases under study. One example is
238 DRINKING WATER AND H"LTH the difficulty in associating exposure with effect because of the long and variable latency periods for manifestation of chronic illnesses, such as cancers (which are of particular concern because of their potential rela- tionships to low-dose toxic exposures). Moreover, clinical features of chronic illnesses rarely provide any clues to specific etiology. Cases of cancer linked to ionizing radiation, for instance, are usually clinically indistinguishable from those linked to chemical exposures. For some dis- eases, many agents may lead to the same illness, thus precluding direct linkage of particular cases to particular toxic agents. It is also difficult to identify suitable control populations. In contrast to laboratory experiments, variables such as nutrition, medication use, or cigarette smoking cannot be easily controlled or matched in populations under epidemiological study. Accurate information on such variables must be collected if they could be potentially confounding elements (Davis et al., 1983~. For example, it is inappropriate to try to estimate the risk of lung cancer in a group of smokers in comparison to a general population that includes a substantial number of smokers and former smokers. In addition, comparisons confin ~ to people aged 20 to 65 exclude many people with longer exposures to some toxic chemicals. In view of the fact that half of all cancer cases arise in people over 65, such inappropriate comparisons dilute the real effect of exposure. McMichael (1976) sug- gested that occupational risk may be understated by as much as a factor of two when using the general population as a control group. For example, it is frequently asserted that the favorable mortality experience of some occupational groups is due to a healthy worker effect. That is, those at work have a lower prevalence of chronic diseases or lower mortality than does the general population, which includes persons not at work due to incapacitating illness. Epidemiological studies are susceptible to certain biases, which must be minimized. Of particular importance to case-comparison studies is response bias or recall bias. Persons with an illness or with known ex- posures tend to remember events associated with those illnesses or ex- posures better than do persons not ill or not aware of the exposure. However large or otherwise well designed an epidemiological study may be, its capacity to provide sound answers to questions of biological risk depends on the accuracy of the data collected. The report of a study must state the extent to which determinations of exposure levels and biological outcomes may be imprecise, subjective, unverified, or unverifiable. For some ex- posures, the responses may be subjective symptoms that are often rather hard to establish and measure effectively. The greater the uncertainty of exposure or outcome data, the less likely it is that a true association or effect will be correctly identified.
Data on Humans 239 Inference and Quantification Epidemiological data usually lack the incisiveness of laboratory data, making it difficult to assert with assurance that an epidemiological study has furnished proof of some cause-and-effect relationship. As noted, there are often circumstances surrounding human exposuresometimes exten- sive exposures that do not permit epidemiological investigations. How- ever, the lack of incisiveness can be offset under two sets of circumstances. First, when the effect is very large (as in the relationship between cigarette smoking and lung cancer), the effect can be clearly discerned above the background of confounding elements. Second, if one epidemiological study is possible, often more than one can be conducted. The observed cause-and-effect relationship gains credibility when similar results are obtained in several studies conducted under somewhat different circum- stances and affected by different confounding variables of greater or lesser importance. It is much easier to determine whether or not an effect is a consequence of some exposure than it is to quantify that response, that is, to answer the question: How much response for how much exposure? Many studies are able to distinguish between exposed and unexposed persons, but are unable to say how much exposure there was. This is largely due to the fact that industrial hygiene and environmental monitoring data were rarely collected in the past. Moreover, epidemiological studies necessarily in- clude people who have had multiple toxic exposures. But scientists' un- derstanding of the ways multiple toxic exposures affect health is very limited. This is reflected by the fact that no existing model of carcino- genesis is yet able to deal adequately with the possible interactions or synergism (or antagonism) associated with multiple exposures (see Chapter 51. Molecular or biochemical epidemiology, combining the precision of laboratory measurements of dose and early response with standard epi- demiological methods, holds considerable promise in the elucidation of some of these problems (Hattie, in press; Perera and Weinstein, 1982~. The molecular assays may elucidate which of the toxicants attain biolog- ically effective levels, how the toxicants interact in the tissues, and what the levels of toxicants are in various organs. Even for extensively studied materials, such as asbestos, exposure es- timates frequently have been developed not from direct measurements of concentrations at time of actual exposure but rather from ingenious re- constructions of earlier working conditions (taking into account ventilation and other factors) in select study populations (NRC, 1984~. For inadvertent or ambient exposures of the general population, estimates are even more difficult to obtain. People are often quite unaware of the exposures, if
240 DRINKING WATER AND H"LTH any, they have had, and the range of exposures in a population may be great. The exposure history for individuals whether intermittent or con- tinuous, whether at relatively constant levels or with peaks and valleys is rarely known. Yet, differences in exposure patterns may affect biological response. For example, the age at first or only exposure may also be of considerable importance. Young women exposed to ionizing radiation at Hiroshima were at higher risk of developing breast cancer than women who were older at the time of exposure (Land et al., 19801. Nature of Populations Studied Many useful epidemiological studies are conducted on industrial pop- ulations for which measurement of exposure can sometimes be obtained. Workers, however, are generally relatively young, healthy males whose exposures are limited to working hours and are governed by workplace conditions. Workers are sometimes able to remove themselves (i.e., find other jobs) if they find exposure disturbing usually leaving the less affected, less disturbed, or less sensitive workers to receive the greater exposure. Therefore, when results of occupational exposure studies are used for risk assessment, they may not be entirely appropriate for esti- mating health risks in the general population exposed under different conditions. They may be especially misrepresentative for pregnant women, children, and elderly persons as well as for people with certain illnesses or biological sensitivities that may lead them to be excluded from industrial populations. As for ionizing radiation, the International Commission on Radiological Protection divides by 10 the occupational exposure levels for a 40-hour, 5-day work week in order to produce estimated safe levels for the general population (ICRP, 19771. This approach has not been uniformly applied. Mechanisms of Biological Action Knowledge of likely or possible biological mechanisms is of consid- erable value in evaluating the findings of specific studies. Unfortunately, these mechanisms are rarely understood in full. The model most often used in risk assessment computations for carcinogenesis, and which is used in this report, is the multistage model of Arrnitage and Doll (19541. This model, discussed in greater detail in Chapter 5, derives from the epidemiological observation that cancer incidence increases with age, im- plying that the generalized cancer process must be of a multistage nature that leads to rapidly increasing age-specific incidence and mortality curves. This epidemiological observation has a parallel in initiator-promoter lab-
Data on Humans 241 oratory studies and in the recent findings that two or more mutations are involved in some carcinogenic transformations (Slaga et al., 1980, 19821. Knowledge of biological mechanisms has not always been essential for taking public health action. On the basis of an epidemiological association, for example, thalidomide was removed from European markets before its mode of action was completely understood (Sjostrom and Nilsson, 19721. When greater biological information exists, however, the precision and depth of epidemiological observations can obviously be much enhanced and can lead to more specific public health measures. RISK ASSESSMENT Relative Risk Epidemiological studies are capable of generating several types of data that might be used to assess risk and to develop public health policy. When there is reason to believe that the exposure-response relationship multiplies the background rates, the increase should be reported as relative risk, as with most studies of cancer. The standard mortality ratio (SMR) is a relative risk measure in which the observed number of deaths is compared with the number of deaths expected for a specified population with a comparable age distribution. The SMR is computed as 100 times the number of observed events divided by the number of expected events. Thus, an SMR of 100 implies no difference in risk from that expected in another comparable population. The expected number is usually derived by applying the rates for age-, sex-, and race-specific groups to the num- bers of persons in the exposed population in the comparable age-, sex-, and race-specific groups, and then summing the products so derived. For example, the relative risk of death from lung cancer for a white male who smokes a pack of cigarettes a day is about 10 times greater than that of a corresponding nonsmoker. The SMR in this case would be 1,000. AHributable Risk Attributable risk can be derived from relative risk when there is infor- mation on the proportion of a population that has been exposed. It measures the maximal proportion of disease incidence in a population that can be attributed to a particular factor, such as cigarette smoking. Attributable risk is useful in suggesting how public health resources should be dis- tributed, the assumption being that the higher the attributable risk, the greater will be the health rewards of controlling it. Originally developed by Levin as a tool to aid decision-makers in un- derstanding epidemiological data (Levin, 1953; Levin and Bertell, 1978),
242 DRINKING WATER AND H"LTH this concept has gained wide acceptance and is now used extensively. The recent cancer prevention initiative undertaken by the Secretary of Health and Human Services, which advocates specific dietary and life-style prac- tices, is based on assessment of the attributable risk of cancer estimated to be due to certain nutritional practices and cigarette smoking. For in- stance, up to 90% of the lung cancer cases among men in industrialized countries is now attributed to smoking (Davis et al., 1981; Doll and Peto, 1981). Additivity In some circumstances, the biological process resulting from an ex- posure may not appear to lead to a multiplication of background rates. At times, the risk may be simply a direct addition to, and completely inde- pendent of, the background rates. Some mathematical models are used in laboratory studies of carcinogenesis to estimate excess risks above back- ground. The appropriateness of these models to humans will depend upon the specific agent-response interaction. For low-level exposures, resulting in small increments in risk, the dis- tinction between additive and multiplicative responses (i.e., between ad- ditive and relative risks) is not important and probably cannot be observed. Thus, if a relative risk for some demographic group is 1.2 and some exposure produces a further 20% increase in risk (i.e., another 1.2), under multiplicative conditions the risk would be 1.2 x 1.2, or 1.44. If the second risk were additive, the total risk would be 1.2 + 0.2, or 1.4, which is not significantly distinguishable from 1.44. Therefore, for low- level exposures, the increased risks due to additive and multiplicative processes are comparable. Exposures to more than one substance may result in a synergistic re- sponse. For example, cigarette smokers who experience a lifetime of exposure to asbestos in the ambient environment have more than 10 times as much risk as asbestos-exposed nonsmokers of contracting lung cancer (NRC, 1984, pp. 21 1-222~. Application to Drinking Water Studies On occasion, results of epidemiological research are useful in the de- velopment of drinking water standards. For example, epidemiological studies indicated that adverse health effects developed in humans exposed to foods or water supplies contaminated with lead, cadmium, and mercury (Calabrese, 19831. Although drinking water constitutes a small percentage of total daily exposure to many substances, it may play a determinative role nonetheless.
Data on Humans 243 TABLE 7-3 Chapter 9 Epidemiological Studies of Compounds Reviewed in Study Findings Dibromochloropropane (DBCP) Azoospermia or severe oligospermia in DBCP production workers. References Sperm count reductions in workers exposed during agricultural application of DBCP. Sperm count reduction, elevated serum follicle- stimulating hormone (FSH) and luteinizing hormone (LH) levels, and reduced or absent spermatogenic cells in men with history of industrial exposure to DBCP. Reduced sperm count, elevated FSH and LH, and decreased testicular volume in a population of workers exposed to DBCP. Increased frequency of Y-chromosome nondisjunction in the sperm of DBCP-exposed workers. Increased number of sperm containing two Y-chromosomes in DBCP- exposed workers. Increased frequency of spontaneous abortions in wives of DBCP-exposed agricultural workers in Israel. No or little recovery in sperm production in DBCP-exposed workers who ceased spermatogenesis as a result of exposure. Complete recovery of sperm production in DBCP-exposed workers with reduced spermatogenesis when exposure was removed. No change in sperm count in agricultural workers exposed to DBCP concentrations up to 1.8 ppm. NIOSH concluded that a 1-ppm exposure level has no observable effects on male fertility. Egnatz et al., 1980 Lipshultz et al., 1980 Milby and Whorton, 1980 Potashnik et al., 1979 Whorton et al., 1979 Glass et al., 1979 Sandifer et al., 1979 Biava et al., 1978 Potashnik et al., 1979 Egnatz et al., 1980 Kapp et al., 1979 Kapp et al., 1979 Kharrazi et al., 1980 Lantz et al., 1981 Whorton and Milby, 1980 Whorton et al., 1979 NIOSH, 1978
244 DRINKING WATER AND H"LTH TABLE 7-3 (Continued Study Findings Ethylene dibromide (EDB) No significantly greater mortality in workers employed at EDB synthesis plants between 1940 and 1976. Standardized birth ratios not significantly reduced among EDB- exposed workers and their wives, regardless of amount of exposure. Pentachlorophenol (PCP) Sweating, weight loss, and gastrointestinal disorders in factory workers exposed to PCP at a Winnipeg plant. Tachycardia, respiratory distress, and liver aberrations in neonates exposed to PCP in diapers and linen. Two of the 20 exposed died. Elevated SOOT, SGPT, and LDH, and low-grade infections or inflammations of the skin, eye, and upper respiratory tract in chronically PCP-exposed workers In Hawaii. No significant difference in chromosome aberrations between PCP- exposed workers and controls. Plasma protein levels elevated in workers chronically exposed to PCP in Hawaii. References Ott et al., 1980 Wong et al., 1979 Higher immunoglobulin levels in PCP- exposed workers in the Federal Republic of Germany than in controls. TrichloJfon Excess of short-lived chromosome breaks and exchange along with a significant increase in stable chromosome alterations in five people exposed to trichlorfon. Bergner et al., 1965 Robson et al., 1969 Klemmer et al., 1980 Wyllie et al., 1975 Takahashi et al., 1976 Zober et al., 1981 van Bao et al., 1974 This is illustrated by the studies of Calabrese and Tuthill (1977), who found that blood pressure measurements differed significantly in matched groups of students living in two communities with different levels of sodium in the drinking water. The investigators also conducted an ex-
Data on Humans 245 penmental, 3-month study in which bottled, low-sodium water was sup- plied to one of the high-sodium community subgroups of students with elevated blood pressure. The study demonstrated reductions in blood pres- sure, but only in girls (Calabrese and Tuthill, 19801. Since people who previously consumed large amounts of salt are now towering their intake, additional monitoring of their blood pressure levels and sodium intake, including that from drinking water, could provide useful information re- garding the effects of elevated levels of sodium in drinking water on human health. Table 7-3 summarizes the epidemiological studies on the pollutants assessed in Chapter 9. The committee found no studies on drinking water exposure to the contaminants listed in the table, presumably because of the many limitations to epidemiological investigations discussed above. The committee was therefore forced to extrapolate data from studies of occupational, airborne, and skin exposures to estimate risks for exposure by ingestion of water. Some clinical reports did include exposure to these compounds through ingestion. Several of them related to accidental poi- sonings or suicides. Such reports can be especially valuable in suggesting important metabolic pathways in humans. As discussed in Chapter 6, knowledge of toxicokinetics derived from studies in humans may ulti- mately transform the process of risk assessment. In the meantime, epi- demiological and clinical studies will continue to play a generally supportive role in the evaluation of risks to humans from drinking water contaminants. REFERENCES Apfeldorf, R., and P. F. Infante. 1981. Review of epidemiologic study results of vinyl chloride-related compounds. Environ. Health Perspect. 41:221-226. Armitage, P., and R. Doll. 1954. The age distribution of cancer and a multi-stage theory of carcinogenesis. Br. J. Cancer 7:1-12. Arnold, F. A., H. T. Dean, P. Jay, and J. W. Knutson. 1953. Effect of fluoridated public water supplies on dental caries prevalence. Public Health Rep. 68:141-148. Bergner, H., P. Constantinidis, and J. H. Martin. 1965. Industrial pentachlorophenol poisoning in Winnipeg. Can. Med. Assoc. J. 92:448-451. Bernstein, D. S., N. Sadowsky, D. M. Hegsted, C. D. Guri, and F. J. Stare. 1966. Prevalence of osteoporosis in high- and low-fluoride areas in North Dakota. J. Am. Med. Assoc. 198:499-504. Biava, C. G., E. A. Smuckler, and D. Whorton. 1978. The testicular morphology of individuals exposed to dibromochloropropane. Exp. Mol. Pathol. 29:448-458. Black, R. E. 1980. Cholera. Pp. 231-236 in J.M. Last, ed. Maxcy-Rosenau Public Health and Preventive Medicine, 11th ed. Appleton-Century-Crofts, New York. Borgofio, J. M., P. Vicent, H. Venturino, and A. Infante. 1977. Arsenic in the drinking water of the city of Antofagasta: Epidemiological and clinical study before and after the installation of a treatment plant. Environ. Health Perspect. 19:103-105.
246 DRINKING WATER AND H"LTH Calabrese, E. J. 1983. Role of epidemiologic studies in deriving drinking water standards for metals. Environ. Health Perspect. 52:99-106. Calabrese, E. J., and R. W. Tuthill. 1977. Elevated blood pressure and high sodium levels in the public drinking water: Preliminary results of a study of high school students. Arch. Environ. Health 32:200-202. Calabrese, E. J., and R. W. Tuthill. 1980. The influence of elevated levels of sodium in drinking water on elementary and high school students in Massachusetts. J. Environ. Pathol. Toxicol. 4(2,3):151-165. Cannon, S. B., J. M. Veazey, Jr., R. S. Jackson, V. W. Burse, C. Hayes, W. E. Straub, P. J. Landrigan, and J. A. Liddle. 1978. Epidemic kepone poisoning in chemical workers. Am. J. Epidemiol. 107:529-537. Cantor, K. P., R. Hoover, P. Hartge, T. J. Mason, D. T. Silverman, and L. I. Levin. 1985. Drinking water source and risk of bladder cancer: A case-control study. Pp. 143- 150 in R. L. Jolley, R. J. Bull, W. P. Davis, S. Katz, Jr., M. H. Roberts, and V. A. Jacobs, eds. Water Chlorination: Chemistry, Environmental Impact and Health Effects. Vol. 5. Lewis Publishers, Chelsea, Mich. Carlos, J. P., A. M. Gittelsohn, and W. Haddon, Jr. 1962. Caries in deciduous teeth in relation to maternal ingestion of fluoride. Public Health Rep. 77:658-660. Cohen, J. 1977. Statistical Power Analysis for the Behavioral Sciences. Academic Press, New York. 474 pp. Creech, J. L., Jr., and M. N. Johnson. 1974. Angiosarcoma of liver in the manufacture of polyvinyl chloride. J. Occup. Med. 16:150-151. Davis, D. L., and B. Mandula. 1985. Airborne asbestos and public health. Annul Rev. Public Health 6:195-221. Davis, D. L., K. Bridbord, and M. Schneiderman. 1981. Estimating cancer causes: Prob- lems in methodology, production, and trends. Pp. 285-316 in R. Peto and M. Schnei- derman, eds. Quantification of Occupational Cancer. Banbury Report 9. Cold Spring Harbor Laboratory, Cold Spring Harbor, N.Y. Davis, D. L., K. Bridbord, and M. Schneiderman. 1983. Cancer prevention: Assessing causes, exposures, and recent trends in mortality for U.S. males, 1968-1978. Int. J. Health Serv. 13:337-372. Davis, D. L., B. Mandula,~and J. Van Ryzin. 1985. Assessing the power and quality of epidemiologic studies of asbestos-exposed populations. Tox. Ind. Health 1(4):93-110. Dean, H. T., F. A. Arnold, Jr., P. Jay, and J. W. Knutson. 1950. Studies on mass control of dental caries through fluoridation of the public water supply. Public Health Rep. 65: 1403-1408. DeRouen, T. A., and J. E. Diem. 1975. The New Orleans drinking water controversy. A statistical perspective. Am. J. Public Health 65:1060-1062. Doll, R., and R. Peto. 1981. The causes of cancer: Quantitative estimates of avoidable risks of cancer in the United States today. J. Natl. Cancer Inst. 66:1191-1308. Dorsch, M. M., R. K. R. Scragg, A. J. McMichael, P. A. Baghdrst, and K. F. Dyer. 1984. Congenital malformations and maternal drinking water supply in rural South Australia: A case-control study. Am. J. Epidemiol. 119:473-486. Egnatz, D. G., M. G. Ott, J. C. Townsend, R. D. Olson, and D. B. Johns. 1980. DBCP and testicular effects in chemical workers: An epidemiological survey in Midland, Mich- igan. J. Occup. Med. 22:727-732. Geever, E. F., N. C. Leone, P. Geiser, and J. Lieberman. 1958a. P~ologic studies in man after prolonged ingestion of fluoride in drinking water. I. Necropsy findings in a community with a water level of 2.5 ppm. J. Am. Dental Assoc. 56:499-507.
Data on Humans 247 Geever, E. F., N. C. Leone, P. Geiser, and J. E. Lieberman. 1958b. Pathologic studies in man after ingestion of fluoride in drinking water. II. Findings in bones in communities with water levels from 1.0 to 4.0 ppm fluoride. Public Health Rep. 73:721-731. Glass, R. I., R. N. Lyness, D. C. Mengle, K. E. Powell, and E. Kahn. 1979. Sperm count depression in pesticide applicators exposed to dibromochloropropane. Am. J. Epidemiol. 109:346-351. Harrington, J. M., J. P. Middaugh, D. L. Morse, and J. Housworth. 1978. A survey of a population exposed to high concentrations of arsenic in well water in Fairbanks, Alaska. Am. J. Epidemiol. 108:377-385. Hattis, D. B. In press. The promise of molecular epidemiology for quantitative risk as- sessment. Risk Anal. Heath, C. W., Jr., H. Falk, and J. L. Creech, Jr. 1975. Characteristics of cases of angiosarcoma of the liver among vinyl chloride workers in the United States. Ann. N.Y. Acad. Sci. 246:231-236. Hill, A. B. 1953. Observation and experiment. N. Engl. J. Med. 248:995-1001. ICRP (International Commission on Radiological Protection). 1977. Radiation Protection. Recommendations of the International Commission on Radiological Protection. ICRP Publication No. 26. Ann. ICRP 1(3):1-52. Kapp, R. W., Jr., D. J. Picciano, and C. B. Jacobson. 1979. Y-chromosomal nondisjunction in dibromochloropropane-exposed workmen. Mutat. Res. 64:47-51. Karstadt, M., and R. Bobal. 1982. Availability of epidemiologic data on humans exposed to animal carcinogens. II. Chemical uses and production volume. Teratogen. Carcinogen. Mutagen. 2: 151-167. Karstadt, M., R. Bobal, and I. J. Selikoff. 1981. A survey of availability of epidemiologic data on humans exposed to animal carcinogens. Pp. 223-245 in R. Peto and M. Schnei- derman, eds. Quantification of Occupational Cancer. Banbury Report 9. Cold Spring Harbor Laboratory, Cold Spring Harbor, N.Y. Kharrazi, M., G. Potashnik, and J. R. Goldsmith. 1980. Reproductive effects of dibromo- chloropropane. Isr. J. Med. Sci. 16:403-406. Klemmer, H. W., L. Wong, M. M. Sato, E. L. Reichert, R. J. Korsak, and M. N. Rashad. 1980. Clinical findings in workers exposed to pentachlorophenol. Arch. Environ. Con- tam. Toxicol. 9:715-725. Land, C. E., J. D. Boice, Jr., R. E. Shore, J. E. Norman, and M. Tokunaga. 1980. Breast cancer risk from low-dose exposure to ionizing radiation: Results of parallel analysis of three exposed populations of women. J. Natl. Cancer Inst. 65:353-376. Lantz, G. D., G. R. Cunningham, C. Huckins, and L. I. Lipshultz. 1981. Recovery from severe oligospermia after exposure to dibromochloro-propane. Fertil. Steril. 35:46-53. Leone, N. C., M. B. Shimkin, F. A. Arnold, C. A. Stevenson, E. R. Zimmerman, P. A. Geiser, and J. Lieberman. 1954. Medical aspects of excessive fluoride in a water supply. A 10 year study. Public Health Rep. 69:925-936. Leone, N. C., C. A. Stevenson, T. F. Hilbish, and M. C. Sosman. 1955. A roentgenologic study of a human population exposed to high-fluoride domestic water. A ten-year study. Am. J. Roentgenol. Radium Ther. Nucl. Med. 74:874-885. Leone, N. C., C. A. Stevenson, B. Besse, L. E. Hawes, T. R. Dawber, and W. J. Claffey. 1960. The effects of the absorption of fluoride. II. A radiological investigation of 546 human residents in an area in which the drinking water contained only a minute trace of fluoride. Am. Med. Assoc. Arch. Ind. Health 21:326-327. Letz, G. A., S. M. Pond, J. D. Osterloh, R. L. Wade, and C. E. Becker. 1984. Two fatalities after acute occupational exposure to ethylene dibromide. J. Am. Med. Assoc. 252:2428-2431.
248 DRINKING WATER AND H"LTH Levin, M. L. 1953. The occurrence of lung cancer in man. Acta Unio Int. Contra Cancrum 9:531-541. Levin, M. L., and R. Bertell. 1978. Simple estimation of population attributable risk from case-control studies. Am. J. Epidemiol. 108:78-79. Lipshultz, L. I., C. E. Ross, D. Whorton, T. Milby, R. Smith, and R. E. Joyner. 1980. Dibromochloropropane and its effect on testicular function in man. J. Urol. 124:4(i4468. Maltoni, C., and G. Lefemine. 1974. Carcinogenicity bioassays of vinyl chloride. I. Re- search plan and early results. Environ. Res. 7:387-405. Markovitz, A., and W. H. Crosby. 1984. Chemical carcinogenesis. A soil fumigant, 1,3- dichloropropene, as possible cause of hematologic malignancies. Arch. Intern. Med. 144: 1409-1411. Marsh, G. M. 1983. Mortality among workers from a plastics producing plant: A matched case-control study nested in a retrospective cohort study. J. Occup. Med. 25:219-230. McMichael, A. J. 1976. Standardized mortality ratios and the "healthy worker effect": Scratching beneath the surface. J. Occup. Med. 18: 165-168. Milby, T. H., and D. Whorton. 1980. Epidemiological assessment of occupationally related, chemically induced sperm count suppression. J. Occup. Med. 22:77-82. Morton, W., G. Starr, D. Pohl, J. Stoner, S. Wagner, and P. Weswig. 1976. Skin cancer and water arsenic in Lane County, Oregon. Cancer 37:2523-2532. Murphy, R., and C. Harvey. 1985. Residues and metabolites of selected persistent halo- genated hydrocarbons in blood specimens from a general population survey. Environ. Health Perspect. 60:115- 120. NIOSH (National Institute for Occupational Safety and Health). 1978. A Recommended Standard for Occupational Exposure to Dibromochloropropane. DHEW (NIOSH) Pub- lication No. 78-115. U.S. Department of Health, Education, and Welfare, National Institute for Occupational Safety and Health, Cincinnati. 14 pp. NRC (National Research Council). 1980. Drinking Water and Health. Vol. 3. National Academy Press, Washington, D.C. 415 pp. NRC (National Research Council). 1984. Asbestiform fibers: Nonoccupational Health Risks. National Academy Press, Washington, D.C. 334 pp. Ott, M. G., H. C. Scharnweber, and R. R. Langner. 1980. Mortality experience of 161 employees exposed to ethylene dibromide in two production units. Br. J. Ind. Med. 37: 163- 168. Page, T., R. H. Harris, and S. S. Epstein. 1976. Drinking water and cancer mortality in Louisiana. Science 193:55-57. Perera, F. P., and I. B. Weinstein. 1982. Molecular epidemiology and carcinogen-DNA adduct detection: New approaches to studies of human cancer causation. J. Chronic Dis. 35:581-600. Potashnik, G., I. Yanai-Inbar, M. I. Sacks, and R. Israeli. 1979. Effect of dibromochlo- ropropane on human testicular function. Isr. J. Med. Sci. 15:438-442. Potashnik, G., J. Goldsmith, and V. Insler. 1984. Dibromochloropropane-induced reduction of the sex-ratio in man. Andrologia 16:213-218. Roberts, D. R. 1983. Summary Report: NIOSH-EPA Interagency Agreement for the As- sessment of Human Health Effects from Exposure to Di-2-(ethylhexyl) phthalate. NIOSH Division of Surveillance, Hazard Evaluations and Field Studies, Cincinnati. 15 pp. Robson, A. M., J. M. Kissane, N. H. Elvick, and L. Pundavela. 1969. Pentachlorophenol poisoning in a nursery for newborn infants. I. Clinical features and treatment. J. Pediatr. 75:309-316. Sandifer, S. H., R. T. Wilkins, C. B. Loadholt, L. G. Lane, and J. C. Eldridge. 1979. Spermatogenesis in agricultum1 workers exposed to dibromochloropropane (DBCP). Bull. Environ. Contam. Toxicol. 23:703-710.
Data on Humans 249 Sjostrom, H., and R. Nilsson. 1972. Thalidomide and the Power of the Drug Companies. Penguin, Baltimore. 280 pp. Slaga, T. J., S. M. Fischer, K. Nelson, and G. L. Gleason. 1980. Studies on the mechanism of skin tumor promotion: Evidence for several stages in promotion. Proc. Natl. Acad. Sci. USA 77:3659-3663. Slaga, T. J., S. M. Fischer, C. E. Weeks, A. J. P. Klein-Szanto, and J. Reiners. 1982. Studies on the mechanisms involved in multistage carcinogenesis in mouse skin. J. Cell. Biochem. 18:99-119. Snow, J. 1855. On the Mode of Communication of Cholera, 2nd ed. John Churchill, London. (Reproduced in Snow on Cholera. 1936. The Commonwealth Fund, New York.) Takahashi, W., E. R. Reichert, G. C. Fung, and Y. Hokama. 1976. Acute phase proteins and pesticides exposure. Life Sci. 19: 1645-1652. Tillotson, J. L., and S. B. Hulley, eds. 1985. Materials and Methods for a Cardiovascular Disease Risk Factor Reduction Program. A Resource from the Multiple Risk Factor Intervention Trial (MRFIT). DHHS Publication No. (NIH) 85-1267. National Heart, Lung, and Blood Institute, National Institutes of Health, U.S. Department of Health and Human Services, Bethesda, Md. [265 pp.] (Available from the National Technical In- formation Service, Springfield, Va., as Publication No. PB85-215630.) Tseng, W.-P. 1977. Effects and dose-response relationships of skin cancer and blackfoot disease with arsenic. Environ. Health Perspect. 19:109-119. Tseng, W.-P., H. M. Chu, S. W. How, J. M. Fang, C. S. Lin, and S. H. Yeh. 1968. Prevalence of skin cancer in an endemic area of chronic arsenism in Taiwan. J. Natl. Cancer Inst. 40:453-463. van Bao, T., I. Szabo, P. Ruzicska, and A. Czeizel. 1974. Chromosom~ aberrations in patients suffering acute organic phosphate insecticide intoxication. Humangenetik 24:33-57. Wagner, J. C., C. A. Sleggs, and P. Marchand. 1960. Diffuse pleural mesothelioma and asbestos exposure in North Western Cape Province. Br. J. Ind. Med. 17:260-271. Whorton, D., R. M. Krauss, S. Marshall, and T. H. Milby. 1977. Infertility in male pesticide workers. Lancet 2:1259-1261. Whorton, D., T. H. Milby, R. M. Krauss, and H. A. Stubbs. 1979. Testicular function in DBCP exposed pesticide workers. J. Occup. Med. 21:161-166. Whorton, M. D., and T. H. Milby. 1980. Recovery of testicular function among DBCP workers. J. Occup. Med. 22:177-179. Williamson, S. J. 1981. Epidemiological studies on cancer and organic compounds in U.S. drinking waters. Sci. Total Environ. 18:187-203. Wong, O., H. M. D. Utidjian, and V. S. Karten. 1979. Retrospective evaluation of reproductive performance of workers exposed to ethylene dibromide (EDB). J. Occup. Med. 21 :98-102. Wyllie, J. A., J. Gabica, W. W. Benson, and J. Yoder. 1975. Exposure and contamination of the air and employees of a pentachlorophenol plant, Idaho 1972. Pestic. Monit. J. 9: 150-153. Zober, A., K. H. Schaller, K. Gossler, and H. J. Krekeler. 1981. Pentachlorophenol and liver-function: A pilot-study on occupationally exposed collectives. Int. Arch. Occup. Environ. Health 48:347-356. (In German; English summary)
