8
Latency and Cancer Risk
One of the topics of special interest to the Department of Veterans Affairs (DVA) is the potential effect of herbicide exposure on cancer latency. The term "latency" is used in a variety of ways to denote the effect of the timing of exposure on the subsequent risk of disease. The importance of latency effects as well as other time-related factors, such as age at exposure, in determining cancer risk has long been recognized (Armenian, 1987). The following are some important practical questions at the heart of the investigation of these time-related factors: (1) How long does it take after exposure to detect an increase in disease risk? (2) How long do the effects of exposure last? (3) How does the effect of exposure vary with the age at which it was received? (4) Does a given carcinogen act at an early or a late stage of the carcinogenic process?
Often, because of either poor exposure assessment or the desire to report a simple summary measure of association, measures of exposure such as ever/ never exposed or cumulative exposure are used to summarize exposure histories. Although such measures can be useful for detecting whether or not there is an association between exposure and disease, it is well known that the timing of exposure often plays an important role in determining when and by how much the eventual disease risk is increased (or decreased) by the exposure.
In response to the DVA's request to explore the latency issues related to Agent Orange, in this chapter the committee (1) presents a methodology to address the four questions listed above concerning the timing of herbicide exposure and the risk of cancer; (2) reviews the literature on herbicide exposure and cancers classified in the sufficient and limited/suggestive categories for results describing how the relative risks vary with time since exposure began or ended;
and (3) discusses the relevance of these data for cancer risk among Vietnam veterans.
ANALYSIS OF LATENCY IN EPIDEMIOLOGIC STUDIES
To discuss latency issues, we need to establish what is meant by the ''effect of exposure over time." First, for purposes of epidemiologic research and quantification, we are interested in the rate of disease among exposed individuals compared to the rate that would be expected if the subjects had not been exposed, which is discussed more fully below. Thus, we are interested in the relative or excess rate of disease as the measure of comparison. Because diseases such as cancer may take a long time to develop (i.e., years or even decades) an analysis of the effects of exposure must consider the "latency period," or time between the exposure and the measurement of disease. The effect of any exposure on a population, whether measured as relative or excess rates, may change with "time since exposure." Typically, after exposure to a carcinogen, no excess cancer rate will be observed for a time. Then there will be a rise in the excess until it reaches a peak, at which point it may fall back down. For exposures of short duration, time since exposure is in many cases easy to define. This situation holds for environmental exposures from industrial accidents, particularly if the exposure involves chemicals that are not retained in the body tissues. If the exposure occurred over a long period of time (a protracted exposure), as with production workers and pesticide applicators, the time since exposure is more difficult to quantify, since there were many exposure times. It is important to note that 2,3,7,8-tetrachlorodibenzo-p-dioxin (2,3,7,8-TCDD, TCDD, or dioxin) and other chlorinated herbicides are retained in some body tissues for a long time (e.g., decades), so that even after external exposure ceases, internal exposure continues. Thus, even a brief exposure such as occurred in Seveso, Italy, can involve protracted exposure of many organs of the body. Conceptually, we think of the effect of exposure at a particular time in the past as the resulting change in risk today that is ascribable to that exposure. Although this may be overly simplified, the effect of an entire exposure history can usefully be thought of as the sum of the effects from exposure at each time point in the past.
To adequately study the effects of protracted exposure, detailed exposure histories for each study subject, including the dates at which the individual was exposed and, ideally, the level of exposure, are needed. Appropriate statistical methods have been developed for investigating the effects of exposure accrued as a function of time since exposure (Thomas, 1983, 1988; Breslow and Day, 1987), but these have not been used to analyze of any of the cancer studies reviewed here.
In general, the ability to investigate the issue of timing of exposure in a given data set will depend on the quality of the exposure measure, the quality of the timing of exposure information, the number of people developing the disease,
and the variation of exposure over time within the study group. These aspects of study quality are of course important in evaluating any epidemiologic study. Special problems arise in the evaluation of time-related factors (Enterline and Henderson, 1973; Peto, 1985; Thomas, 1987).
Need to Control for the Effects of Aging
Progression along the scale of time since exposure is paralleled by increasing age. Since the rate of most cancers increases dramatically with age, for any given study group the expected number of cancers per unit of time will increase with time since exposure, simply because the study group is aging. Thus, an examination of the absolute number of cases versus time since exposure would be misleading. We would expect to see an increase even if there were no change related to the exposure. Thus, it is standard epidemiologic practice to "normalize" the comparison by reference to an appropriate unexposed group. "Appropriate" in this context means a group with the same age structure and otherwise having similar risk of disease as the study group; thus we are measuring the effects of exposure over and above those changes in cancer rates that would be expected if the study group were simply aging without having been exposed.
Correlations Among Various Time Factors
It is important to keep in mind that duration of exposure, time since exposure, age of exposure, and exposure level itself may well be correlated with each other, so that an observed pattern for one of these factors could actually be due to correlation with one or more of the others. Stratifying the analysis or modeling both effects is commonly needed to disentangle the confounding of time-related variables. Realistically, however, most data sets do not have sufficient data to explore these issues, and for the most part, these considerations merely serve as a cautionary note in the interpretation of results. The following hypothetical situations help to illustrate how various time-related factors are intertwined. For illustration, we the exposure of interest is referred to as "the agent."
-
Two people are born in 1945. Both are exposed continuously to the agent for 10 years. One person begins exposure at age 20 (continuing to age 30); the other begins at age 30 (continuing to age 40). In 1995, at age 50, the first person has 30 years since first exposure, whereas the second has only 20 years. Thus, the age at first exposure and the time since first exposure, for a given duration of exposure, are linked.
-
One person begins exposure to the agent in 1970; a second starts in 1975. They are the same age at first exposure and exposure continues until 1985. When evaluated in 1995, the first person has both a longer duration of exposure (15 years versus 10 years) and a longer time since first exposure (25 years versus 20
-
years), which illustrates the link between these two factors for a given age at first exposure.
-
Two people are born in 1945. One begins exposure at age 20; the other, at age 30. Exposure stops for both at age 40. The first person has a younger age at first exposure, a longer duration of exposure , and a longer time since first exposure, showing that all three factors may potentially be linked.
-
Two people are exposed to the same concentration of the agent. The person with the longer duration of exposure will, by definition, have a higher cumulative exposure.
Although it is possible to construct counterexamples to the above, these are fairly typical examples of what happens in many occupational settings. They are presented primarily to illustrate that examining the effects of one time-related factor may be difficult without information about the others.
Although it can be difficult to disentangle these interrelated effects, it is not impossible. Many occupational studies, for instance, have shown that the strongest effects of industrial chemicals on cancer occur 10 to 20 years after exposure begins, after age and calendar time have been controlled. Several radiation-related cancers including leukemia, gastrointestinal cancers, and breast cancer show age at exposure to be a strong determinant of risk (NAS, 1990).
Mortality and Incidence Studies for Examining Latency
If an agent is carcinogenic, it may increase the chance of cancer occurring, or it may accelerate development of the cancer so that it occurs at a younger age than it otherwise would have. The agent may also influence the likelihood that the cancer results in death or may shorten the time between occurrence of the disease and death caused by that disease. Which of these processes occurs may depend not only on the agent, but also on the site of cancer. For example, lung cancer tends to be fatal in a very high percentage of those who develop it, and death usually comes swiftly. For this site, therefore, a study of mortality is unlikely to provide different results from a study of incidence. In contrast, prostate cancer is fatal in a fairly small proportion of cases (incidence rates are five or more times higher than mortality rates) (Merrill and Brawley, 1997). For this reason, a study of prostate cancer mortality would be less likely to detect the effect of a carcinogenic agent than would a study of prostate cancer incidence, unless the agent increased the severity of disease. On the other hand, since prostate cancer is so common and occurs with an increasing frequency as men age, any study of prostate cancer incidence should examine whether those exposed to the agent of interest develop the cancer at an earlier age than those not exposed. This type of analysis could be accomplished using age-specific rates. Caution would have to be exercised in interpreting incidence studies because of the recent introduction of prostate-specific-antigen (PSA), a marker for prostate tumors that are not clinically detectable,
as a screening tool. Differences across subpopulations in the extent to which PSA is used could confound results (Gann, 1997).
In the investigation of latency, changes in relative risks with time since exposure will occur later for mortality studies than for incidence studies, by an amount of time approximately equal to the average time from occurrence of the cancer to death. If the agent has no effect on the probability of death or the age at death from the cancer, then mortality studies will result in a pattern of relative risks "shifted to the right" of the pattern that would have been observed in the corresponding incidence studies. In other words, the pattern with time since exposure will be similar, but the latency period will be longer. As a result, at any given point in the follow-up period a study of mortality will record fewer events than a study of incidence and consequently will have lower statistical power, even if the exposed and unexposed cases have the same prognosis. The problem of the mortality studies with lower statistical power is magnified for cancer sites that have long survival times or tend not to be fatal (e.g., prostate cancer). Most of the herbicide studies that reported latency results were mortality studies.
Measurement Errors That Are Time Related
In epidemiologic studies, a common problem in data quality involves errors in the assignment of exposure. These errors can occur when exposed individuals are erroneously categorized as unexposed, or when unexposed individuals are categorized as exposed. These errors can also occur when determining how large an exposure an individual received: high exposures may be assessed as lower than they really are, and vice versa. These errors can be classified in several ways:
-
One way of classifying errors considers whether they are related to the true exposure: either the errors are independent of true exposure (if highly exposed individuals are just as likely to be erroneously assessed as those who truly had low exposures) or they can depend on true exposure (e.g., if those receiving low exposure are not well assessed, but those at high exposure levels are assessed correctly).
-
Another way to categorize errors is whether they are random or systematic: random errors are just as likely to overestimate as to underestimate true exposure. By contrast, systematic errors occur when there is a tendency for the measured exposure to be lower or higher than the true exposure.
-
A third way to categorize errors is according to whether they are more likely or less likely—as opposed to equally likely—to occur in the nondiseased than in the diseased population (this problem arises more frequently in case-control studies than in cohort studies).
In general, most types of errors will distort the evidence, sometimes causing the analysis to show a weaker effect than is actually occurring and other times
showing an effect that is not real, or a stronger effect than the true one. In certain situations, namely when the misclassification of exposure status is unrelated to a person's ultimate disease status and neither time-related factors nor levels of exposure are examined (exposure is simply considered either present or absent), then the distortion will usually result in observing a weaker effect than truly exists. In an analysis evaluating how time-related factors such as duration of exposure or cumulative exposure influence risk, the effect of misclassification is difficult to predict. Finally it should be pointed out that errors can occur not just in exposure assessment, but also in ascertainment of the outcome. For example, the classification of diseases changes over time as new techniques for diagnosis are developed. The recent development of and widespread screening for PSA have seemingly "indicated" a large increase in the incidence of early, localized prostate cancer. In reality, we are simply moving our time of diagnosis to an earlier stage in the development of the cancer.
FOUR QUESTIONS ADDRESSED BY THE COMMITTEE
For each question outlined in the introduction to this chapter, the committee discusses the measures it is seeking in the reports of study results, how these measures are examined to address the particular question, the types of data a study would need to be informative about this question, and problems associated with the measures chosen by the committee.
How Long Does It Take After Exposure to Detect an Increase in Disease Risk?
Measures of Interest
Relative risks for specific intervals of time since exposure are the appropriate measures of interest for this question. One must examine the pattern of relative risks, looking for the earliest indication of an increase in risk relative to the unexposed comparison group. For protracted exposures, it is customary to examine the relative risks by time since first exposure, because the earliest detectable increase in relative risk may be a manifestation of the earliest exposure. In fact, relative risks for specific times since first exposure are often the only measures of latency reported for studies of protracted exposure to herbicides.
Data Requirements
The critical data item for this measure is the date of first exposure for each subject. With this date, the investigator can determine the time that each subject spent in each time since first exposure category. If full exposure histories are available, more sophisticated analyses are possible.
Potential Problems with this Approach
The "earliest indication of an increase in relative risk" is difficult to measure and will be refined as more data are collected. First, it is likely that latency periods vary among individuals; as a result, changes in risk in a population occur continuously rather than suddenly jumping from "normal" to ''above normal." Actual changes in relative risk probably would occur earlier than indicated by the analysis, but because of limitations in study designs, this increase might not be detectable. In other words, the degree to which an increased risk is statistically detectable depends on the size of the particular data set, as well as the magnitude of both the background level of risk (which in turn depends on the age of distribution of study subjects) and the relative increase in risk (which in turn depends on the exposure level received, variation in susceptibilities, length of follow-up in the study, and true distribution of latency periods among the exposed population). It should be noted that if the latency periods are highly variable among individuals, an analysis by time since first exposure may be somewhat insensitive, because the increased risk will appear slight and will occur quite gradually. Additionally, if the effect of time since exposure is modified by the intensity of exposure or the age at exposure, these other factors would have to be accounted for in the analysis. For example, the latency might be longer for a low-level exposure than for a higher one, in which case a study that examined only time since first exposure might encounter greater variability in latency periods and hence have less ability to assess how long it takes to observe an effect of exposure. Such effect modification would also limit the generalizability of results from one study to a different population or to another exposure scenario.
Limitations in the Data Available
Studies that examined changes in risk by time since first exposure used different categories of time, so that results are not always directly comparable. In many of the studies, when specific cancer sites were examined by time since first exposure, the number of deaths in each category became quite small, leading to less stable estimates. In most studies, the analyses by time since first exposure did not adjust for other time-related factors such as duration of exposure, intensity of exposure, or age at start of exposure, so the apparent effect of time since first exposure could very well be confounded by these other factors. In the case of Australian veterans who served in Vietnam, latency analyses examined the time since start of service, which may not have corresponded to the time since first exposure; for some veterans, the time since first exposure would have been shorter than this surrogate variable.
While recognizing these limitations in its evaluations, the committee has determined the earliest increases in relative risk reported in the literature. These are, of course, subject to change as more information becomes available.
How Long Does the Effect of Exposure Last?
Measure of Interest
Relative risks for specific intervals of time since last exposure are used to address this question. The pattern of relative risks is examined for the latest indication that the relative risk is greater than one.
Data Requirements
Dates of each start and stop of exposure are required to answer this question. These are needed to classify the subjects' time spent in each time-since-exposure category. If full exposure histories are available, more sophisticated analyses are possible. However, if the critical issue is "time since exposure stopped," multiple starts and stops will be difficult to analyze.
Potential Problems with this Approach
If exposure is protracted, time since exposure must be analyzed in the proper time-dependent fashion (Clayton and Hills, 1993). Tight adjustment for age is also necessary. To achieve adequate power and precision, a study group must have a sufficient number of subjects with long times since exposure ended. If exposure has been protracted, then much longer time periods of follow-up are needed than for addressing the previous question.
Limitations in the Data Available
Most of the studies reporting latency data for cohorts with protracted exposure examined only the time since first exposure (not time since last exposure). The study of Vietnam era veterans examined changes in mortality by years since they last served in Vietnam. For the Seveso cohort, the time since exposure ended is, for most subjects, quite close to the time since the accident occurred. Hence, all other factors being equal, these two studies would be the most amenable to answering the question of how long the effect of exposure lasts.
How Does the Effect of Exposure Vary with the Age at Which It Was Received?
Measures of Interest
Relative risks for exposure beginning at various ages are the critical measures needed to address this question. One must examine the pattern of relative risks associated with exposure beginning at various ages and com-
pare the patterns of relative risks by time since exposure across age at exposure categories.
Data Requirements
Dates of exposure and date of birth are the critical data required to construct these measures. These are needed to classify subjects as exposed or unexposed in each age category. The date of birth of study participants is generally known in epidemiologic studies. If information about level of exposure is available, it would be used in preference to the simple exposed/unexposed categorization. For the relative risks stratified by time since exposure, the data requirements include those described above.
Potential Problems with this Approach
The problems with this approach parallel those for the previous questions. A large study with long follow-up is more likely than a small study to detect differential age effects. Sample size becomes a practical problem, since analysis within age groups requires more data than pooling all age groups in exposure categories. To examine the time since exposure within an age group, comments about the investigation of relative risk by time since exposure apply here as well.
Limitations in the Data Available
The committee found no studies that report the results needed to address this question, namely, changes in cancer risk by age at exposure (or age at first exposure) to herbicides.
Does the Exposure Appear to Act at an Early or a Late Stage of the Carcinogenic Process?
Measures of Interest
The key statistical measures needed to address this question are the relative risks by age at exposure, the time since exposure began, and the time since exposure ended or, alternatively, the parameters in one of several models of carcinogenesis. In the multistage model of carcinogenesis, a healthy cell is presumed to go through a series of stages before becoming a cancer cell (Armitage and Doll, 1961; Chu, 1987). This model predicts specific patterns of relative risks by age and time since exposure, depending on whether the agent acts at an early or late stage of the carcinogenic process (Whittemore, 1977; Thomas, 1988). Further, the parameters in the multistage model or other mechanistic models, such as the two-event "initiator-promoter" model of Moolgavkar and Venzon (1979), may be estimated from cohort data to distinguish early-and late-stage effects.
Data Requirements
To construct these measures, complete exposure histories and the date of birth are required. The study group must include subjects with protracted exposures, and there must be variation with respect to exposure histories. A sizable study cohort is needed.
Potential Problems with this Approach
Large studies with high-quality data on exposure history are needed. Even when such data are available, it is difficult to distinguish early-from late-stage effects, possibly because many carcinogens have effects at more than one stage in the carcinogenic process or because the differences in susceptibility and latency period among individuals mask such effects.
Limitations in the Data Available
No studies in the published literature attempted to conduct analyses to determine the stage(s) at which exposure to herbicides such as TCDD exerts an effect.
REVIEW OF THE SCIENTIFIC LITERATURE
For purposes of this discussion, the review of the literature on herbicide exposure and cancer was focused on cancers in the categories of "sufficient" and "limited/suggestive" evidence of association, as found in both Veterans and Agent Orange: Health Effects of Herbicides Used in Vietnam (henceforth called VAO) (IOM, 1994) and Veterans and Agent Orange: Update 1996 (henceforth called Update 1996) (IOM, 1996)—that is, those cancers for which there was some evidence of an association. These are soft-tissue sarcoma, non-Hodgkin's lymphoma, Hodgkin's disease, prostate cancer, respiratory cancer, and multiple myeloma. Although VAO, Update 1996, and Chapter 7 of this report review the entire relevant literature on herbicide exposure, this chapter discusses only those articles that provide results the committee believes reflect, with reasonable accuracy, the timing of herbicide exposure, and that studied a sufficient number of cases to make some judgment about the patterns of relative risks reported. In addition, the discussion is restricted to those cancer sites for which the data were at least limited/suggestive of an association with the herbicides used in Vietnam.
Limitations of the Literature Review Approach
In Chapter 4, the committee considers the problem of using a literature review to determine whether an association exists between herbicides and disease. The committee concludes that for overall questions of association between
exposure and disease, the published literature would adequately report results, whether "positive" or "negative" with respect to association, from the studies that have been carried out to date. Thus, there should be little "publication bias" (the tendency for positive results to be published more frequently than negative) in the literature for association.
In a specific investigation of timing issues based on a review of the literature, the same question of publication bias has to be addressed. That is, is it more likely that results of investigations of timing issues will be published depending on the outcome of these investigations? Unlike measures of association (particularly relative risk) that are universally reported, results of investigations of timing issues are not reported routinely. Indeed, although it is not possible to determine the reasons that timing is or is not reported, it is quite plausible that negative results (i.e., no differential effect of timing) are reported less frequently than positive results. One likely scenario is that if no association is found between exposure and disease, then either timing issues were not investigated or they were investigated and only "interesting" results (i.e., large changes over time intervals) were reported; so-called "uninteresting" results (no association over all time intervals) were not reported. Thus, the committee recognizes that there is a potential for its review to reflect publication bias.
Overview of the Findings
Update 1996 reported results on the timing of exposure in relation to two cancer sites: respiratory and prostate, with considerably more information about the former than the latter. However, even for these cancers, the reports of some potentially informative studies did not include latency results, which suggests a potential for publication bias, although one cannot always know whether researchers did an analysis and failed to report it because the results were uninteresting or simply did not conduct the analyses. Also, both prostate and respiratory cancers are in the "limited/suggestive" evidence category, indicating the committee's belief that the evidence for association between herbicide exposure and these cancers is not conclusive. This view has not changed after the investigation of latency issues.
Since Update 1996 several more reports provide latency information on lung cancer. New data have also appeared regarding prostate cancer. With regard to the outcomes, several studies also report latency analyses for all cancer deaths combined, and there are now data on non-Hodgkin's lymphoma, multiple myeloma, lymphatic and hematopoietic cancers combined, and a few other sites. However, only those cancer sites for which at least two studies have provided latency analyses are discussed in this chapter. In addition, for any particular site, individual studies that do not have an overall excess of cancer at that site are not evaluated for the patterns of risk with respect to latency.
Studies using the proportionate mortality ratio (PMR)—that is, those that
enumerated only the deaths, not the number of individuals at risk—are ignored for the purpose of evaluating latency. This decision is based on the inherent weakness of PMR studies, in which the results in terms of one outcome are influenced by the results of another. (If an exposure causes an increase in both a more common outcome and a rarer one, the increase in mortality for the rarer outcome could be obscured in a PMR study.)
Table 8-1 summarizes the newly published papers that provide data on latency periods, the time since exposure categories they used, and the cancer sites they reported. Respiratory cancer, prostate cancer, and non-Hodgkin's lymphoma are discussed below, because these are the only sites that either were reviewed in Update 1996 or have more than one new study (not based on PMRs) that reports analyses by different latencies.
RESPIRATORY CANCER
Background
There is a substantial body of literature that explores issues of timing of exposure and respiratory cancer, because of its relatively high incidence and because numerous carcinogenic agents have been identified. Some of the studies are summarized here to provide a background for the examination of these issues.
Gamma Rays. In an investigation of latency issues for radiation exposure in atomic bomb survivors, it was found that the relative risk of respiratory cancer began to rise 5 to 10 years after exposure and reached a plateau about 15 years after exposure. Thirty years after exposure there was no evidence of a decrease in relative risk (Land, 1987). In addition, the effects of age at exposure are quite pronounced for some sites (NAS, 1990) (e.g., leukemia, digestive cancers, and breast cancer).
Radon Daughters. For miners exposed to radon daughters (radon decay products), the relative risk of lung cancer was seen to peak 5 to 10 years after first exposure, then slowly decline, although the risk still appears to be elevated even 30 years after exposure (Lubin et al., 1994; Thomas et al., 1994). In addition, the effect of exposure varies with age at exposure: a given exposure level results in a lower relative risk in older workers than it does in younger workers.
Smoking. Analyses of lung cancer indicate that the relative risks begin to rise substantially after about 20 years from the initiation of cigarette smoking. Among ex-smokers, the relative risk declines to about 50 percent of that of smokers by 12 years after cessation but then remains fairly constant (and elevated relative to those who never smoked). Among continuing smokers, for the same cumulative amount smoked, the relative risk declines with age at the start of smoking.
TABLE 8-1 New Studies with Latency Data
Reference |
Measures of Association |
Latency Periods* |
Population |
Cancer Outcomes |
Michalek et al., 1998 |
SMR 20+ |
0-20 |
Ranch Hands |
Respiratory (other sites presented for <20 or 20+ years, but not for both) |
Bertazzi et al., 1997 |
RR (Mortality) |
0-15 |
Seveso residents |
All cancers Digestive cancer Rectal cancer (males only) Lymphatic and hematopoietic Stomach (females only) Leukemia (males only) Multiple myeloma (females only) |
Crane et al., 1997 |
SMR |
<10 11-15 16-20 21-25 >25 |
Australian Vietnam veterans |
All cancers Lung cancer |
Kogevinas et al., 1997 |
SMR |
0-9 >20 |
IARC cohort 10-19 |
All neoplasms (140-208) Lung cancer NHL Soft tissue sarcoma |
Becher et al., 1997 |
SMR |
0- <10 10-<20 20+ |
German production workers NHL |
All neoplasms (140-208) Buccal cavity/pharynx Lung All lymphatic and hematopoietic |
Watanabe and Kang, 1996 |
PMR |
0-10 11-15 > 16 |
U.S. Vietnam-era veterans |
All cancers (170-174, 185-209) Pancreas Larynx Lung Connective tissue Skin Prostate Testis NHL Hodgkin's disease Multiple myeloma |
* Refers to years since start of exposure, with the exception of the study by Watanabe and Kang (1996), who examined time since last year in Vietnam, and Michalek et al. (1998), who did not specify what the latency represented. |
Arsenic. In a cohort of workers from a copper smelter in Montana, relative risks were observed to increase with time after exposure, reaching a maximum between 15 and 20 years after exposure, after which they slowly declined (Breslow and Day, 1987). There was little change in relative risk with age at first exposure.
Asbestos. In a cohort of workers exposed only briefly to high levels of asbestos during World War II, the relative risk rose sharply between 5 and 10 years after exposure, after which it remained constant up to 40 years after exposure (U.S. EPA, 1986). The relative risks are independent of age at exposure.
Nickel. In a cohort study of nickel refiners in England and Wales, the relative risk for lung cancer peaks less than 20 years after exposure, then decreases sharply. After 50 years, however, the risk is still elevated, except in the low-exposure group. The relative risks are more or less constant across age at first exposure (Kaldor et al., 1986). It is interesting to note that in contrast to these results, the same author reported quite a different pattern of relative risks for nasal sinus cancer. It was found that the relative risks for nasal sinus cancer continued to increase slowly with time since exposure, but increased markedly with age at first exposure.
Thus, for all of these exposures, increases in relative risk either reached a plateau or peaked within 20 years after exposure. This indicates that the first detectable increases occurred somewhat earlier than this. The pattern of relative risks after reaching the peak and the pattern with age at exposure vary greatly across the agents, probably reflecting different mechanisms of action.
Review of the Scientific Literature
Since respiratory cancer is fairly common, the committee has focused on studies with at least 10 cases. Five studies have reported timing effects related to herbicide exposure and respiratory cancer. The National Institute for Occupational Safety and Health (NIOSH) study of chemical production workers gives the most detailed account of timing effects and exposure to TCDD (Fingerhut et al., 1991). Standardized mortality ratios (SMRs) for lung cancer were 0.8, 1.0, and 1.2 for 0-9, 10-19, and 20+ years since first exposure to TCDD, based on a total of 85 cases. SMRs for time since first exposure are further stratified by duration of exposure, as reproduced in Table 8.2. An association between TCDD exposure and respiratory cancer is not observed in years 0-9 after first exposure. Effects begin to be observed in the second decade after exposure began among those with at least 5 years of exposure, and they have not disappeared 20 or more years after exposure. The latency may be longer for those with shorter durations of exposure.
Data from Seveso provided by Bertazzi et al. (1989a,b) and summarized in Tables 8-3A and 8-3B indicate that respiratory cancer mortality was not in-
TABLE 8-2 NIOSH Study: Respiratory Cancer Relative Mortality by Time Since First Exposure and Duration of Exposure to TCDD
|
Duration of Exposure to TCDD (years) |
|||||||||
Time Since First Exposure |
< 1 |
|
1-4 |
|
5-14 |
|
15 + |
|
Overall |
|
|
Obs |
SMR |
Obs |
SMR |
Obs |
SMR |
Obs |
SMR |
Obs |
SMR |
0-9 years |
3 |
0.8 |
3 |
1.0 |
1 |
0.8 |
0 |
0.0 |
7 |
0.8 |
10-19 years |
6 |
0.7 |
5 |
0.8 |
9 |
1.8 |
1 |
1.4 |
21 |
1.0 |
>20 years |
17 |
1.0 |
17 |
1.3 |
14 |
1.5 |
9 |
1.6 |
57 |
1.2 |
Total |
26 |
0.9 |
25 |
1.1 |
24 |
1.5 |
10 |
1.5 |
85 |
1.1 |
SOURCE: Fingerhut et al., 1991, Table 4. |
TABLE 8-3a Seveso Study: Lung Cancer Mortality Ratios in Men by Calendar Period
|
RR |
|||
Time Since Exposure |
Zone A |
Zone B |
Zone R |
|
0-5 years |
0.0 |
1.1 |
0.7 |
|
6-10 years |
2.0 |
1.8 |
0.9 |
|
11-15 years* |
1.0 |
1.0 |
1.0 |
|
SOURCE: Bertazzi et al., 1997, Table 3, and Bertazzi et al., 1989b, Tables 4, 5, and 7. * Relative risks have been calculated using data from the two published reports. |
TABLE 8-3b Seveso Study: Lung Cancer Mortality in Males for 15-Year Follow-Up
|
Observed |
Expected |
RR |
Zone A |
4 |
4.2 |
1.0 |
Zone B |
34 |
27.6 |
1.2 |
Zone R |
178 |
194.4 |
0.9 |
SOURCE: Bertazzi et al., 1997, Table 3. |
creased among those in the exposed areas during the period from 0 to 5 years after the accident but was increased in years 6-10 for the most proximate residential areas. In the 15-year follow-up of the Seveso cohort, no additional data are presented on latency for respiratory cancer (Bertazzi et al., 1997), but given the results from several publications, the committee has calculated the relative risk for years 11-15 as 1.0 in all three zones.
In an 18-year follow-up of Finnish herbicide applicators, Asp et al. (1994) give the SMRs for respiratory cancer relative to the Finnish male age-and calendar-year-specific rates in such a way that SMRs could be calculated by time since first exposure for 0-9, 10-15, and >15 years. These data show that there is no clear pattern according to time since first exposure, but there is also no overall association with respiratory cancer, probably because the exposures were on average only four weeks' duration.
In a report on four occupational cohorts involved in phenoxy herbicide and chlorophenol manufacturing in Germany, with 47 lung cancer deaths and an overall SMR of 1.4, Becher et al. (1996) showed the highest relative risk in the first decade (SMR = 1.80), declining thereafter (SMR = 1.38 between 10 and 20 years after exposure, and 1.35 thereafter). These data are presented in Table 8-4. The same cohorts were included in the much larger International Agency for Research on Cancer (IARC) multicohort occupational study (Kogevinas et al., 1997), which similarly found the highest relative risks in the first 10 years: SMRs
TABLE 8-4 German Phenoxy Herbicide and Chlorophenol Manufacturing Workers Study: Lung Cancer Observed and Expected Deaths and SMRs for Men by Time Since First Exposure
Time Since Exposure |
Observed |
Expected |
SMR |
<10 years |
8 |
4.4 |
1.8 |
10 to <20 years |
14 |
10.1 |
1.4 |
>20 years |
25 |
18.4 |
1.4 |
SOURCE: Becher et al., 1996, Table 4. |
for 0-9, 10-19, and 20 or more years were 1.2, 1.0, and 1.2, respectively, based on 34, 64, and 127 lung cancer deaths. The IARC results are shown in Table 8-5.
The study of Ranch Hands (Michalek et al., 1998) does examine latency for several cancer sites but does not define whether this involves time since first service, since last service, since start of service in Vietnam, or since last service in Vietnam. This group of veterans experienced fewer respiratory cancer deaths than expected in the first 20 years (3 observed, 5.6 expected) and a slight excess after 20 years (9 observed, 7.2 expected).
A report on the Australian veterans who served in Vietnam provides additional information on the time since first year of service (Crane et al., 1997). Since the first year of service may have been earlier than the first year in Vietnam, or the first year of exposure, any latency observed in these data would be longer than the actual latency. The pattern of SMRs for lung cancer deaths during 1980-1994 (no lung cancers were observed before 1980) was as follows: 2.5, 0.9, 1.3, 1.3, and 1.1 for the periods <10, 11-15, 16-20, 21-25, and >25 years respectively since the start of service. Note, however, that the SMR of 2.5 in the early period is based on only 3 lung cancer deaths, whereas the remaining periods had 17, 60, 95, and 35 lung cancer deaths, respectively. These results can be found in Table 8-6.
TABLE 8-5 IARC International Study of Workers Exposed to TCDD or Higher Chlorinated Dioxins: Lung Cancer Observed and Expected Deaths and SMRs for Men by Time Since First Exposure
Time Since First Exposure |
Observed |
Expected |
SMR |
0-9 years |
34 |
27.9 |
1.2 |
10-19 years |
64 |
61.5 |
1.0 |
>20 years |
127 |
110.4 |
1.2 |
SOURCE: Kogevinas et al., 1997, Table 5. |
TABLE 8-6 Australian Vietnam Veterans Study: Lung Cancer Observed and Expected Deaths and SMRs for Men by Time Since Start of Military Service
Time Since Start of Service |
Observed |
Expected |
SMR |
<10 years |
3 |
1.2 |
2.5 |
11-15 years |
17 |
19.7 |
0.9 |
16-20 years |
60 |
45.9 |
1.3 |
21-25 years |
95 |
73.0 |
1.3 |
>25 years |
35 |
32.2 |
1.1 |
SOURCE: Crane et al., 1997, Table E-19. |
Conclusions
Perhaps because respiratory cancers are the most common type of cancer in all of the cohort studies, there is more latency information available for this site than for any other. However, based on the review of the evidence in VAO, Update 1996, and Chapter 7 of this report, respiratory cancer is in the ''limited/suggestive" evidence category, indicating that the committee believes the evidence for association between herbicide exposure and these cancers is not conclusive. Although an investigation of latency effects could result in a change in the categorization of evidence, in this case it did not. The fact that the committee reviewed the literature for latency effects does not imply an a priori belief on the part of the committee that the association is definitive.
How Long Does It Take After Exposure to Detect an Increase in Disease Risk?
If the association between TCDD exposure and respiratory cancer is causal, then the evidence in the literature suggests that the time between exposure to TCDD and an increased risk of respiratory cancer may be less than 10 years. Although the NIOSH study (Fingerhut et al., 1991) does not begin to show an effect until 10 years after exposure, the Seveso cohort (Bertazzi et al., 1989a,b, 1997) data show an increased occurrence of death from respiratory cancer beginning 6-10 years after initiation of an exposure, and the IARC cohort (Kogevinas et al., 1997) demonstrates the highest increase in the first decade. Australian Vietnam veterans (Crane et al., 1997) also showed an elevated risk of lung cancer mortality in the first decade, but this finding is based on a small number of deaths. The latest report on Ranch Hands (Michalek et al., 1998) shows a reduced risk in the first 20 years since exposure.
The committee also finds evidence in the literature that the time between exposure and the detection of respiratory cancer probably depends on the magnitude of exposure. This evidence is seen in the Fingerhut et al. (1991) study, which was the only analysis that presented a cross-classification of time since first expo-
sure with duration of exposure. With latency depending on the level of exposure, one would not necessarily expect to see the same pattern for time since exposure in all studies. Nor would one expect the pattern of risk over time since exposure to be the same for Vietnam veterans as it was for those exposed in manufacturing plants or through accidental environmental releases of the same chemicals.
When exposure is not protracted, a pattern with time since exposure could be due to confounding by another exposure that has a similar trend among the exposed, but no such trend among the unexposed. When exposure is protracted, as in two of the occupational cohorts, an even more complex pattern would have to occur for confounding to explain the results. Although one can hypothesize that a particular pattern of risk with latency could be due to confounding, evidence for differential confounding by years-since-first-exposure may be difficult to find, particularly for cohorts with a protracted exposure.
How Long Do the Effects of Exposure Last?
If there is, in fact, a causal association between TCDD exposure and respiratory cancer, the literature suggests that the risk can be elevated beginning at least as early as 6 years after exposure, but the literature is less clear on how long the effect lasts. In the NIOSH study (Fingerhut et al., 1991), risks were most elevated 20 or more years after exposure began, even for those with only 1-4 years of exposure (i.e., 16-19 years after exposure ended). The SMR in the IARC study (Kogevinas et al., 1997) for workers exposed to TCDD or higher chlorinated dioxins dropped to 1.0 from 10 to 19 years after first exposure, and rose to 1.2 for 20 or more years since first exposure (95 percent confidence interval [95% CI] 1.0-1.4), but no analyses are presented by years since last exposure. The most recent follow-up of the Seveso cohort (Bertazzi et al., 1997) did not provide any data on lung cancer latency. The Ranch Hands (Michalek et al., 1998) showed an SMR of 1.3 for 20 or more years of latency, which, depending on how latency was defined, could represent approximately a few years shorter time since service ended. Among Australian veterans (Crane et al., 1997), SMRs were 1.3 for 16-25 years after service began and 1.3 for 25 or more years after service began, but again this analysis was not for years since service ended or since leaving Vietnam. Given the scant data, the committee cannot determine how long it takes before the relative risks return to one. The lack of conclusive data on timing parallels the lack of definitive data on whether exposures to TCDD and other herbicides are causally associated with respiratory cancer.
How Does the Effect of Exposure Vary with the Age at Which It Was Received?
None of the available studies provides information on the variation of the effect of exposure with age.
Does the Carcinogen Appear to Act at an Early or a Late Stage of the Carcinogenic Process?
None of the available studies address this issue.
PROSTATE CANCER
Background
There do not appear to be environmental exposures other than herbicides associated with prostate cancer for which latency issues have been investigated. Although there are new data on prostate cancer since Update 1996, there are few new data on latency for prostate cancer.
Review of the Scientific Literature
The NIOSH study of chemical production workers exposed to TCDD (Fingerhut et al., 1991) reports SMRs for prostate cancer for the entire cohort, as well as for 20+ years since first exposure, by the duration of exposure: short = <1 and long = 1 + years. The presentation of results in their paper did not allow a comparison of SMRs for <20 and 20+ years since first exposure within duration-of-exposure categories. Based on the material presented, SMRs were calculated according to years since first exposure; these are listed in Table 8-7. No difference in SMRs was observed for time since first exposure. The wide categories of time since exposure limit the degree to which any inferences can be drawn about latency.
If the exposure to TCDD after the Seveso accident was of relatively short duration, the time since the accident is essentially the same as the time since exposure. The mortality studies by Bertazzi et al. (1989a,b) provide results relevant to the timing of exposure for prostate cancer mortality. By combining the data from earlier reports with those published in 1997 (Bertazzi et al., 1997), relative risks were calculated for the 11-15 years since the accident. If a low rate of immigration is assumed, the calendar-period relative risks will approximate
TABLE 8-7 NIOSH Production Workers Study: Prostate Cancer Observed and Expected Numbers of Deaths and SMRs by Time Since First Exposure to TCDD
Time Since First Exposure |
Observed |
Expected |
SMR |
<20 years |
6 |
5.0 |
1.20 |
>20 years |
11 |
8.9 |
1.23 |
SOURCE: Derived from Fingerhut et al., 1991, Table 2. |
those for three categories of time since exposure. The relative risks are summarized in Tables 8-8A and 8-8B. There are no cases in zone A, 6 cases in zone B, and 39 in zone R. In zones B and R, there is a decrease in the relative risk with calendar period, although the small number of cases and the fact that this is a mortality rather than an incidence study preclude strong statements about the actual pattern of relative risks.
A recent update of the Ranch Hand study reported prostate cancer for >20 years since the start of exposure but not for <20 years (Michalek et al., 1998). Hence it provides no information about how time since exposure might be related to prostate cancer risk.
Conclusions
The committee's review of the literature yielded only two sets of articles (Fingerhut et al., 1991; Bertazzi et al., 1989a,b, 1997) on prostate cancer that presented latency-related results and a sufficient number of cases for statistical analysis. Prostate cancer is in the category of limited/suggestive evidence, so it is important to keep in mind that the committee believes the evidence for an association between herbicide exposure and prostate cancer is not conclusive. Although the investigation of latency effects could result in a change in the categorization of evidence, in this case it did not. The fact that the committee reviewed the literature for latency effects does not imply an a priori belief on the part of the
TABLE 8-8a Seveso Study: Prostate Cancer Relative Mortality in Men by Calendar Period
|
RR |
||
Time Since Exposure |
Zone A |
Zone B |
Zone R |
0-5 years |
no cases |
2.8 |
1.9 |
6-10 years |
no cases |
1.5 |
1.2 |
11-15 years |
no cases |
0.9 |
1.0 |
SOURCE: Bertazzi et al., 1997, Table 3, and Bertazzi et al., 1989b, Tables 4, 5, and 7. |
TABLE 8-8b Seveso Study: Prostate Cancer Relative Mortality in Men for 15-Year Follow-Up
|
Observed |
Expected |
RR |
Zone A |
0 |
0.7 |
0.0 |
Zone B |
6 |
4.8 |
1.2 |
Zone R |
39 |
33.0 |
1.2 |
SOURCE: Bertazzi et al., 1997, Table 3. |
committee that the association is definitive. A further caveat is the concern that evidence based on mortality studies may have no relation to the latency period that might apply to the incidence of prostate cancer.
How Long Does It Take After Exposure to Detect an Increase in Disease Risk?
The limited data from the NIOSH study (Fingerhut et al., 1991) are uninformative; they provide no information about any pattern of relative risk in the 0-20 years since exposure began and show no difference between the first 20 years and the subsequent years since exposure. The Seveso studies (Bertazzi et al., 1989a,b, 1997) suggest that the relative risk for prostate cancer mortality is higher in the early period after exposure begins and declines 11-15 years after exposure. As for respiratory cancer, the extrapolation of latency across studies is very uncertain, since it can vary according to the exposure level and other factors such as age at exposure.
How Long Do the Effects of Exposure Last?
The available evidence is limited to the results from the Seveso cohort (Bertazzi et al., 1989a,b, 1997). Since external exposure was for a brief period, the findings are the same as for time since first exposure, with evidence suggesting little risk if any, between 10 and 15 years after the end of exposure.
How Does the Effect of Exposure Vary with the Age at Which It Was Received?
None of the studies provides information on the variation of the effect of exposure with age.
Does the Carcinogen Appear To Act at an Early Or a Late Stage of the Carcinogenic Process?
None of the studies address this issue.
NON-HODGKIN'S LYMPHOMA
Background
In Update 1996, non-Hodgkin's lymphoma (NHL) was not reviewed for time-related factors because of the lack of published data. Three recent reports provide latency data on non-Hodgkin's lymphoma in relation to herbicide exposures, but one of these relied on PMRs (Watanabe and Kang, 1996).
Review of the Scientific Literature
Two occupational cohort studies address the latency issue for non-Hodgkin' s lymphoma. The first is a report from four cohorts in Germany, with a total of six deaths, shown in Table 8-9 (Becher et al., 1996). None were observed in the 10 years since first exposure, 2 were observed in the second decade, and 4 in the third decade or later, for SMRs of 0, 3.6, and 4.3, respectively. Only the latter is significantly elevated. These data are included in the large IARC cohort of herbicide manufacturing workers (Kogevinas et al., 1997), which observed 2, 8, and 14 deaths from this disease, yielding SMRs of 0.6 (95% CI 0.1-2.3), 1.5 (95% CI 0.6-2.9), and 1.6 (95% CI 0.9-2.7). These data are presented in Table 8-10.
Conclusions
The committee's review of the literature yielded only a few papers with data on non-Hodgkin's lymphoma that provided latency-related results and sufficient numbers of cases for statistical analysis. Non-Hodgkin's lymphoma is in the category of having sufficient evidence of an association with exposures to the herbicides used in Vietnam, which means that the committee finds strong evidence of an association and is convinced that the association is not due to bias or confounding from other factors.
TABLE 8-9 German Phenoxy Herbicide and Chlorophenol Manufacturing Workers Study: Non-Hodgkin's Lymphoma Observed and Expected Deaths and SMRs for Men by Time Since First Exposure
Time Since First Exposure |
Observed |
Expected |
SMR |
< 10 years |
0 |
0.4 |
0 |
10 to <20 years |
2 |
0.6 |
3.6 |
>20 years |
4 |
0.9 |
4.3 |
SOURCE: Becher et al., 1996, Table 4. |
TABLE 8-10 IARC International Study of Workers Exposed to TCDD or Higher Chlorinated Dioxins: Non-Hodgkin's Lymphoma Observed and Expected Deaths and SMRs for Men by Time Since First Exposure
Time Since First Exposure |
Observed |
Expected |
SMR |
0-9 years |
2 |
3.2 |
0.6 |
11-19 years |
8 |
5.5 |
1.5 |
>20 years |
14 |
8.6 |
1.6 |
SOURCE: Kogevinas et al., 1997, Table 5. |
How Long Does It Take After Exposure to Detect an Increase in Disease Risk?
The data suggest that the increase in risk is not immediate. Occupational cohorts do not begin to show an excess of this type of cancer until the second decade after initial exposure.
How Long Do the Effects of Exposure Last?
The available evidence suggests that the effect of herbicide exposure on the risk of non-Hodgkin's lymphoma lasts for more than 20 years. No data are available to examine latencies of 30 or more years.
How Does the Effect of Exposure Vary with the Age at Which It Was Received?
None of the studies provides information on the variation of the effect of exposure with age.
Does the Carcinogen Appear to Act at an Early or a Late Stage of the Carcinogenic Process?
None of the studies address this issue.
RELEVANCE OF LATENCY IN ASSESSING THE EFFECT OF HERBICIDES ON CANCER RISK IN VIETNAM VETERANS
One of the committee's tasks was to assess the likelihood that exposure to herbicides used in Vietnam resulted in or will result in increased risk of disease in Vietnam veterans. Currently, any such inference would have to be based on extrapolation from the findings about disease experience of other groups exposed to TCDD or herbicides generally. Given that we know when the potential exposure to TCDD and other herbicides used in Vietnam began and ended, it would appear reasonable to examine time-related factors for those who served in Vietnam, but to date, no adequate analyses of time-related factors for cancer occurrence in Vietnam veterans have been published. The extrapolation from other types of studies is problematic for several reasons. Brief exposures, such as occurred in Seveso, and chronic occupational exposures may not apply to Vietnam veterans because of the different exposure situation. For example, there is evidence in the literature (e.g., for respiratory cancer) that latency can vary not only among individuals, but also according to other aspects of the exposure scenario, such as the magnitude of exposure. Thus,. if high exposures in an occupational setting result in a certain pattern of relative risks with time since first
exposure, this pattern may not hold for lower-level exposures like those that occurred in Vietnam. Similarly, direct evidence was not presented to evaluate the impact of age at exposure to herbicides. It is possible that the age at which exposure was received could influence the pattern of latency observed (e.g., exposures incurred at younger ages could be more potent, but the impact might not be seen for a longer time period; conversely, exposures at older ages might be more harmful, particularly in the short run). Unfortunately, the data are not available to evaluate the hypothesis that age at exposure is important. A major limitation of the analyses discussed in this chapter is the failure of most studies to conduct analyses of latency that also controlled for factors such as duration of exposure, age, and calendar time of exposure (or analyses of age at exposure that controlled for time since exposure), particularly for occupational cohorts with protracted exposure periods.
Another consideration is the long retention time of TCDD and other highly chlorinated herbicides. Since body burdens from any exposure, no matter how brief, result in continuing exposure of internal organs, the concept of time since exposure ended has a different meaning than for chemical agents that are excreted quickly.
A third issue concerns the distinction between morbidity and mortality. As discussed earlier in this chapter, the latency between exposure and death is composed of two parts: latency until disease is detected and time between disease occurrence and death. For diseases with low survival, such as respiratory cancer, the time between disease occurrence and death is generally short, and therefore, a study focusing on mortality will give a good approximation of the latency period. However, for diseases that are not always fatal or that have a long survival time, such as prostate cancer, it is preferable to examine incidence rather than mortality. Thus, further data on the incidence of prostate cancer would be of great help, since relatively few men with prostate cancer die from it.
Overall, the data on latency do not alter the committee's conclusions with regard to the categories of evidence for individual cancer sites, but they do provide some information on how long the effects of herbicide exposures last. The evidence suggests that if respiratory cancer does result from exposures to the herbicides used in Vietnam, the greatest relative risk for lung cancer may be in the first decade after exposure, but until further follow-up has been carried out for some of the cohorts, it will not be possible to put an upper limit on the length of time these herbicides could exert their effect. For prostate cancer, the published data are largely uninformative, and conclusions must await more definitive studies, preferably using incidence rather than mortality. For non-Hodgkin's lymphoma, effects are seen in the second decade after exposure begins and continue to be observed more than 20 years after external exposure ends. Because of the long retention times of TCDD, internal exposures can continue long after external exposures cease.
REFERENCES
Armenian HK. 1987. Incubation periods in cancer epidemiology. Journal of Chronic Diseases 40:9-16.
Armitage P, Doll R. 1961. Stochastic models for carcinogenesis. In Neyman J ed. Proceedings of the 4th Berkeley Symposium on Mathematical Statistics and Probability. Berkeley: University of California Press.
Asp S, Riihimaki V, Hernberg S, Pukkala E. 1994. Mortality and cancer morbidity of Finnish chlorophenoxy herbicide applicators: an 18-year prospective follow-up. American Journal of Industrial Medicine 26:243-253.
Becher H, Flesch-Janys D, Kauppinen T, Kogevinas M, Steindorf K, Manz A, Wahrendorf J. 1996. Cancer mortality in German male workers exposed to phenoxy herbicides and dioxins. Cancer Causes and Control 7:312-321.
Bertazzi PA, Zochetti C, Pesatori AC, Guercilena S, Sanarico M, Radice L. 1989a. Mortality in an area contaminated by TCDD following an industrial incident. Medicina Del Lavoro 80:316-329.
Bertazzi PA, Zocchetti C, Pesatori AC, Guercilena S, Sanarico M, Radice L. 1989b. Ten-year mortality study of the population involved in the Seveso incident in 1976. American Journal of Epidemiology 129:1187-1200.
Bertazzi PA, Zochetti C, Guercilena S, Consonni D, Tironi A, Landi MT, Pesatori AC. 1997. Dioxin exposure and cancer risk: a 15-year mortality study after the "Seveso accident." Epidemiology 8:646-652.
Breslow NE, Day NE. 1987. Statistical Methods in Cancer Research, Vol II: The Design and Analysis of Cohort Studies. Lyon: International Agency for Research on Cancer. Oxford University Press.
Chu K. 1987. A non-mathematical view of mathematical models of cancer. Journal of Chronic Diseases 40:163-170.
Clayton D, Hills M. 1993. Statistical Models in Epidemiology. New York: Oxford University Press.
Crane PJ, Barnard DL, Horsley KW, Adena MA. 1997. Mortality of Vietnam Veterans: The Veteran Cohort Study: A Report of the 1996 Retrospective Cohort Study of Australian Vietnam Veterans. Canberra: Department of Veterans' Affairs.
Enterline P, Henderson V. 1973. Type of asbestos and respiratory cancer in the asbestos industry. Archives of Environmental Health 27:312-317.
Fingerhut MA, Halperin WE, Marlow DA, Piacitelli LA, Honchar PA, Sweeney MH, Greife AL, Dill PA, Steenland K, Suruda AJ. 1991. Cancer mortality in workers exposed to 2,3,7,8-tetrachlorodibenzo-p-dioxin . New England Journal of Medicine 324:212-218.
Gann PH. 1997. Interpreting recent trends in prostate cancer incidence and mortality. Epidemiology 8:117-120.
Institute of Medicine (IOM). 1994. Veterans and Agent Orange: Health Effects of Herbicides Used in Vietnam. Washington, DC: National Academy Press.
Institute of Medicine. 1996. Veterans and Agent Orange: Update 1996. Washington, DC: National Academy Press.
Kaldor J, Peto J, Easton D, Doll R, Hermon C, Morgan L. 1986. Models for respiratory cancer in nickel refinery workers. Journal of the National Cancer Institute 77:841-848.
Kogevinas M, Becher H, Benn T, Bertazzi PA, Boffetta P, Bueno-de-Mesquita HB, Coggon D, Colin D, Flesch-Janys D, Fingerhut M, Green L, Kauppinen T, Littorin M, Lynge E, Mathews JD, Neuberger M, Pearce N, Saracci R. 1997. Cancer mortality in workers exposed to phenoxy herbicides, chlorophenols, and dioxins. American Journal of Epidemiology 145:1061-1075.
Land C. 1987. Temporal distributions of risk for radiation-induced cancers. Journal of Chronic Diseases 40:45-58.
Lubin J, Boice J, Edling C, Hornung R, Howe G, Kunz E, Kusiak R, Morrison H, Radford E, Samet J, Tirmarche M, Woodward A, Xiang Y, Pierce D. 1994. Radon and Lung Cancer Risk: A Joint Analysis of 11 Underground Miners Studies. U.S. Department of Health and Human Services, Public Health Service, National Institutes of Health, Bethesda.
Merrill RM, Brawley OW. 1997. Prostate cancer incidence and mortality rates among white and black men. Epidemiology 8:126-131.
Michalek JE, Ketchum NS, Akhtar FZ. 1998. Post-service mortality of Air Force veterans occupationally exposed to herbicides in Vietnam: 15 year follow-up. American Journal of Epidemiology 148(8):786-792.
Moolgavkar S, Venzon D. 1979. Two-event models for carcinogenesis: incidence curves for childhood and adult tumors. Mathematical Biosciences 47:55-77.
National Academy of Sciences (NAS), National Research Council. Committee on the Biological Effects of Ionizing Radiation (BEIR V). 1990. Health Effect of Exposures to Low Levels of Ionizing Radiation. Washington, DC: National Academy Press
Peto J. 1985. Some problems in dose-response estimation in cancer epidemiology. In Voug V, Butler G, Hoel D, Peakall D, eds. Methods for Estimating Risk of Chemical Injury: Human and Non-Human Biota and Ecosystems. New York: Wiley.
Thomas DC, Pogoda J, Langholz B, Mack W. 1994. Temporal modifiers of the radon-smoking interaction. Health Physics 66:257-262.
Thomas DC. 1983. Statistical methods for analyzing effects of temporal patterns of exposure on cancer risks. Scandinavian Journal of Work, Environment, and Health 9:353-366.
Thomas DC. 1987. Pitfalls in the analysis of exposure-time-reponse relationships. Journal of Chronic Disease 40:70-78.
Thomas DC. 1988. Models for exposure-time-response relationships with applications to cancer epidemiology . Annual Review of Public Health 9:451-482.
U.S. Environmental Protection Agency (EPA). 1986. Airborne Asbestos Health Assessment Update. Research Triangle Park, NC: Environmental Criteria and Assessment Office.
Watanabe KK and Kang HK. 1996. Mortality patterns among Vietnam veterans: a 24-year retrospective analysis. Journal of Occupational and Environmental Medicine 38:272-278.
Whittemore AS. 1977. The age distribution of human cancer for carcinogenic exposures of varying intensity. American Journal of Epidemiology 106:418-432.