5
Recommendations Regarding Epidemiologic Studies Using the Exposure Assessment Model

With this chapter the committee addresses the central question put to it by the Department of Veterans Affairs (VA): What are the best ways to use the Stellman team’s herbicide exposure assessment model? The committee has concluded that epidemiologic studies applying the exposure assessment model should be considered, and in this chapter it discusses the potential contributions of and pitfalls in conducting such studies. The chapter discusses design and statistical considerations for studies, including the selection of health outcomes, potential study populations, and other criteria for optimal use of the model.

SHOULD STUDIES BE DONE USING THE STELLMAN TEAM’S MODEL?

The committee carefully considered the question of whether, given the shortcomings of the model in its current form and its inherent limitations even with further refinements, it holds promise for generating informative epidemiologic studies of herbicides and health among Vietnam veterans. The committee concluded that the answer is yes. It reached this decision based on two key considerations: (1) The exposure assessment model is applicable to the population of ultimate interest, namely Vietnam veterans; and (2) most previous studies of this population have been so severely limited with respect to exposure assessment that a more accurate yet imperfect method would advance the current understanding of whether herbicide exposure is associated with adverse health outcomes among Vietnam vet-



The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 84
5 Recommendations Regarding Epidemiologic Studies Using the Exposure Assessment Model W ith this chapter the committee addresses the central question put to it by the Department of Veterans Affairs (VA): What are the best ways to use the Stellman team’s herbicide exposure assessment model? The committee has concluded that epidemiologic studies applying the exposure assessment model should be considered, and in this chapter it discusses the potential contributions of and pitfalls in conducting such studies. The chapter discusses design and statistical considerations for studies, including the selection of health outcomes, potential study populations, and other criteria for optimal use of the model. SHOULD STUDIES BE DONE USING THE STELLMAN TEAM’S MODEL? The committee carefully considered the question of whether, given the shortcomings of the model in its current form and its inherent limitations even with further refinements, it holds promise for generating informative epidemiologic studies of herbicides and health among Vietnam veterans. The committee concluded that the answer is yes. It reached this decision based on two key considerations: (1) The exposure assessment model is applicable to the population of ultimate interest, namely Vietnam veterans; and (2) most previous studies of this population have been so severely lim- ited with respect to exposure assessment that a more accurate yet imperfect method would advance the current understanding of whether herbicide exposure is associated with adverse health outcomes among Vietnam vet- 

OCR for page 84
 RECOMMENDATIONS REGARDING EPIDEMIOLOGIC STUDIES erans. We recognize the logistical challenges to be faced in obtaining the needed model inputs (troop locations, unit membership), but based on ongoing and completed activities, these challenges appear to be surmount- able. At the same time, while we appreciate the potential contributions from applying the model, we also believe that the best use of the model requires that its limitations be understood and considered when making decisions about the conduct of epidemiologic studies and particularly in interpreting the results of those studies. CONTRIBUTIONS AND PITFALLS OF STUDIES USING THE MODEL Studying the Population of Interest Efforts to determine whether herbicide exposure among Vietnam veterans is associated with adverse health outcomes follow one of two approaches: (1) information from other populations, such as pesticide applicators or chemical production workers, is extrapolated to the experi- ence of Vietnam veterans, an approach that has the advantage of being grounded in information about higher exposures that are more accurately characterized; or (2) studies are conducted among the Vietnam veterans themselves. The quality of the data and the statistical power of the first group of studies have been superior and will likely continue to generate more accurate information on herbicide exposure and health in general, particularly for dioxin, but the disadvantage of this approach is that there is an inevitable loss of information value when results are extrapolated from one population to another. Furthermore, there is an inherent value in examining the issue of health effects of herbicide exposure in the Vietnam veterans themselves. The question posed by the VA is not the global one, “Is herbicide exposure asso- ciated with adverse health outcomes?”, but rather the narrow one, “Did herbicide exposure affect the health of Vietnam veterans?” Because the herbicide exposures experienced by most Vietnam veterans are imperfectly characterized and presumably low compared with occupationally exposed populations, studies on these veterans can make only limited contributions to answering global questions regarding herbicides and health. However, studies on veterans have much greater potential than other studies to answer the narrower question about their exposure and their health. No other group has the confluence of exposures and exposure circumstances experienced by the Vietnam veterans, so given an adequate tool, it is impor- tant to ask the question in this group.

OCR for page 84
 PROXIMITY-BASED HERBICIDE EXPOSURE ASSESSMENT Improvement in Estimation of Exposure to Herbicides Even with the advantage of studying exposure and health in the popula- tion of interest—Vietnam veterans—the Stellman team’s herbicide exposure assessment model must offer incremental improvement over the methods of exposure assessment used in previous epidemiologic studies of Vietnam veterans if it is to be valuable. The committee concluded that it does offer such improvement. As noted in Chapter 3, many of the studies carried out in Vietnam veterans have relied on “service in Vietnam” as the exposure of interest in examining health outcomes. That approach in effect classifies all those who served in Vietnam as having been exposed to herbicides. Such studies have been able to make only limited contributions to the question of the health effects of exposure to herbicides in Vietnam because many of those considered “exposed” are likely to have had little or no contact with herbicides. Accurate exposure classification is frequently the limiting factor in epidemiologic studies, particularly regarding environmental exposures. In simple terms, errors are introduced when those who are truly exposed are misidentified as unexposed (false negatives) and when those who are truly unexposed are misidentified as exposed (false positives). The shortcomings of an exposure assignment are sometimes explicit, as when service in Viet- nam is used to define “exposure to herbicides” despite the certainty that many false positives will occur (unexposed veterans classified as exposed). In general, such errors, if independent of health outcomes, will tend to dilute any true associations between exposure and health. This will both bias measures of relative risk to be closer to the null value of 1.0 (i.e., no association) and increase the relative variance of estimated effects. Although the actual proportion of Vietnam veterans who had potential for meaningful herbicide exposure is not known, given the temporal and geographic distribution of spraying, it seems likely that exposure was not universal or equal for all those who served. For instance, the Ranch Hand and Army Chemical Corps veterans, who handled the herbicides and pre- sumably had higher exposures on average than other troops, can be studied with confidence as being exposed populations. But the challenge of address- ing the broader population of Vietnam veterans remains. The Stellman team’s model offers the possibility of establishing a gradient of exposure opportunity among the larger population of veterans. In many instances requiring historical estimation of exposure, it is not possible to generate accurate, quantitative indices, and epidemiologists instead pursue a more modest goal of rank-ordering levels or probabilities of exposure. That is, without being able to specify a quantity of exposure, it is possible to place individuals into groups that have higher and lower levels or probabilities of exposure and examine patterns of health across

OCR for page 84
 RECOMMENDATIONS REGARDING EPIDEMIOLOGIC STUDIES those groups to determine whether the more highly exposed have excess risk of adverse health events. Such studies can identify associations even if they are unable to provide information on quantitative dose-response relationships. The exposure assessment hierarchy discussed in Chapter 3 helps to illustrate both the potential contributions and the limitations of the model’s exposure metrics. It is very likely that exposure metrics based on proxim- ity to herbicide spraying provide a better estimate of exposure than does simple presence in Vietnam during certain years. The metrics are also likely to be an improvement over self-reported exposure to herbicides because individuals are unlikely to know when or if an area was sprayed before they arrived and could have mistaken insecticide spraying for herbicide spraying. Approximating target tissue doses in quantitative terms would be an ideal exposure measure, but it is unattainable. Thus, relative to an ideal, the Stellman team’s herbicide assessment model is inherently limited, and even with major refinements will not approximate exposure classifica- tion based on precise individual doses. Still, there is incremental benefit in moving from poor to somewhat better approaches. As a simple illustration of this point, consider the hypothetical results of a study that considers all veterans as exposed versus a study that cor- rectly identifies some fraction of them as not exposed. For purposes of the example, assume that both exposure and disease are classified simply as either occurring or not occurring, and assume a true relative risk of 1.4. Our estimates of relative risk are accurate only if the classification of expo- sure is 100 percent accurate—that is, if all those who were presumed to be exposed truly were exposed, and all those presumed not to be exposed truly were not. With these assumptions, the value of correctly identifying at least some veterans as not exposed becomes clear: If all of those who served in Vietnam are classified as exposed (as many studies have done) when in truth only 20 percent of the troops who served were exposed, the relative risk would be calculated to be only 1.08—far less than the actual value. If exposure classification could be improved such that 80 percent of those designated as exposed actually were exposed (e.g., by applying an expo- sure classification model that more effectively isolates the troops who had potential herbicide exposure) and no one who was exposed was classified as not exposed (a false negative), the relative risk would be calculated as 1.32, which is much closer to the true value in this highly simplified example. Although these results are hypothetical, they illustrate the principle that when the true risk among those exposed is only modestly increased, mis- classification may well hinder identification of elevated risks. Improvements in estimation, even with error still present, would move the observed results in the direction of the value that would be obtained with no misclassification and be more likely to identify any associations between herbicide exposure

OCR for page 84
 PROXIMITY-BASED HERBICIDE EXPOSURE ASSESSMENT and health that are truly present. By converting a dichotomous classifica- tion of exposure (yes/no) into an exposure opportunity score, even with measurement error, there are further gains in accuracy and statistical power. We expect that the exposure opportunity score, even if imprecise, will better identify the highly exposed individuals, improving the specificity1 of the exposure assignment and therefore reducing the exposure misclassification bias. As noted above, improvements in specificity have the largest effect on reducing misclassification and attenuation of the relative risk (Flegal et al., 1986). Although application of the Stellman team’s model to estimate expo- sure can be expected to reduce some aspects of exposure misclassifica- tion, others will inevitably remain. It is possible that the Stellman team’s exposure metrics and software tool could be improved with the incorpo- ration of more detailed chemical fate and transport models. But moving from simple proximity to consideration of fate and transport will involve incorporating either more detailed assumptions or additional data regard- ing spray patterns, drift, meteorology, and ground cover, among others. Furthermore, the current model does not account for individual-level factors such as contact with contaminated plants and soil or consumption of tainted food or water, which may be an additional source of error in the exposure estimate. However, unlike the factors related to the flight path, which affect a whole unit’s potential exposure, these individual-level factors influence how individual-level exposure is distributed around a group (unit) average. Therefore, omitting them should cause little or no bias, but it will generally tend to increase the confidence limits around the exposure estimates. In the absence of individual-level data on behavior or suspected systematic variations in individual behavior between locations or units, any additional modeling of exposure beyond fate and transport, such as different routes of exposure, is unlikely to appreciably change the rank order of exposure estimates. The ultimate benefits of the Stellman team’s exposure metrics depend, of course, on the reliability of the infrastructure around which they are built—the data on the location histories of spraying and military units or personnel. The committee concluded that the infrastructure is sufficiently reliable to justify use of the data, as long as sources of potential exposure misclassification are recognized. In particular, the documentation of her- bicide spraying by fixed-wing aircraft appears to be more complete than that for the spraying by helicopters and ground equipment. In addition, analyses have shown high soil contamination with 2,3,7,8-tetrachloro- dibenzo-p-dioxin (TCDD) at some former military bases (Dwernychuck 1In this context, specificity refers to how well truly unexposed people are classified as unexposed.

OCR for page 84
 RECOMMENDATIONS REGARDING EPIDEMIOLOGIC STUDIES et al., 2002; Dwernychuck, 2005; Hatfield Consultants and Office of the National Steering Committee 33, MNRE, 2007), suggesting spills or other discharges of herbicide that would not be reflected in the spraying data. As a result, service members may have herbicide exposures that cannot be captured by the model’s exposure metrics. Another consideration is that many of the herbicides used in Vietnam, or products similar to them, were also in widespread use at lower concentrations in the United States begin- ning before the Vietnam War, so veterans may have had some exposure to them and any TCDD contaminants before or after their military service. TCDD and other dioxins have also been present in the United States since the Vietnam War as chemical contaminants or byproducts of combustion and manufacturing processes. The exposure sources that are not captured by the model would result in underestimation of herbicide exposures and the potential for false negatives—people designated as not exposed by the model who were in fact exposed to herbicides. If this exposure were randomly distributed across the study population, it would be expected to bias the relative risk toward the null, making true associations of health effects with herbicide sprayed from fixed-wing aircraft more difficult to discern. Interpretation of Studies Using the Model The committee was also concerned with how the results of studies of Vietnam veterans using the Stellman team’s model should be interpreted, given the strengths and limitations of the approach. For established associa- tions between herbicide exposure and health outcomes, finding a positive association in a study that applies this model to Vietnam veterans would provide evidence in support of the veterans having experienced health problems as a result of herbicide exposure. It would also add credibility to the ability of the model to classify exposure. On the other hand, inter- preting a study that failed to find such associations would be much more challenging. In fact, at least four interpretations would be possible: (1) an association is truly present in the population, but the model failed to cap- ture herbicide exposure with sufficient accuracy for the association to be detected; (2) study subjects were accurately assigned to exposure categories, but their herbicide exposure was at levels that are not associated with the health outcome; (3) study subjects’ exposures are associated with such small levels of risk that they are not detectable using epidemiologic methods; or (4) the presumption that herbicides truly cause the health outcome studied is in error. Note that these explanations pertain to the use of the model specifically to evaluate herbicide exposure and the health of Vietnam veterans. A sepa- rate issue is the contribution such studies might make to broader questions

OCR for page 84
0 PROXIMITY-BASED HERBICIDE EXPOSURE ASSESSMENT concerning the effects of herbicides on health. In those cases where there is a sizable body of high-quality epidemiologic research based on large popu- lations with relatively high exposures, the incremental benefit of studies of herbicides and health among veterans would be small, given their relatively low and imperfectly documented herbicide exposure. Even with improve- ments in exposure assignment, their experience is less readily documented than that of some other groups that have already been studied. On the other hand, where the current body of research on specific herbicides and health outcomes is less extensive—which is often the case for health outcomes other than cancer—the application of the Stellman team’s exposure assessment model to Vietnam veterans does hold potential for advancing the larger body of knowledge concerning herbicides and health. That is, where the baseline level of information is lower, studies on Vietnam veterans may advance knowledge despite the known limitations in exposure levels and measurement. The ongoing research by the Stellman team addressing herbicides and amyotrophic lateral sclerosis (ALS) is one example of such a situation, as there has been a limited amount of work done on this topic to date. But any new associations observed in studies carried out using the model in Vietnam veteran populations will need to be confirmed in other herbicide-exposed populations. Applying the model to approximate TCDD exposure is especially prob- lematic. The variable (and unknown) concentrations of TCDD across her- bicide batches make even a perfectly accurate assignment of proximity to spraying of limited value as a TCDD exposure proxy. The CDC (1988, 1989) Agent Orange study, which included only 646 Vietnam veterans, all of whom served no more than 1 year in Vietnam, is the most comprehen- sive study of veterans’ serum TCDD levels performed to date. It found that serum TCDD levels in veterans who served in Vietnam were similar to those of veterans who did not serve there, but the study may not have captured a sufficient number of the most highly exposed soldiers. It is unlikely that meaningful new studies of TCDD levels in serum can be performed in the future because of the long period since the end of the war. This evidence and other studies (e.g., CDC, 1988; Mocarelli et al., 1991; Pirkle and Houk, 1992; Michalek et al., 1996; Sweeney et al., 1997–1998) suggest that the TCDD exposure of many Vietnam veterans was lower than that of other groups that have been studied epidemiologically, such as Air Force Ranch Hand personnel, herbicide production workers, or some of the residents of Seveso, Italy (e.g., those in Zone A). For this reason, studies of Vietnam veterans are not likely to contribute much to the overall knowledge of TCDD and health, regardless of how accurate the methods of exposure classification might become. In summary, epidemiologic studies that use the Stellman team’s expo- sure model have the potential to be informative about the health of Vietnam

OCR for page 84
 RECOMMENDATIONS REGARDING EPIDEMIOLOGIC STUDIES veterans, but they should not necessarily be expected to offer additional insight into the effects of TCDD at the toxicologic level because of inherent limitations in the accuracy of individual exposure assignment. However, the committee does see a possibility that more might be learned about the long- term health effects of exposure to the other herbicides that have not been as extensively studied and that were more consistent in composition. GENERAL CONSIDERATIONS FOR CONDUCTING STUDIES To make the best use of the Stellman team’s exposure assessment model and produce meaningful epidemiologic findings for Vietnam vet- erans, several issues need to be considered that bear on the feasibility of studies and the information value of their results. Representativeness of Vietnam Veterans The committee has assumed that the main purpose of conducting studies using the exposure assessment model is to learn more about if and how Vietnam veterans’ health was affected by herbicide spraying. Given that assumption, then study populations should reflect in a general way the breadth of experience of Vietnam veterans. And because the goal is to produce findings that would be applicable to all those who served, research should encompass the spectrum of experiences of potentially exposed troops. While formal random sampling is not necessary, broad represen- tation is needed. Furthermore, oversampling of those most likely to have been exposed to herbicides would be helpful in enhancing statistical power. Specific candidate populations for study are considered below. Broad Range of Health Outcomes Research addressing proximity to spraying and health outcomes should provide a comprehensive view of the health of veterans and thus address a wide range of health outcomes. Studying outcomes believed to be related to herbicide exposure is of interest, because finding such associations in the veteran population would support the supposition that their exposures were high enough to produce health effects and that the exposures were represented by the model with sufficient accuracy for those effects to be detected. Addressing a wide range of health outcomes also has advantages in terms of efficiency. Depending on the sources of health outcome informa- tion, when information on one health outcome is gathered, it may not take much additional work to gather information on other outcomes. In the most recent review by the Institute of Medicine (IOM) Com- mittee on Veterans and Agent Orange (IOM, 2007), sufficient evidence

OCR for page 84
 PROXIMITY-BASED HERBICIDE EXPOSURE ASSESSMENT for an association2 with exposure to herbicides has been reported for the following conditions: chloracne, chronic lymphocytic leukemia, Hodgkin’s disease, non-Hodgkin’s lymphoma, and soft tissue sarcoma. Limited or sug- gestive evidence of an association was reported for laryngeal cancer; cancer of the lung, bronchus, or trachea; prostate cancer; multiple myeloma; AL amyloidosis; early-onset transient peripheral neuropathy; porphyria cutanea tarda; hypertension; Type 2 diabetes mellitus; and, in the offspring of exposed people, spina bifida. Studies would also be of value for health outcomes that have even weaker evidence for an association with herbicide exposure. Among these are the health outcomes currently classified by IOM committees reviewing the literature on herbicides and health as having “inadequate or insuffi- cient evidence” to determine an association with herbicide exposure. The most recent IOM review (IOM, 2007) includes in this category numerous types of cancer as well as lupus, stroke, and movement disorders, such as Parkinson’s disease. Breast cancer is also in this category, but studying it poses special challenges because of the small number of women who served in Vietnam.3 In selecting health outcomes for study, researchers may want to take into account the ability of the Stellman team’s exposure assessment soft- ware to generate separate exposure metrics for each of the herbicides used in Vietnam, including the arsenical cacodylic acid of Agent Blue and picloram in Agent White. The biennial reviews carried out by IOM contain careful reviews of the toxicologic and epidemiologic literature on the herbi- cides and can offer a source of guidance on outcomes of potential interest. As noted above, the contributions that the model can make to advancing general knowledge of herbicides and health depend to a great extent on the baseline level of understanding. Therefore, research on less extensively studied herbicides and health outcomes holds more promise for advancing 2Conclusions from the IOM reviews of evidence concerning associations between herbicides and health outcomes fall primarily into one of three categories (IOM, 2007): (1) Sufficient evidence of an association means that a positive association has been observed in studies in which chance, bias, and confounding can be ruled out with reasonable confidence. (2) Limited or suggestive evidence of an association means that the body of evidence suggests an associa- tion, but chance, bias, or confounding cannot be confidently ruled out. (3) Inadequate or insufficient evidence to determine an association means that available studies have inconsistent findings or are of insufficient quality or statistical power to support a conclusion regarding the presence of an association. 3The IOM committee that conducted the most recent review of the evidence concerning herbicides and health outcomes was unable to reach consensus on the classification for breast cancer, melanoma, and ischemic heart disease (IOM, 2007). As a default, these three condi- tions were assigned to the category of “inadequate or insufficient evidence” of an association rather than to the category of “limited or suggestive evidence,” which indicates somewhat stronger support for an association.

OCR for page 84
 RECOMMENDATIONS REGARDING EPIDEMIOLOGIC STUDIES general knowledge of the health effects of a herbicide than does additional research on topics already well studied. It may also be worthwhile to consider whether proximity to herbicide spraying can serve as a proxy for combat stress or other factors that might be relevant in studying such outcomes as suicide, posttraumatic stress dis- order, and cardiovascular disease. Completeness and Timeliness of Health Data Studying the relationship between herbicide exposure in Vietnam and health outcomes that are often not fatal, such as diabetes or prostate cancer, might be strengthened by using data on morbidity in addition to mortality data. In these cases it is important to determine health outcomes accu- rately and comprehensively both for the sake of precision (to maximize the number of cases) and for validity (to ensure that the measures of association between herbicide exposure and health are accurate). The issue of statistical power is addressed more fully below, but to the extent that there is under- ascertainment of health outcomes in veterans, it seems likely that the degree of under-ascertainment will be unrelated to exposure opportunity scores, meaning that there will be either no bias or a bias towards the null in rela- tive measures of association. As discussed in Chapter 4, researchers will face challenges in access- ing morbidity data stemming from the dispersed nature of health care in the United States. These challenges are not unique to studies of veterans, however, and both the VA and Medicare systems are leading possibilities for consideration. With Vietnam veterans reaching older ages and their overall likelihood of health problems therefore increasing, studies should aim to ascertain health status up to recent time periods. Reanalysis of health outcome data that were collected for past studies has the appeal of being somewhat sim- pler than initiating a new study: The principal task becomes developing location histories for the study subjects whose health outcome data have already been collected. But, depending on how long ago the study was origi- nally conducted, this approach would miss the opportunity to incorporate additional, more recent data on the study population’s health experience. Statistical Power Proposed studies will need to have adequate statistical power to detect an association between health effects and exposure to herbicides in Viet- nam, if in fact one exists. Statistical power depends on several factors: the risk of error considered acceptable in concluding that no effect is present, the risk of concluding that an effect is present when it is not, the size of

OCR for page 84
 PROXIMITY-BASED HERBICIDE EXPOSURE ASSESSMENT the effect detected relative to the amount of variability in the population, and the size of the sample. The smaller the effect anticipated, the larger the sample size needed to reliably detect it. The committee expects that a very large study population (in the tens of thousands) will be needed to conduct cohort studies of health effects in relation to Vietnam veterans’ exposure to herbicide spraying, particularly for relatively rare outcomes such as soft tissue sarcoma or ALS. For cancer endpoints, where we have the most prior knowledge, three factors suggest that the magnitude of the effect (e.g., the relative risk) is likely to be small in studies of Vietnam veterans: (1) the excess of cancer was relatively small in many previous studies of more highly exposed populations, including those involved in Operation Ranch Hand and even pesticide production workers and applicators (see IOM, 2007); (2) the exposure to herbicides among the majority of Vietnam veterans is likely to be low relative to these previously studied occupational and environmental cohorts; and (3) the degree of accuracy of the model in representing differences in veterans’ exposure is uncertain. For these reasons, studies should be designed to have sufficient power to detect even modest effects. Confounding Another major consideration in designing an epidemiologic study to evaluate possible adverse health effects of herbicide exposures in Vietnam veterans is the need to control for potential confounders, such as tobacco, alcohol, or drug consumption. Some Vietnam veterans, particularly among those with posttraumatic stress disorder, are likely to have had high rates of alcohol or drug use (e.g., McFall et al., 1992). It is also possible that exposure to combat stress might be correlated with exposure to spraying operations. Confounding may be controlled by including potential confounding factors in the statistical analyses, for example, or by restricting analyses to veterans with similar potential for exposure to combat-related stress and with similar socioeconomic backgrounds (Stellman and Stellman, 2003). Analytic Approaches Researchers planning to use the Stellman team’s exposure metrics in an epidemiologic study should recognize and seek ways to mitigate the impact of the many unmeasured determinants of exposure operating at the unit level and affecting individuals within units. The model’s predictions for unit exposures are subject to error, and statistical methods to account for such limitations at the level of unit assignment and individuals within units merit further consideration and refinement. Multilevel (or hierarchical) models

OCR for page 84
0 PROXIMITY-BASED HERBICIDE EXPOSURE ASSESSMENT have only one tour of duty in Vietnam. Another concern is that the study populations may not be large enough to permit the detection of risk for rarer health effects, such as soft tissue sarcomas or NHL. Units Identified as More Highly Exposed Based on the Stellman Team’s Exposure Assessment Model Other potential targets for study that warrant consideration are mem- bers of military units for which location histories have already been tracked and for which the exposure assessment model suggests the likelihood of higher exposure compared with other units. In other words, units with high EOI values could be selected for further investigation. Alternatively, groups with the potential for greater herbicide exposure could be identified by examining the map of sprayed areas in conjunction with approximately known locations of military units. The cancelled CDC study of the 1980s used the approach of selecting units that had served in areas of heavy spraying. However, future researchers who take a similar approach may want to include study participants with more than one tour of duty and take full advantage of the knowledge gained by the Stellman team in their efforts to characterize exposures by service unit. Even with unit location histories, researchers will still need to compile full unit assignment histories for individual subjects and ascertain their health experience. The committee notes that the Air Force Ranch Hand personnel and Army Chemical Corps personnel are two groups of veterans that are not suitable study populations when the Stellman team’s model is to be used, even though their exposures to herbicides are likely to be among the high- est of all veterans. Most of their herbicide exposures were a direct result of duties that required handling or applying herbicides. By contrast, the model is designed to assess the exposure opportunity that would result from unintended proximity to herbicide spraying. Australian and Korean Veterans of Vietnam Approximately 59,200 Australian men have been identified as veterans of military service in Vietnam (Wilson et al., 2005b), and numerous studies have been made of the morbidity and mortality of Australian veterans (e.g., see list in Wilson and Horsley, 2003). A third round of retrospective cohort studies commissioned by the Australian government was completed in 2005 (Wilson et al., 2005a,b,c). Some studies of Australian veterans of the Vietnam War have relied on comparisons with the general Australian population to assess health risks, with the most recent reports showing elevated risks of certain cancers (e.g.,

OCR for page 84
0 RECOMMENDATIONS REGARDING EPIDEMIOLOGIC STUDIES lung and prostate) and liver disease (Wilson et al., 2005b,c). For many other conditions, veterans have lower risks than the general population. In studies specifically of conscripted veterans, comparisons have been made between those who served in Vietnam (approximately 19,200 men) and those who did not (24,700 men) (e.g., Fett et al., 1987; Crane et al., 1997; Wilson et al., 2005a). The most recent of these studies found that Vietnam veterans had significantly higher incidence and mortality rates of lung and pancreatic cancers and a higher incidence of head and neck cancer (Wilson et al., 2005a). The effects of herbicide exposure have been a concern among Aus- tralian veterans, and, as with many studies of U.S. veterans, service in Vietnam has often constituted the exposure criterion in Australian studies. Thus Australian veterans would seem to present another opportunity for application of the Stellman team’s exposure assessment model. A search- able roster of these veterans has been compiled and is accessible through the Internet (http://www.vietnamroll.gov.au/). In communications with Dr. Keith Horsley (2007a,b) of the Australian Institute of Health and Welfare, the committee also learned that records of ground troop locations are poor for land-based Australian Navy and Air Force personnel and generally good for most Australian Army units (the Australian Task Force). According to Dr. Horsley, location information for the Task Force, roughly 38,000 men, is frequently available down to the section level of nine people. The Australian military records are in paper form and stored in the archives of the Australian Department of Defense (Horsley, 2007a). Reconstructing troop locations for Australian service members would seem to pose some of the same challenges as for U.S. troops. The committee did not explore access issues for these records but can see application of the exposure assessment model as a reasonable possibility. Approximately 320,000 South Korean soldiers also participated in the Vietnam War, and more than 92,000 soldiers have reported having health problems related to Agent Orange exposure (Kim et al., 2003). A research group conducted a study of morbidity and mortality of Korean soldiers using EOI values from the Stellman team’s model (Yonsei Medical Institu- tion, 2006).5 To the best of the committee’s knowledge, this is the only completed analysis to date that has used the current version of the Stellman team’s model. Complete information on study methods and findings was not available to the committee at the time this report was written, but the approach is of particular interest as an example of the application of the exposure assessment model. As details become available, the design and results of this study may be of use to other researchers planning to apply the model. 5 Personal communication, S.-W. Yi, Kwandong University, August 17, 2007.

OCR for page 84
0 PROXIMITY-BASED HERBICIDE EXPOSURE ASSESSMENT Residents of Vietnam The Stellman team’s model may also be of use in estimating potential herbicide exposures and studying health outcomes in Vietnamese soldiers and civilians. With the possible exception of the U.S. Air Force personnel participating in Operation Ranch Hand and the Army Chemical Corps, it is likely that the Vietnamese population experienced higher levels of exposure than did U.S. troops because of the greater opportunity of the Vietnamese to be exposed through multiple routes, including inhalation, ingestion of local food and water, and dermal contact with soil and sediments in sprayed areas. Although the possibility of applying the Stellman team’s model to studies of the Vietnamese is of interest and potentially informative about the health effects of herbicides, the committee saw such studies as contribut- ing only indirectly to a better understanding of the health of U.S. veterans, which was the focus of its charge. Committee Observations The committee sees the potential for researchers to pursue studies in a variety of Vietnam veteran populations, while noting that each population presents different trade-offs in terms of magnitude of effort, the avail- ability of data on locations of service in Vietnam, representation of those more likely to have been exposed, and access to data on health outcomes. Researchers will need to direct their efforts based on the particulars of their research questions and on available resources. The committee believes that the most promising studies are either those that use groups where higher exposure to herbicides appears likely based on analysis of existing unit or troop movement information, or case-control studies nested in large cohorts already assembled, possibly including the VA roster of 3 million Vietnam veterans. ONGOING VA WORK TO APPLY THE EXPOSURE ASSESSMENT MODEL Dr. Han Kang and colleagues at VA have several projects under way to explore use of the Stellman team’s model (VA, 2007). The work “aims to evaluate the validity and utility of this model . . . using the databases already collected from previous health studies on Vietnam veterans.” They have described three specific areas of inquiry: 1. Does the model generate internally consistent index exposure scores that are consistent with variation in military mission? To test this, five military units stationed in Vietnam during the height of Agent Orange

OCR for page 84
0 RECOMMENDATIONS REGARDING EPIDEMIOLOGIC STUDIES spraying in 1968 will be tracked for geographic location in Vietnam for one year, using the databases already collected by Dr. Stellman and her team. 2. Does the model demonstrate a positive association between the index exposure scores and the prevalence or incidence of health outcomes that are considered truly associated with herbicides, such as soft tissue sarcoma, non-Hodgkin’s lymphoma, and Hodgkin’s disease? To test this, veteran study subjects from three existing case-control study databases and three Vietnam veterans’ mortality study databases will be used. Geographic locations and time of service in Vietnam will be tracked for the cancer cases and the controls and then applied to the model. 3. Does application of the model to existing epidemiologic study databases generate additional information on the health effects of exposure to herbicide in Vietnam beyond what is already known? How much time and resources are required to conduct a health study of Vietnam veterans using the model? The model will be applied to three existing databases: the Agent Orange Registry of 409,000 Vietnam veterans, Marine Corps Vietnam Veterans, and Women Vietnam Veterans. When Dr. Kang and his colleagues met with the committee in spring 2007, they reported that efforts were under way to construct location his- tories for the veterans who would be included in the reanalyses. Calculation of exposure metrics using the Stellman team’s model had not yet begun. The committee sees the value of VA’s work in investigating the feasibil- ity of obtaining the needed model inputs and generating exposure scores. Because of its unique insights into military records and databases, the VA research group has the potential for achieving success that can be docu- mented and shared with other researchers beyond VA. Completion of this work should thus prove worthwhile and informative for methodologic pur- poses in addition to generating epidemiologic findings that relate estimated exposure opportunity to particular health outcomes. Nevertheless, as has previously been observed (IOM, 2003), such studies are of limited value in moving toward the stated goal of validat- ing the model. If positive, these studies would lend support to the model’s potential use. If the results are negative, though, the findings will be dif- ficult to interpret. They cannot disprove the value of the proximity model because the levels of exposure may be lower than would cause adverse health effects or because the study’s power may be insufficient to address the question adequately. Although the work that the VA researchers are conducting will make an important contribution, its scope falls short of the complete array of work that the committee believes would be appropriate. The committee understands that not all the efforts it proposes can be undertaken at once, but it does have some specific concerns. In the morbidity studies that are

OCR for page 84
0 PROXIMITY-BASED HERBICIDE EXPOSURE ASSESSMENT to be revisited, the range of health outcomes is limited, the ascertainment of those health outcomes ended for the most recent of the studies in 1985,6 and the health data are limited to events identified among veterans who participated in the VA health care system. There do not appear to be plans to conduct sensitivity analyses or explore ways in which the model might be extended. In addition, there is no indication of plans to create a resource that other investigators, including those outside VA, could expand upon. For these reasons, the committee sees the VA work as insufficient by itself to constitute the best use of the exposure assessment model. CONTRIBUTION TO THE ONGOING IOM REVIEWS OF THE ASSOCIATION BETWEEN HERBICIDE EXPOSURE AND HEALTH OUTCOMES The biennial reviews carried out by IOM committees to examine poten- tial associations between herbicide exposure and health effects have had to rely in large measure on studies in populations other than Vietnam veterans to reach conclusions regarding the strength of such associations. Applica- tion of the herbicide exposure assessment model in epidemiologic studies of Vietnam veterans has the potential to contribute to the body of work that future committees can evaluate. No single study should be expected to pro- vide definitive evidence, but results derived from studies of the population of interest will be a welcome addition. The committee reiterates, however, that because of the potential for exposure misclassification and concerns about statistical power, studies of Vietnam veterans that make use of the proximity-based exposure surrogates must always be viewed as informative rather than definitive. CONCLUSIONS The committee reached the following specific conclusions regarding the use of the Stellman team’s herbicide exposure assessment model in epidemiologic studies: 1. Epidemiologic studies of health outcomes among Vietnam veterans that use the exposure opportunity index are capable of characterizing the health of veterans in relation to their proximity to herbicide spraying. Such studies would provide an improvement in evaluating potential effects of herbicide exposure compared to studies based on Vietnam service in gen- 6 Case ascertainment for the case-control studies carried out by VA extended to the following years: NHL, 1985 (Dalager et al., 1991); Hodgkin’s disease, 1985 (Dalager et al., 1995); and soft tissue sarcoma, 1983 (Kang et al., 1986).

OCR for page 84
0 RECOMMENDATIONS REGARDING EPIDEMIOLOGIC STUDIES eral. This improvement may permit observation of associations between herbicide exposure and health effects in the Vietnam veteran population that were not identifiable in previous studies. 2. The ongoing efforts by the VA investigators to apply the model in existing case-control study populations are useful for characterizing the logistical challenges and the magnitude of effort needed to apply the model, but they are limited and insufficiently accessible to future researchers and the broader research community and so do not constitute, in isolation, the best use of the exposure assessment model. 3. The most promising study designs for applications of the model would be (1) to selectively sample units based on exposure potential, gath- ering relevant information from unit records, individual service records, and sources of individual health outcomes data; or (2) to build on large cohorts already assembled and pursue nested case-control studies of outcomes of interest with linkage to units, unit locations, and exposure potential. 4. Efforts to validate the model solely by examining health outcomes believed to be related to herbicide exposure are of limited value. If posi- tive, such studies would add support to the model’s potential value, but, if negative, the model’s value would not be disproved because the levels of exposure may be lower than would cause adverse health effects or the study’s power may be insufficient to address the question adequately. RECOMMENDATIONS The committee’s conclusions from its consideration of the Stellman team’s herbicide exposure assessment model led it to make the following recommendations: 1. VA should sponsor epidemiologic studies of Vietnam veterans that take into account the criteria below regarding the appropriate characteristics of informative research on herbicide exposure and health outcomes in this population. VA should draw on the criteria as the basis for developing a request for proposals. Specifically, to make the best use of the exposure assessment model, epidemiologic studies of Vietnam veterans should have the follow- ing characteristics: a. The study population should be broadly representative of Viet- nam veterans, with care taken to include sufficient numbers of study participants with relatively higher exposure. b. A broad range of health outcomes should be considered, not just those that are suspected of being related to herbicide expo-

OCR for page 84
0 PROXIMITY-BASED HERBICIDE EXPOSURE ASSESSMENT sure. Where feasible, morbidity should be studied in addition to mortality. c. The health data should be as complete and up-to-date as possible. d. The study should have sufficient statistical power to address the range of health outcomes of concern. e. To isolate the effects of herbicide exposure, potential con- founding factors need to be carefully addressed in the study design or the analytic approach. f. Analyses should be conducted to evaluate how sensitive the estimated associations between exposure opportunity and health outcomes are to the uncertainty in the exposure oppor- tunity metrics and to varying approaches to estimating her- bicide exposure, possibly including alternative approaches to exposure assignment as discussed in Chapter 3. g. Opportunities to conduct research using the exposure assess- ment model should be open to investigators beyond the VA sys- tem to allow for the benefits of engaging the broader research community and to enhance public acceptance and credibility. 2. In support of the recommended epidemiologic studies, VA should work with DoD and NARA to facilitate health research uses of military records that are sub- • ject to access barriers arising from privacy laws, and arrange for assistance from DoD and NARA staff with appro- • priate expertise to aid researchers in the location and interpre- tation of military records for health research uses. REFERENCES Barrett, D. H. 2007. Letter to L. Joellenbeck, with accompanying materials. Atlanta, GA, June 22. Breslin, P., H. K. Kang, Y. Lee, V. Burt, and B. Shepard. 1988. Proportionate mortality study of US Army and US Marine Corps veterans of the Vietnam War. Journal of Occupational Medicine 30(5):412–419. Bullman, T. A., K. K. Watanabe, and H. K. Kang. 1994. Risk of testicular cancer associated with surrogate measures of Agent Orange exposure among Vietnam veterans on the Agent Orange Registry. Annals of Epidemiology 4:11–16. CDC (Centers for Disease Control and Prevention). 1985. Exposure assessment for the Agent Orange study. Interim Report Number 2. Typescript. CDC. 1988. Serum 2,3,7,8-tetrachlorodibenzo-p-dioxin levels in U.S. Army Vietnam-era vet- erans. The Centers for Disease Control Veterans Health Studies. Journal of the American Medical Association 260(9):1249–1254.

OCR for page 84
0 RECOMMENDATIONS REGARDING EPIDEMIOLOGIC STUDIES CDC. 1989. Comparison of serum levels of ,,,-tetrachlorodibenzo-p-dioxin with indirect estimates of Agent Orange exposure among Vietnam veterans: Final report. Atlanta, GA: Agent Orange Projects, Center for Environmental Health and Injury Control. Crane, P. J., D. L. Barnard, K. W. Horsley, M. A. Adena. 1997. Mortality of national service Vietnam veterans: A report of the  retrospective cohort study of Australian Vietnam veterans. Canberra, Australia: Department of Veterans’ Affairs. Dalager, N. A., H. K. Kang, V. L. Burt, and L. Weatherbee. 1991. Non-Hodgkin’s lymphoma among Vietnam veterans. Journal of Occupational Medicine 33(7):774–779. Dalager, N. A., H. K. Kang, V. L. Burt, and L. Weatherbee. 1995. Hodgkin’s disease and Vietnam service. Annals of Epidemiology 5(5):400–406. Dominici, F., S. L. Zeger, and J. M. Samet. 2000. A measurement error model for time-series studies of air pollution and mortality. Biostatistics 1(2):157–175. Dwernychuk, L. W. 2005. Dioxin hot spots in Vietnam. Chemosphere 60:998–999. Dwernychuk, L. W., H. D. Cau, C. T. Hatfield, T. G. Boivin, T. M. Hung, P. T. Dung, and N. D. Thai. 2002. Dioxin reservoirs in southern Viet Nam—A legacy of Agent Orange. Chemosphere 47(2):117–137. Fett, M. J., J. R. Nairn, D. M. Cobbin, and M. A. Adena. 1987. Mortality among Australian conscripts of the Vietnam conflict era. II. Causes of death. American Journal of Epide- miology 125:878–884. Flegal, K. M., C. Brownie, and J. D. Haas. 1986. The effects of exposure misclassification on estimates of relative risk. American Journal of Epidemiology 123(4):736–751. Greenland, S. 2000. Principles of multilevel modeling. International Journal of Epidemiology 29:158–167. Hatfield Consultants and Office of the National Steering Committee 33, MNRE (Ministry of Natural Resources and Environment, Vietnam). 2007. Assessment of dioxin contamina- tion in the environment and human population in the vicinity of Da Nang Airbase, Viet Nam: Final report. West Vancouver, British Columbia, Canada. Horsley, K. 2007a. Re: Inquiry regarding Australian veterans and Agent Orange studies. E-mail to L. Joellenbeck, June 4. Horsley, K. 2007b. Re: Inquiry regarding Australian veterans and Agent Orange studies. E-mail to L. Joellenbeck, June 28. IOM (Institute of Medicine). 1994. Veterans and Agent Orange: Health effects of herbicides used in Vietnam. Washington, DC: National Academy Press. IOM. 2003. Characterizing exposure of veterans to Agent Orange and other herbicides used in Vietnam: Interim findings and recommendations. Washington, DC: The National Academies Press. IOM. 2007. Veterans and Agent Orange: Update 00. Washington, DC: The National Academies Press. Kang, H. K. 2007a. Data resources within VA for an epidemiological study of Vietnam veterans. PowerPoint presentation to the IOM Committee on Making Best Use of the Agent Orange Reconstruction Model, Meeting 2, April 30–May 1, Washington, DC. Kang, H. K. 2007b. Vietnam roster veterans: Demographic/military service characteristics. Unpublished document submitted to the IOM Committee on Making the Best Use of the Agent Orange Reconstruction Model, July 11. Kang, H. K., L. Weatherbee, P. Breslin, Y. Lee, and B. Shepard. 1986. Soft tissue sarcomas and military service in Vietnam: A case comparison group analysis of hospital patients. Journal of Occupational and Environmental Medicine 28(12):1215–1218. Kim, H. A., E. M. Kim, Y. C. Park, J. Y. Yu, S. K. Hong, S. H. Jeon, K. L. Park, S. J. Hur, and Y. Heo. 2003. Immunotoxicological effects of Agent Orange exposure to the Vietnam War Korean veterans. Industrial Health 41:158–166.

OCR for page 84
0 PROXIMITY-BASED HERBICIDE EXPOSURE ASSESSMENT Mahan, C. M., T. A. Bullman, and H. K. Kang. 1997. A case-control study of lung cancer among Vietnam veterans. Journal of Occupational and Environmental Medicine 39(8):740–747. McFall, M. E., M. E. Mackay, and D. M. Donovan. 1992. Combat-related posttraumatic stress disorder and severity of substance abuse in Vietnam veterans. Journal of Studies on Alcohol 53(4):357–363. Michalek, J. E., S. P. Caudill, and R. C. Tripathi. 1996. Pharmacokinetics of TCDD in veterans of Operation Ranch Hand: 10-year follow-up. Journal of Toxicology and Environmental Health 47:209–220. Mocarelli, P., L. L. Needham, A. Marocchi, D. G. Patterson, Jr., P. Brambilla, P. M. Gerthoux, L. Meazza, and V. Carreri. 1991. Serum concentrations of 2,3,7,8-tetrachlorodibenzo-p- dioxin and test results from selected residents of Seveso, Italy. Journal of Toxicology and Environmental Health 32:357–366. Peng, R. D., F. Dominici, and S. L. Zeger. 2006. Reproducible epidemiologic research. Ameri- can Journal of Epidemiology 163(9):783–789. Pirkle, J. L., and V. N. Houk. 1992. Use of epidemiologic studies to assess human risk from exposure to 2,3,7,8-tetrachlorodibenzo-p-dioxin. Chemosphere 25(7–10):1109–1115. Stayner, L., M. Vrijheid, E. Cardis, D. O. Stram, I. Deltour, S. J. Gilbert, and G. Howe. 2007. A Monte Carlo maximum likelihood method for estimating uncertainty arising from shared errors in exposures in epidemiological studies of nuclear workers. Radiation Research 168(6):757–763. Stellman, J. M. 2007a. A data resource for the health and environmental consequences of the Vietnam War. Unpublished document submitted to the IOM Committee on Making the Best Use of the Agent Orange Reconstruction Model, July 31. Stellman, J. M. 2007b. Responses to IOM 00. Unpublished document submitted to the IOM Committee on Making the Best Use of the Agent Orange Reconstruction Model, September 14. Stellman, J. M. 2007c. Separate modeling and reconstruction issues. PowerPoint presentation to the IOM Committee on Making Best Use of the Agent Orange Reconstruction Model, Meeting 3, June 13–14, Washington, DC. Stellman, J. M., and S. D. Stellman. 2003. Contractor’s final report: Characterizing exposure of veterans to Agent Orange and other herbicides in Vietnam. Submitted to the National Academy of Sciences, Institute of Medicine, in fulfillment of Subcontract VA-5124-98- 0019, June 30, 2003. Sweeney, M. H., G. M. Calvert, G. A. Egeland, M. A. Fingerhut, W. E. Halperin, and L. A. Piacitelli. 1997–1998. Review and update of the results of the NIOSH medical study of workers exposed to chemicals contaminated with 2,3,7,8-tetrachlorodibenzodioxin. Teratogenesis, Carcinogenesis, and Mutagenesis 17(4–5):241–247. VA (U.S. Department of Veterans Affairs). 2007. Evaluation of Dr. Stellman’s herbicide exposure reconstruction model. http://www.va.gov/wriisc-dc/research/studies_ongoing. asp#Stellman (accessed February 22, 2007). Watanabe, K. K., and H. K. Kang. 1995. Military service in Vietnam and the risk of death from trauma and selected cancers. Annals of Epidemiology 5:407–412. Watanabe, K. K., and H. K. Kang. 1996. Mortality patterns among Vietnam veterans. Journal of Occupational and Environmental Medicine 38(3):272–278. Wilson, E. J., and K. W. A. Horsley. 2003. Health effects of Vietnam service. ADF Health 4:59–65. Wilson, E. J., K. W. Horsley, and R. van der Hoek. 2005a. Australian National Service Viet- nam Veterans Mortality and Cancer Incidence Study 00. Canberra, Australia: Depart- ment of Veterans’ Affairs.

OCR for page 84
 RECOMMENDATIONS REGARDING EPIDEMIOLOGIC STUDIES Wilson, E. J., K. W. Horsley, and R. van der Hoek. 2005b. Australian Vietnam Veterans Mor- tality Study 00. Canberra, Australia: Department of Veterans’ Affairs. Wilson, E. J., K. W. Horsley, and R. van der Hoek. 2005c. Cancer Incidence in Australian Vietnam Veterans Study 00. Canberra, Australia: Department of Veterans’ Affairs. Yonsei Medical Institution. 2006. [Final report: Dynamic study for defoliant (Agent Orange) damages.] Report to the Ministry of Patriots and Veterans Affairs, Republic of Korea.

OCR for page 84