The assessment of human exposure continues to be a key element in addressing two of the charges that guide the work of this committee. This chapter first presents background information on the military use of herbicides in Vietnam from 1961 to 1971 with a review of our knowledge about the exposures of those who served in Vietnam and of the Vietnamese population to the herbicides and to the contaminant 2,3,7,8-tetrachlorodibenzo-p-dioxin, which is referred to in this report as TCDD (and commonly referred to as dioxin) and is the most toxic congener of the tetrachlorodibenzo-p-dioxins. Two modeling approaches to estimating exposure of ground troops to herbicides are presented. The fact that they lead to quite different conclusions demonstrates the difficulties of assessing exposure in the complex environment that characterized Vietnam during the period of interest. The application of one of these models in an epidemiologic context is discussed in considerable detail. The chapter concludes by reviewing several key methodologic issues in human population studies: disease latency, possible misclassification based on exposure, and the exposure specificity required for the scientific evaluation of study results.
The exposure of human populations can be assessed in a number of ways, including the use of historical information, questionnaires and interviews, measurements in environmental media, and measurements in biologic specimens. Researchers often rely on a mixture of qualitative and quantitative information to derive such estimates (Armstrong et al., 1994; Checkoway et al., 2004). The most basic approach compares members of a presumably exposed group with the general population or with a nonexposed group; this method of classification offers simplicity and ease of interpretation. A more refined method assigns each study subject to an exposure category—such as high, medium, or low exposure—and
calculates the disease risk for each group separately and then compares that with the risk for a reference or non-exposed group; this method can identify the presence or absence of an exposure–response trend. In some cases, more detailed information is available for quantitative exposure estimates that can be used to construct what are sometimes called exposure metrics. The metrics integrate quantitative estimates of exposure intensity (such as the chemical concentration in air or the extent of skin contact) with exposure duration to produce an estimate of cumulative exposure. Exposure can also be assessed by measuring chemicals and their metabolites in human tissues. Such biologic markers of exposure integrate absorption from all exposure routes, but their interpretation requires knowledge of pharmacokinetic processes. All of those exposure-assessment approaches have been used in studies of Vietnam veterans.
The military use of herbicides in Vietnam took place from 1962 through 1971. Tests conducted in the United States and elsewhere that were designed to evaluate defoliation efficacy were used to select specific herbicides (IOM, 1994; Young and Newton, 2004). Four compounds were used in the herbicide formulations in Vietnam: 2,4-dichlorophenoxyacetic acid (2,4-D), 2,4,5-trichlorophen-oxyacetic acid (2,4,5-T), 4-amino-3,5,6-trichloropicolinic acid (picloram), and dimethylarsinic acid (DMA or cacodylic acid). The chemical structures of those compounds are presented in Chapter 2 (see Figure 2-1). These herbicides were used to defoliate inland hardwood forests, coastal mangrove forests, cultivated lands, and zones around military bases. A National Resource Council committee estimated the amount of herbicides sprayed from helicopters and other aircraft by using records gathered from August 1965 through February 1971 (NRC, 1974). That committee calculated that about 18 million gallons (about 69 million liters) of herbicide were sprayed over about 3.6 million acres (about 1.5 million hectares) in Vietnam during that period. The amount of herbicides sprayed on the ground to defoliate the perimeters of base camps and fire bases and the amount sprayed by Navy boats along river banks were not estimated.
A revised analysis of spray activities and of the exposure potential of troops emerged from a study overseen by a committee of the Institute of Medicine (IOM, 1997, 2003b,c). That work yielded new estimates of the amounts of military herbicides used in Vietnam from 1961 through 1971 (Stellman et al., 2003a). The investigators reanalyzed the original data sources that were used to develop herbicide-use estimates in the 1970s and identified errors that inappropriately removed spraying missions from the dataset. They also added new data on spraying missions that took place before 1965. Finally, a comparison of procurement records with spraying records found errors that suggested that additional spraying had taken place but had gone unrecorded at the time. The new analyses led to a revision of the estimates of the amounts of the agents applied, as indicated
|Code Name||Chemical Constituentsa||Concentration of Active Ingredienta||Years Useda||Amount Sprayed|
|VAO Estimateb||Revised Estimatea|
|Pink||60% n-butyl ester, 40% isobutyl ester of 2,4,5-T||961–1,081 g/L acid equivalent||1961, 1965||464,817 L (122,792 gal)||50,312 L sprayed; 413,852 L additional on procurement records|
|Green||n-butyl ester of 2,4,5-T||—||1961, 1965||31,071 L (8,208 gal)||31,026 L on procurement records|
|Purple||50% n-butyl ester of 2,4-D, 30% n-butyl ester of 2,4,5-T, 20% isobutyl ester of 2,4,5-T||1,033 g/L acid equivalent||1962–1965||548,883 L (145,000 gal)||1,892,733 L|
|Orange||50% n-butyl ester of 2,4-D, 50% n-butyl ester of 2,4,5-T||1,033 g/L acid equivalent||1965–1970||42,629,013 L (11,261,429 gal)||45,677,937 L (could include Agent Orange II)|
|Orange II||50% n-butyl ester of 2,4-D, 50% isooctyl ester of 2,4,5-T||910 g/L acid equivalent||After 1968||—||Unknown; at least 3,591,000 L shipped|
|White||Acid weight basis: 21.2% triisopropanolamine salts of 2,4-D, 5.7% picloram||By acid weight, 240 g/L 2,4-D, 65 g/L picloram||1966–1971||19,860,108 L (5,246,502 gal||20,556,525 L )|
|Blue powder||Cacodylic acid (dimethylarsinic acid) sodium cacodylate||Acid, 65% active ingredient; salt, 70% active ingredient||1962–1964||—||25,650 L|
|Blue aqueous solution||21% sodium cacodylate + cacodylic acid to yield at least 26% total acid equivalent by weight||Acid weight, 360 g/L||1964–1971||4,255,952 L (1,124,307 gal||4,715,731 L )|
|Total, all formulations||—||—||—||67,789,844 L (17,908,238 gal)||76,954,766 L (including procured)|
in Table 3-1. The new research effort estimated that about 77 million liters were applied, about 9 million liters more than the previous estimate.
Herbicides were identified by the color of a band on 55-gallon shipping containers and were called Agent Pink, Agent Green, Agent Purple, Agent Orange, Agent White, and Agent Blue. Agent Green and Agent Pink were used in 1961 and 1965, and Agent Purple in 1962–1965. Agent Orange was used in 1965–1970, and a slightly different formulation (Agent Orange II) probably was used after 1968. Agent White was used in 1966–1971. Agent Blue was used in powder form in 1962–1964 and as a liquid in 1964–1971. Agent Pink, Agent Green, Agent Purple, Agent Orange, and Agent Orange II all contained 2,4,5-T and were contaminated to some extent with TCDD. Agent White contained 2,4-D and picloram. Agent Blue (powder and liquid) contained cacodylic acid. The chlorinated phenoxy acids 2,4-D and 2,4,5-T persist in soil for only a few weeks; picloram is much more stable, persisting in soil for years; and cacodylic acid is nonvolatile and stable in sunlight (NRC, 1974). More details on the herbicides used are presented in the initial IOM report, Veterans and Agent Orange: Health Effects of Herbicides Used in Vietnam,1 referred to as VAO (IOM, 1994).
TCDD is formed during the manufacture of 2,4,5-T in the following manner: trichlorophenol (2,4,5-TCP), the precursor for its synthesis, is formed by the reaction of tetrachlorobenzene and sodium hydroxide (see Figure 3-1a); 2,4,5-T is formed when 2,4,5-TCP reacts with chloroacetic acid (see Figure 3-1b); small amounts of TCDD are formed as a byproduct of the intended main reaction (see Figure 3-1b) when a molecule of 2,4,5-TCP reacts with the tetrachlorobenzene stock (see Figure 3-1c) instead of with chloroacetic acid. In each step in the reaction, a chlorine atom is replaced with an oxygen atom, and this leads to the final TCDD molecule (NRC, 1974). In the class of compounds known as polychlorinated dibenzo-p-dioxins (PCDDs), 75 congeners can occur, depending on the number and placement of the chlorine atoms. Cochrane et al. (1982) noted that TCDD had been found in pre-1970 samples of 2,4,5-TCP. Other PCDDs—2,7-dichloro-dibenzo-p-dioxin and 1,3,6,8-tetrachloro-dibenzo-p-dioxin—were measured in the same samples. The concentration of TCDD in any given lot of 2,4,5-T depended on the manufacturing process (FAO/UNEP, 2009; Young et al., 1976).
The manufacture of 2,4-D is a different process: Its synthesis is based on dichlorophenol, a molecule formed from the reaction of phenol with chlorine
1Despite loose usage of “Agent Orange” by many people, in numerous publications, and even in the title of this series, this committee uses “herbicides” to refer to the full range of herbicide exposures experienced in Vietnam, while “Agent Orange” is reserved for a specific one of the mixtures sprayed in Vietnam.
a. Trichlorophenol, the precursor for the synthesis of 2,4,5-T, is formed by the reaction of tetrachlorobenzene and sodium hydroxide (NaOH).
b. The herbicide 2,4,5-T is formed when a reactive form of trichlorophenol (2,4,5-trichlo-rophenoxide) reacts with chloroacetic acid.
c. TCDD is formed when a molecule of trichlorophenol reacts with its own precursor, tetrachlorobenzene. Two intermediate steps are shown in this diagram. At each step, an oxygen–carbon bond forms as a chlorine atom is released. This reaction does not occur in the synthesis of 2,4-D because precursors with adjacent chlorines are not used in its production.
FIGURE 3-1 TCDD formation during 2,4,5-T production.
(NZIC, 2009). Neither tetrachlorobenzene nor trichlorophenol is formed during this reaction, so TCDD is not normally a byproduct of the manufacturing process. However, other, less toxic PCDDs have been detected in pre-1970 commercial-grade 2,4-D (Cochrane et al., 1982; Rappe et al., 1978; Tosine, 1983). Cochrane et al. (1982) found multiple PCDDs in isooctyl ester, mixed butyl ester, and dimethylamine salt samples of 2,4-D. It has also been noted that a cross-contamination of 2,4-D with 2,3,7,8-TCDD occurred in the operations of at least one major manufacturer (Lilienfeld and Gallo, 1989).
TCDD concentrations in individual herbicide shipments were not recorded, but they were known to vary from batch to batch and between manufacturers. TCDD concentrations in stocks of Agent Orange remaining after the conflict, which either had been returned from South Vietnam or had been procured but not shipped, ranged from less than 0.05 ppm to almost 50 ppm and averaged 2–3 ppm in two sets of samples (NRC, 1974; Young et al., 1978). In 1974, domestic manufacturing standards for 2,4,5-T required that TCDD not be present at over 0.05 ppm (NRC, 1974).
Originally, data from Young and Gough were used to estimate the amount of TCDD in the various herbicide formulations (Gough, 1986; Young, 1992; Young et al., 1978). Young et al. (1978) estimated that Agent Green, Agent Pink, and Agent Purple—which all were used early in the program (through 1965)—contained 16 times the mean TCDD content of the Agent Orange formulations used in 1965–1970, which had mean TCDD concentrations estimated at 2 ppm. Gough (1986) estimated that about 167 kg of TCDD were sprayed in Vietnam from 1962 to 1970.
Later analysis by researchers at Columbia University benefited from access to military spray records that had not been available earlier, and it resulted in substantial revisions of the estimates (Stellman et al., 2003a). The investigators were able to incorporate newly found data on spraying in the early period of the war (1961–1965) and to document that larger volumes of TCDD-containing herbicides were used in Vietnam than had been estimated previously. They also found the earlier estimates of TCDD contamination in the herbicide formulations to be low, noting that the original estimates were based on samples whose concentrations were at the lower end of the distribution. The researchers concluded that the mean TCDD concentration in Agent Orange was closer to 13 ppm than to the earlier estimate of 3 ppm. They therefore proposed 366 kg of TCDD as a plausible estimate of the total amount of TCDD applied in Vietnam during 1961–1971.
Determining the exposures of US military personnel who served in Vietnam has been perhaps the greatest challenge in the study of the health effects associated with herbicides and TCDD. Some military personnel stationed in cities or
on large bases may have received little or no herbicide exposure, whereas troops who moved through defoliated areas soon after treatment may have been exposed through soil contact, drinking water, or bathing. In most cases it is not possible to make reliable estimates of the magnitude and duration of such exposures because of the lack of contemporaneous chemical measurements, the lack of a full understanding of the movement and behavior of the defoliants in the environment, and the lack of records of individual behaviors and locations. Consequently, most studies have focused on populations that had well-defined tasks that brought them into contact with the agents. It is believed that the subjects of those studies, primarily Air Force personnel involved in fixed-wing aircraft spraying activities (often referred to as Operation Ranch Hand [ORH]) and members of the US Army Chemical Corps (ACC), may have had among the highest exposures. As described below, the exposures of ground troops are difficult to define, so this group has not been studied as intensively. As illustrated by Figure 1-1 in Chapter 1, the median TCDD levels in veterans who had worked in Operation Ranch Hand were higher than those measured in their own comparison group or in ground troops, which both had median levels in the unitary ppt range of contemporaneous background levels, but about an order of magnitude less than herbicide production workers, who in turn had levels about two orders of magnitude less than individuals who resided near the site of the industrial explosion in Seveso, Italy (Pirkle et al., 1995).
In accordance with Congress’s mandated presumption of herbicide exposure of all Vietnam veterans, VAO committees have treated Vietnam-veteran status as a proxy for some herbicide exposure when more specific exposure information is not available.
Exposure of Herbicide Handlers
Military personnel who came into direct contact with the herbicidal chemicals through mixing, loading, spraying, and clean-up activities had relatively high exposures to them. The US Environmental Protection Agency refers to such personnel as pesticide handlers and provides special guidance for preventing or minimizing their exposure during those activities in its worker-protection standard for pesticides (EPA, 1992). The number of US military personnel who handled herbicides directly is not known precisely, but two groups have been identified as high-risk subpopulations among veterans: Air Force personnel involved in ORH and members of the ACC who used hand-operated equipment and helicopters to conduct smaller-scale operations, including defoliation around special-forces camps; clearing of the perimeters of airfields, depots, and other bases; and small-scale crop destruction (NRC, 1980; Thomas and Kang, 1990; Warren, 1968). Additional units and individuals handled or sprayed herbicides around bases or lines of communication; for example, Navy river patrols were reported to have used herbicides to clear inland waterways, and engineering personnel used
herbicides to remove underbrush and dense growth in constructing fire-support bases. The latter groups have not been the subject of epidemiologic studies. The herbicides used in Vietnam were not thought to present an important human health hazard at the time, so few precautions were taken to prevent the exposure of personnel (GAO, 1978, 1979); that is, military personnel did not typically use chemical-protective gloves, coveralls, or protective aprons, so substantial skin exposure almost certainly occurred in these populations in addition to exposure by inhalation and incidental ingestion (such as by hand-to-mouth contact).
The Air Force personnel who participated in ORH were the first Vietnam-veteran population to receive special attention with regard to herbicide exposure. In the Air Force Health Study (AFHS), job and work history, biomarkers, and the health outcomes of members of this Ranch Hand cohort were contrasted with Air Force personnel who had served elsewhere in Southeast Asia during the Vietnam era. The AFHS began in 1979 (IOM, 2006a). The exposure index that was initially proposed relied on military spray records for the TCDD-containing herbicides (Agent Orange, Agent Purple, Agent Pink, and Agent Green); these records also helped identify the members of the cohort. The subjects were further characterized by military occupation, and the exposure in the cohort and the comparison group was evaluated by measurements of TCDD in blood (serum) samples drawn in 1987 or later. A general increase in serum TCDD was detected in people whose jobs involved more frequent handling of herbicides, but there was no clear demarcation between the distributions of serum TCDD concentrations in the Ranch Hand subjects and those in the comparison group (AFHS, 1991a). Several methods for estimating the herbicide exposure of members of the cohort were developed on the basis of questionnaires, and they focused on such factors as the number of days of skin exposure, the percentage of skin area exposed, and the concentration of TCDD in the different herbicidal formulations (Michalek et al., 1995). Analyses of the AFHS data have typically relied on serum TCDD concentration as the primary exposure metric for epidemiologic classification (Kern et al., 2004; Michalek et al., 2001a, 2003; Pavuk et al., 2003). Pavuk et al. (2014) examined the serum concentrations of several other dioxins and dioxin-like compounds (i.e., PCDDs, polychlorinated dibenzofuran [PCDFs], and polychlorinated biphenyls [PCBs]) in serum samples gathered in 2002 from 777 ORH participants and 1,173 Air Force veterans in the comparison group. While the median TCDD levels were more than twice as high in the ORH subjects as in the comparison veterans (5.0 and 2.2 pg/g, respectively), no substantial differences were found between these groups for the other compounds. When contrasted with the serum levels measured in men in their age range during the 2001–2002 cycle of NHANES, the concentrations of the ORH subjects were similar except for TCDD, which demonstrated the specificity of the dioxin exposure experienced from contact in Vietnam with military herbicides. (Although serum TCDD measurements in 2002 were still sufficiently elevated to distinguish exposed and unexposed veterans at the group level, with the passage of several
more TCDD half-lives of about 7 years, newly drawn serum samples will cease to be useful metrics for assessing health outcomes in surviving Vietnam veterans, occupational cohorts, or Seveso residents. Factors influencing TCDD’s half-life are discussed in Table 4-1, which documents the variability in TCDD half-life observed in various circumstances.)
Members of the ACC performed herbicide-spraying operations on the ground and by helicopter and were thereby involved in the direct handling and distribution of Agent Orange and other herbicides in Vietnam. They were not identified for detailed study of health effects related to herbicide exposure until the late 1980s (Thomas and Kang, 1990). An initial feasibility study recruited Vietnam veterans and non-deployed Vietnam-era veterans from within the ACC (Kang et al., 2001). Blood samples collected from 50 Vietnam veterans in 1996 showed an association between veterans reporting having sprayed herbicides and higher serum TCDD concentrations; this finding was confirmed in a follow-up study of a larger fraction of the cohort (Kang et al., 2006). Modeling efforts (Ross et al., 2015a,b) have also found that higher exposures were probably experienced by those involved with mixer, loader, and applicator activities than by bystanders because of the fact that those in the first group were generally in closer proximity to and had more frequent contact with the herbicides.
Other veteran populations may also have been involved in handling herbicides although probably to a small degree. As discussed in Young (2009), for example, in 1971 the US Department of Defense (DOD) initiated Operation PACER IVY, which was responsible for removing stocks of Agent Orange from Vietnam to Johnston Island in the central Pacific Ocean. Operation PACER IVY was the responsibility of the 7th Air Force with assistance from Ranch Hand units and the ACC. PACER IVY procedures included the identification of unused herbicides, the transport of the identified herbicides to a central location in Vietnam for relabeling, and, for about half of the barrels, re-drumming before shipment. Potential Agent Orange hot spots included central PACER IVY locations, such as Du Nang, Bien Hoa, and to a small extent Phu Cat and Nha Trang airbases (Young, 2006). Although this is not certain, exposures of Allied troops from PACER IVY may have been low because most of the relabeling, repackaging, and handling of Agent Orange during PACER IVY was overseen and conducted by Chinese contractors, local Vietnamese, and the Vietnamese military. However, there were spills of Agent Orange in the de- and re-drumming and storage areas, which contaminated surrounding soils and asphalt (Young, 2009), and these have been suggested as possible sources of exposure. Other possible points of contamination for Vietnam-era veterans include defoliation tests conducted in South Vietnam as part of Project AGILE; ports in New Orleans, Louisiana; Baltimore, Maryland; Seattle, Washington; Mobile, Alabama; and Gulfport, Mississippi, which served as embarkation points for shipping of Agent Orange to Vietnam; storage locations on Johnston Island, where contamination could have occurred from re-drumming and maintenance of drums that contained Agent Orange; and
at-sea incineration of Agent Orange as part of Operation PACER HO (Young, 2009). Because the Army of the Republic of Vietnam (ARVN) was responsible for handling, transport, and storage of herbicides from the time it was delivered to Vietnam until it was loaded onto Ranch Hand aircraft, the herbicide exposures of Allied troops during these procedures may have been negligible.
Exposure of Ground Troops
In light of the widespread use of herbicides in Vietnam for many years, it is reasonable to assume that many military personnel were inadvertently exposed to the chemicals of concern. In surveys of Vietnam veterans who were not part of the Ranch Hand or ACC groups, 25 to 55 percent said that they believed they had been exposed to herbicides (CDC, 1989b). That belief has been supported by government reports (GAO, 1979) and reiterated by veterans and their representatives in testimony to the VAO committees over the years.
In contrast with those reports and veteran testimony, Young and colleagues provide evidence in a series of papers that is consistent with the veterans having received minimal exposures to herbicides (Young et al., 2004a,b). They used data from unpublished military records and environmental-fate studies to argue that ground troops had little direct contact with herbicide sprays and that TCDD residues in Vietnam had low bioavailability. They also argued that direct exposures of ground troops were relatively low because herbicide-spraying missions were carefully planned, and spraying occurred only when friendly forces were not in the target area.
To resolve the issue, numerous attempts were made in the 1980s to characterize the herbicide exposures of people who served as ground troops in Vietnam (CDC, 1988a; Erickson et al., 1984a; NRC, 1982; Stellman and Stellman, 1986; Stellman SD et al., 1988a). Those efforts combined self-reports of contact with herbicides or military service records with aerial-spray data to produce an Exposure Opportunity Index (EOI). For example, Erickson et al. (1984a) created five exposure categories based on military records in order to examine the risks of birth defects among the offspring of veterans. Those studies were conducted carefully and provided reasonable estimates based on available data, but there were no means of testing the validity of the estimates available at the time.
The search for a validation method led to the development of exposure biomarkers in veterans. Initial studies measured concentrations of dioxin in adipose tissue of veterans (Gross et al., 1984; Schecter et al., 1987). A study sponsored by the New Jersey Agent Orange Commission was the first to link dioxin concentrations in adipose tissue to dioxin concentrations in blood (Kahn et al., 1988). At the same time, the Centers for Disease Control (now the Centers for Disease Control and Prevention) undertook what came to be called the Agent Orange Validation Study, measuring TCDD in the serum portion of blood from a relatively large sample of Vietnam veterans and other Vietnam-era veterans (CDC,
1989a). The study did not find a statistically significant difference in mean serum TCDD concentrations between the groups: The mean values in each group were about 4 parts per trillion (ppt), and only two Vietnam veterans had concentrations greater than 20 ppt (CDC, 1988a). A review of a preliminary report of the work by an advisory panel established through the IOM concluded that the long lag between exposure and the serum measurements (about 20 years) called into question the accuracy of exposure classification based on serum concentrations. The panel concluded that estimates based on troop locations and herbicide-spraying activities might be more reliable indicators of exposure than serum measurements (IOM, 1987).
The report of the first VAO committee (IOM, 1994) proposed further work on exposure reconstruction and the development of a model that could be used to categorize exposures of ground troops. The committee cautioned that serum TCDD measurements should not be regarded as a “gold standard” of exposure, that is, as a fully accurate measure of herbicide exposure. Efforts to develop exposure-reconstruction models for US Vietnam veterans are discussed later in this chapter.
One other effort to reconstruct exposure was reported by researchers in the Republic of Korea who developed an exposure index for Korean military personnel who served in Vietnam (Kim JS et al., 2001, 2003). The exposure index was based on herbicide-spray patterns in military regions in which Korean personnel served during 1964–1973, time–location data on the military units stationed in Vietnam, and an exposure score derived from self-reported activities during service. The researchers were not successful in an attempt to validate their exposure index with serum dioxin measurements.
Exposure of Personnel Who Had Offshore Vietnam Service
US Navy riverine units are known to have used herbicides while patrolling inland waterways (IOM, 1994; Zumwalt, 1993), and it is generally acknowledged that estuarine waters became contaminated with herbicides and dioxin as a result of shoreline spraying and runoff from spraying on land, particularly in heavily sprayed areas that experienced frequent flooding. Thus, military personnel who did not serve on land could have been among those exposed to the chemicals during the Vietnam conflict. In recent years, there has been concern about dioxin exposure among personnel who served offshore but within the territorial limits of the Republic of Vietnam. It has been hypothesized that in addition to possibly experiencing drift from herbicide-spray missions, personnel on those ships that converted seawater by distillation may have been exposed via drinking water. Those concerns were heightened by findings from an Australian study (Muller et al., 2002) that showed that TCDD could be enriched in a simulation of the potable-water distillation process that was used on US Navy and Royal Australian Navy ships during the Vietnam War era. The National Academies convened the Blue Water Navy Vietnam Veterans and Agent Orange Exposure Committee
to address that specific issue; its report (IOM, 2011b) found that information to determine the extent of exposure experienced by Blue Water Navy personnel was inadequate, but that there were possible routes of exposure.
As summarized by Constable and Hatch (1985), Vietnamese researchers have made a number of attempts to characterize the herbicide exposure of residents of Vietnam in the process of trying to assess adverse reproductive outcomes. Some researchers compared residents of the South with residents of the unsprayed North, and others endeavored to compare South Vietnamese people who lived in sprayed and unsprayed villages as determined by observed defoliation. To evaluate reproductive outcomes, the pregnancy outcomes of North Vietnamese women married to veterans who had served in South Vietnam were compared with those of women whose husbands had not. In some cases, records of herbicide spraying have been used to refine exposure measurements. In assessing infant mortality, Dai et al. (1990) considered village residents to have been exposed if an herbicide mission had passed within 10 km of the village center and classified exposure further by length of residence in a sprayed area and the number of times that the area reportedly had been sprayed.
Armitage et al. (2015) used the US Forest Service’s Agricultural Dispersion (AGDISP), an aerial dispersion, to compare the distributions on the forest canopy and in the soil that would be expected following herbicide spraying in South Vietnam. Results were coupled with a chemical fate and transport model and with additional models considering dermal exposure via direct overspray and long-term dietary exposures. The investigators concluded that highly elevated exposures to the people in the upland forests of South Vietnam were not common.
A small number of studies have provided information on TCDD concentrations in Vietnamese civilians who were exposed during the war (Schecter et al., 1986, 2002, 2006). Dwernychuk et al. (2002) emphasized the need to evaluate dioxin contamination around former air bases in Vietnam. Those researchers collected environmental and food samples, human blood, and breast milk from residents of the Aluoi Valley of central Vietnam. The investigators identified locations where relatively high dioxin concentrations remained in soil or water systems. Soil dioxin concentrations were particularly high around former airfields and military bases where herbicides were handled. Fish harvested from ponds in those areas were found to contain high dioxin concentrations. Dwernychuk (2005) elaborated on the importance of “hot spots” as important locations for future studies and argued that herbicide use at former US military installations was the most likely cause of the hot spots. Other hot spots that have been identified include depots of chemical defoliants, airbases used for defoliant spray missions, and areas where chemical defoliants were used extensively. The Vietnamese population has since inhabited the areas in and around many former airbases and
depots, which have become the focus of studies of environmental contamination and bioaccumulation. Considering results of modeling exercises quantifying the dispersion and extent of exposure, Armitage et al. (2015) similarly emphasize hot spots as locations of higher potential exposure to TCDD as compared with areas primarily affected by aerial spraying only. The Bien Hoa Air Base, which is considered a hot spot because of the use of chemical defoliants around the base, was the focus of a study that examined dioxin contamination in soils in Vietnam (Mai et al., 2007). The study found high soil concentrations but did not estimate the exposures of people who lived in the vicinity of the bases. More recently, Hoang et al. (2014) reported that dioxin total toxic equivalent (TEQ) levels in eggs of poultry raised by the Vietnamese population currently living on the former Bien Hoa airbase were found to exceed the adult exposure guideline set forth by the World Health Organization (WHO) by two-fold and the child guideline by five-fold. In Thau Thien-Hue Province, a region affected by defoliant spraying, Banout et al. (2014) reported that although dioxin and furan (PCDD/ PCDF) concentrations were below the WHO recommended guideline for dioxin TEQ in sediments, the concentrations in the muscle and liver of the poultry raised in the region exceeded the WHO guidelines for dioxin content per unit fat mass.
Publications reviewed in earlier updates have reported environmental concentrations and human body burdens of dioxins in various areas throughout Vietnam (Brodsky et al., 2009; Feshin et al., 2008; Hatfield Consultants, 2009a,b,c; Nhu et al., 2009; Saito et al., 2010; Tai et al., 2011). They have found pervasive exposure to dioxins more than a half-century after the Vietnam War. Dioxin concentrations in breast milk reflect the residence location of the mothers, with levels and TEQs being elevated in areas where herbicides sprayed during the war and tending to be still higher in areas where herbicides were stored. Phu Cat airbase, a hot spot in South Vietnam, has recently been the focus of several studies examining dioxin levels in human sera and breast milk and corresponding levels of steroid hormones. In 16 mother–baby pairs from Phu Cat airbase compared to 10 pairs from Kim Bang, Manh et al. (2013) reported significantly higher concentrations of salivary cortisol, cortisone, and dehydroepiandrosterone (DHEA) in primiparous mothers and of dioxin TEQs in their milk. The associations between dioxin TEQ levels in mothers’ breast milk and salivary hormone levels were non-linear: The relationship was U-shaped for estradiol, and an inverted U-shape for cortisol, cortisone, and DHEA. The two groups did not differ significantly in the concentrations of four other salivary steroid hormones (androstenedione, estradiol, progesterone, or testosterone). In men, however, Sun et al. (2014) reported no correlation between serum dioxin TEQ and steroid hormones for either those who had lived in and around the Phu Cat airbase for 50 years or more or those who had lived in the unsprayed Kim Bang district of Ha Nam Province in North Vietnam. Manh et al. (2014) reported a correlation between proximity of residence to Phu Cat airbase and serum dioxin levels among men. When contrasting men from Kim Bang who had spent time in South Vietnam
during or after the war to those who had not, however, Manh et al. (2014) found no significant differences in serum dioxin levels between the two groups. These findings imply a greater body burden of dioxin exists in those currently living in the vicinity of the Phu Cat airbase than remains in those who spent time in areas that had been the target of herbicide spraying.
The above studies are not directly relevant to the present committee’s task, but they may prove useful in future epidemiologic studies of the Vietnamese population and in the development of risk-mitigation policies.
The development of a means of characterizing the exposure of individual Vietnam veterans has long been a prime objective for use in refining epidemiologic investigations of health outcomes in this population. Serum TCDD levels might have been a very useful proxy for harmful exposures to all the components of the herbicides used by the US military in Vietnam. As analytic methods for TCDD have become much more sensitive and somewhat less costly with the passage of time, body burdens in even quite highly exposed individuals by now would have decreased to such an extent over many half-lives that newly gathered samples would be minimally informative. The consideration of records detailing the herbicide spray missions has provided another approach to deriving individual-specific exposure estimates. Two models—a proximity-based EOI model and an aerial spray distribution model (explained and contrasted below)—have been proposed for estimating the exposure of Vietnam veterans. Until this update, neither had actually been applied in an epidemiologic investigation. The use of the EOI model in studying health consequences in a large cohort of Korean veterans who participated in the Vietnam War is assessed in the final portion of this section.
Exposure Opportunity Index Model
The IOM, following up on the recommendations contained in the original VAO report (IOM, 1994), issued a request for proposals seeking individuals and organizations to develop historical exposure-reconstruction approaches suitable for epidemiologic studies of the herbicide exposure of US veterans during the Vietnam War (IOM, 1997). The request resulted in the project Characterizing Exposure of Veterans to Agent Orange and Other Herbicides in Vietnam. The project was carried out under contract by a team of researchers in Columbia University’s Mailman School of Public Health. The Columbia University project integrated various sources of information concerning spraying activities and information on the locations of military units assigned to Vietnam, all compiled into a database. The resulting EOI model (Stellman and Stellman, 2003) generates individualized estimates (EOI scores) of the exposure potential of troops serving in Vietnam.
Mobility factor analysis, a technique used for studying troop movement, was developed for use in reconstructing herbicide-exposure histories. The analysis is a three-part classification system for characterizing the location and movement of military units in Vietnam. It comprises a mobility designation (stable or mobile), a distance designation (usually in kilometers) to indicate how far a unit might travel in a day, and a notation of the modes of travel available to the unit (by air, by water, or on the ground by truck, tank, or armored personnel carrier). A mobility factor was assigned to every unit that served in Vietnam.
The data were combined into a geographic information system (GIS) for Vietnam. Herbicide-spraying records were integrated into the GIS and linked with data on military-unit locations to derive individual EOI scores. The results are the subject of reports by the contractor (Stellman and Stellman, 2003) and the Committee on the Assessment of Wartime Exposure to Herbicides in Vietnam (IOM, 2003b,c). A summary of the findings on the extent and pattern of herbicide spraying (Stellman et al., 2003a), a description of the GIS for characterizing exposure to Agent Orange and other herbicides in Vietnam (Stellman et al., 2003b), and an explanation of the EOI model based on that work (Stellman and Stellman, 2004) have been published in peer-reviewed journals. In those publications the researchers have argued that it is feasible to conduct epidemiologic investigations of veterans who served as ground troops during the Vietnam War. The IOM later issued a report that examined the feasibility of using the EOI model developed by Columbia University (IOM, 2008b). The report concluded that “despite the shortcomings of the exposure assessment model in its current form and the inherent limitations in the approach, the committee agreed that the model holds promise for supporting informative epidemiologic studies of herbicides and health among Vietnam veterans and that it should be used to conduct studies” (p. 2).
As an alternative to the EOI model, Ginevan et al. (2009a) proposed the use of the AgDRIFT Tier III forestry model for estimating the deposition of herbicides via aerial spraying. Hewitt et al. (2002a) presented the history of AgDRIFT’s development as a model to describe deposition and drift patterns resulting from spraying from fixed wing aircraft and helicopters. The National Aeronautics and Space Administration (NASA) sponsored the initial development of the model’s computational approach. The Forest Service of the US Department of Agriculture (USDA) and the US Army oversaw its refinement into the AGDISP model. The final development of AgDRIFT occurred under a cooperative agreement between the Spray Drift Task Force (SDTF) (i.e., a consortium of chemical companies formed in 1990), the US Environmental Protection Agency (EPA), and the USDA. Hewitt et al. (2002b) discussed the Lagrangian modeling of physical properties (such as droplet size, wind speed and direction, and equipment
design) factored into the models and validation efforts that ultimately resulted in the AgDRIFT model.
The AgDRIFT model can provide, among other things, ground and foliar deposition estimates derived from application information such as aircraft speed and altitude, nozzle characteristics, and droplet evaporation and environmental parameters such as canopy density, canopy roughness, and crosswind speed (Ginevan et al., 2009a). AgDRIFT outputs are then used to estimate dermal exposure through both direct deposition and post-application transfer from foliage. Exposures resulting from contact with soil and dust, and through the inhalation route, are considered negligible and are not included. Ginevan et al. (2009a) claim that the resulting estimates are more accurate and more appropriate for estimating aerial herbicide exposure than those from the EOI model because they are quantitative in nature, unlike the EOI model, which was designed to provide rank-ordered exposures.
Comparison and Validation of EOI and AgDRIFT-based Models
Until the current update, the position of earlier committees has been that exposure to herbicides experienced by US troops in Vietnam could not be determined with any certainty. Because the EOI and AgDRIFT models have both been proffered as solutions for estimating the extent of herbicide exposure in Vietnam, a summary of the efforts to compare and validate these models is included in this update. Overall, the committee recognizes that there is little consensus as to how accurate a model should be in order for it to be useful. The necessary accuracy likely differs depending on the application and the implications of using modeled estimates, and validation under one set of assumptions may not indicate validation under a different set of conditions. Furthermore, there is no standard against which these models could now be validated.
In support of validation of the AgDRIFT model, Ginevan et al. (2009a) pointed to Bird et al. (2002) and asserted justification for the use of this model in estimating the exposure to herbicides in Vietnam. Bird et al. concluded that the AgDRIFT model reasonably predicted average field deposition levels of herbicides when compared with 161 low-flight (< 10 m) aerial field trials collected over three field studies by the SDTF. However, the model under predicted mean deposition levels in the near-field and overpredicted at far-field distances. In a peer review of the collection of the SDTF field studies, several critiques emerged on the data used for validation. Several of the reviews that were requested by the Environmental Fate and Effects Division (EFED) of EPA’s Office of Pesticide Programs praised the efforts and agreed that the resulting data represented the state of the art at the time, but they also concluded that there were several missed opportunities to collect data that represent real-world conditions in agricultural settings (Akesson, 1997; EFED, 1997; Fox, 1997; Kirk, 1997; Mulchi, 1997; Zhang et al., 1997). These conditions and suggestions included a broader array
of atmospheric stability classes, deposition on vegetation and canopies rather than relatively bare ground, an experimental design to capture airborne spray and the full extent of the drift plume, adequate methods to estimate dermal and inhalation exposure to non-target organisms, measurement of post-application volatilization and other fate and transport phenomena (for example, spray flux, lift off, attachment to water vapor), and larger field applications.
Perhaps a more appropriate scenario applicable to Vietnam includes high-flight (> 10 m) forestry applications, which are briefly referenced in Bird et al. (2002). Several of these analyses, however, were carried out using the preceding AGDISP or the Forest Service’s near-wake model, which, similar to AgDRIFT, incorporates AGDISP computations (Teske et al., 2002). Testing the near-wake model against 17 aerial spray trials over mixed oak forest, Anderson et al. (1992) found that the model adequately predicted the ensemble average over a large number of spray applications but that it was less successful in predicting values from individual runs and could not replicate the wider variability of measured deposition. In evaluating computational methods for aircraft wake effects, Rafferty and Bowers (1993) compared results from near-wake and AGDISP models to deposition measurements from 15 trials in two field programs. When examining trials over a mixed conifer forest, they found areas of over prediction and under prediction similar to other studies, and only about 10 to 15 percent of modeled values fell within a factor of two of the observed measurements. Additionally, the study uncovered the importance of deposition sampling methods. Statistical differences were found between two difference sampling methods (i.e., spot count and manganese analysis) that were used simultaneously on the same trials. This difference in measuring deposition further complicates attempts at comparing modeled and observed values. Finally, when testing the near-wake model against 12 field tests, Richardson et al. (1995) found that although overall predictions showed good correlation with ground deposition measured to a distance of 300 m, modeled values generally underpredicted measured values (i.e., by a range of 3.3 to 27.0 when measured as the maximum ratio between prediction and measurement), and the model had trouble predicting the location of peak deposition.
It is difficult to judge whether these evaluations are currently appropriate in light of the improvements in AgDRIFT over the years and the various versions used in validation attempts over time. Additionally, while several field studies have been used to test the modeled ground deposition results of AgDRIFT, there do not appear to have been any tracer or field studies to validate the composite model relating deposition to human exposure estimates.
Discrepancies between the modeled AgDRIFT deposition estimates and scores from the EOI model, when applied to similar time points for Vietnam-era exposures, were reported by Ginevan et al. (2009a). For example, the AgDRIFT model predicted a much smaller area under the spray path and herbicide concentrations that are several orders of magnitude lower than EOI estimates for the same set of sample flight paths. The differences between the two exposure
estimation methods were particularly pronounced at points distant from the spray path, with AgDRIFT predicting herbicide exposures up to 20 orders of magnitude lower than the EOI model at a location 4 km away from the flight-path centerline.
In their response to the Ginevan et al. (2009b) critique of the EOI model, Stellman and Stellman (2013) questioned the validity of most, if not all, of Ginevan’s calculations of EOI scores, citing errors regarding the use of “incorrect data and its fundamentally incorrect and negative interpretations.” The Stellmans noted that Ginevan et al. compared raw EOI scores to log-transformed EOI scores. As a result, the variability in raw EOI scores on the flight line was potentially artificially high, but the log-transformed scores produced reasonable values that varied within 10 percent around the mean for each spray mission. The Stellmans also stated that the use of raw EOI scores leads to a host of other incorrect assertions and theories, such as “an incorrect score of 60,791 for one point, when the true score is zero in our [the EOI] system” (p. 2).
Given the lag time since herbicide exposures in Vietnam took place and the lack of direct exposure measurements from that era, it is neither possible to fully validate either the AgDRIFT or EOI models, nor to ascertain the accuracy and precision of estimates from either model or the claims of either Stellman and Stellman (2013) or Ginevan et al. (2009a,b). In addition, because the intent and outcomes of the two models differ substantially, the model results and interpretation would likely differ. The EOI model, for example, predicts potential exposure to troops on the basis of military data on spray history and troop locations. The AgDRIFT model, in contrast, predicts ground concentrations and their spatial dispersion with additional equations extending dispersion to the fraction deposited on skin and transferred from foliage; however, by design, the AgDRIFT model does not consider troop-location data.
The issue of Allied troop presence during spraying is one of the central issues in the debate regarding the use of the EOI model. The EOI model relied on actual military data on spray history and troop locations, which, as pointed out by both Stellman and Stellman (2004) and Young (2009), are limited in their spatial and temporal resolution and accuracy. The accuracy of the records with regard to missions flown, mission locations, and number of gallons sprayed, and other important information was examined by MITRE Corporation (Heizer, 1971). MITRE reported that about 2 percent of the records were missing data, 6 percent of the records had serious transcription or measurement errors, and 23 percent of the records that had complete data were off by 50 percent in the reported distance sprayed (Young, 2009). However, the overall quality of the data was found to be good, and it could be improved with adjustments, as performed by Stellman et al. (2003a,b) and others (ESG, 1985; NRC, 1974). Whether the adjustments improved the quality of the military data is not known. However, the troop movement information in the GIS database compiled by Stellman et al. (2003a,b) for the EOI model was compiled with the assistance of the US Armed Services Center for Research of Unit Records with presumably limited potential
for information bias. These data account for locations and changes in locations of approximately 80 percent of Army troops and most Air Force and Navy personnel during the Vietnam era (IOM, 2003b).
Exposure Estimation in Korean Veterans Health Study
Military personnel of the Republic of Korea served in Vietnam during 1964–1973. Since Update 2012, the committee reviewed several publications from a large epidemiological study of more than 114,000 Korean Vietnam veterans, four of which described how the exposure metrics used were derived (Yi et al., 2013a,b, 2014a,b). This study cohort, referred to here as the “Korean study,” is much larger in scope than any of the other published epidemiological studies conducted among Vietnam veterans. The results of a very large set of health outcomes examined in the Korean study are discussed in subsequent chapters of this report. Exposures to herbicides and their contaminants were estimated using an EOI method developed by Stellman et al. (2003a). This model produced a set of EOI scores, as summarized below, based on the descriptions provided by Yi et al. (2013a,b, 2014a,b). Although the IOM (2008b) acknowledged that it was not feasible to directly validate the accuracy of exposure assignments developed by the EOI method, the committee encouraged efforts to quantify the degree of accuracy and to incorporate those estimates into sensitivity analyses. The Korean study, unfortunately, did not provide information on whether the influence of uncertainty on estimated associations between the EOI metric and specific health outcomes had been evaluated.
The Korean study investigators used a number of methods to estimate potential exposure to herbicides during Vietnam service. First, a self-report perceived exposure index was used to query Korean veterans as to how they might have been exposed to herbicides in Vietnam (Yi et al., 2013a,b). Study participants were asked to respond “yes,” “no,’” or “do not know” to questions regarding perceived exposure to herbicides. The survey results showing the distribution of perceived herbicide exposure among the Korean veterans are presented in Table 3-2. These self-reported perceived exposures are not directly comparable to the objective EOI scores, which were designed to assess the exposure opportunity that would result from unintended proximity to herbicide spraying and not the direct result of duties that required handling or applying herbicides (IOM, 2008b). The perceived herbicide exposure estimates were highly correlated with the health outcomes in Yi et al. (2013a), indicating the possibility of recall bias.
Because concerns about potential inaccuracy or recall bias in self-reports of exposure and disease, as well as empirical observations of inconsistencies when such data are compared to more objective metrics of exposure potential, the committee expressed considerable concern about findings based on self-report exposure and outcome data in the Korean study (Yi, 2013; Yi et al., 2013a).
|2 Groups||Exposure Questions||4 Groups||Prevalence|
|High||1. Sprayed herbicides||High||16.1%|
|2. Handled herbicide spray equipment|
|3. Present during herbicide spraying||Moderate||35.7%|
|4. Got herbicide on skin or clothing|
|Low||5. Walked through sprayed area||Low||13.2%|
|6. Exposed in other ways (not listed above)|
|Answered “no” to all six questions||None||34.9%|
In the second method, an objective EOI score of exposure potential was calculated for each veteran based on the proximity of the veteran’s military unit to herbicide sprayed areas. The Korean investigators obtained locations and calendar date histories for the military units represented in their cohort and provided this information to the Stellman group to use as input to obtain EOI scores from its model, which consolidates all the temporal and spatial information gathered from the original military records on the herbicide spray missions conducted in Vietnam. The investigators classified the resulting EOI scores using two- and four-group categorizations, and multiple aggregations of military units, as summarized in Tables 3-3 and 3-4.
The Korean study aggregated military units at two levels in the development of EOI scores: the larger brigade/division level and the smaller battalion/company level. According to Yi et al. (2013a), the Vietnam post locations and the tactical and operational areas were identified at the battalion level and higher through
|Exposure Category (log10 EOI score)||Yi et al. (2014a)b Division/Brigade or Battalion/Company (n = 111,726)d||Yi et al. (2014b)c Division/Brigade (n = 180,251)e|
|Low (< 4.0)||62.0%||52.4%|
|High (≥ 4.0)||38.0%||47.6%|
bBattalion/company level EOI score assigned for combat units only.
|Exposure Category (log10 EOI score)||Yi et al. (2013a) Division/Brigade (n = 96,126)b||Yi et al. (2013a) Battalion/Company (n = 96,126)||Yi et al. (2014a) Division/Brigade or Battalion/Companyc (n = 111,726)d||Yi et al. (2014b) Division/Brigade (n = 180,251)e|
|None (< 0.1)||20.1%||26.1%||30.9%||25.2%|
|Low (0.1 ≤ EOI < 4.0)||28.2%||33.1%||31.2%||27.2%|
|Med (4.0 ≤ EOI < 5.0)||31.1%||21.5%||20.1%||28.3%|
|High (≥ 5.0)||20.6%||19.3%||17.9%||19.3%|
aDetails of the Four-level exposure classification is described in Ohrr et al. (2006, publication in Korean).
bYi et al. (2013a), 96,126 veterans analyzed for self-reported disease prevalence. Log (EOI score) mean and range not reported.
cBattalion/company level EOI score assigned for combat units only.
dYi et al. (2014a), 111,726 veterans analyzed for disease prevalence. Log (EOI score) mean 2.6 ± 2.2 (range 0.0–6.2).
eYi et al. (2014b), 180,251 veterans analyzed for cancer outcomes. Log (EOI score) range 0.0–5.8.
records review. Because veterans also reported the specific battalion/company in which they served while in Vietnam, this presumably finer unit level was also used in estimating EOI scores. However, no information regarding the validity and reliability of the company-level spatial coordinates relative to military records was provided. The rationale for the two- and four-level exposure categories is explained in more detail only in a Korean-language paper (Ohrr et al., 2006), but the distributions of EOI scores are similar across the Korean study publications (see Tables 3-3 and 3-4) regardless of military unit aggregation. However, the committee noted that proportion of veterans in the “high” exposure category may be too large for optimal detection of associations between exposure and adverse health conditions. Stated another way, the “high” exposure individuals may be too similar to the lower categories, thereby diluting the strength of the associations. An exposure classification that put only the top 10 or 15 percent in the “high” category would perhaps have been better for the purpose of identifying adverse health effects due to exposure.
In summary, the recent Korean study overcame significant logistical challenges in applying the EOI model to a large-scale epidemiologic study of a broad spectrum of health effects. The Korean researchers did not refine the EOI model for the influence of environmental fate and transport between spraying of the herbicides and possible exposure of ground troops, as recommended by the IOM in 2008. Nor were the results subjected to sensitivity testing for variability arising from the parameter values selected and other sources of uncertainty in the exposure assessment method. Nonetheless, compared to the severe constraints on exposure assessment in previous studies of this population of ultimate interest,
this first application of the EOI model represents a “more accurate, if still imperfect, method to increase the specificity of exposure classification” for observing the association between herbicide exposure and health effects among Vietnam veterans (IOM, 2008b, p. 84).
In conclusion, the committee acknowledges that there are undoubtedly sources of error in the EOI method for modeling herbicide exposures of Vietnam veterans, but there is no indication of systematic bias in rank ordering of exposure scores developed by this method. Given that nondifferential misclassification of exposure would bias measures of association toward the null, observed statistically significant relationships between EOI scores and health effects are likely to be real.
The focus here is on several key methodologic issues that complicate the development of accurate estimates of exposure of the Vietnam-veteran population and the other study populations discussed in this report: The latent period between exposure and disease, exposure misclassification, and exposure specificity.
The temporal relationship between exposure and disease is complex and often difficult to define in studies of human populations. Many diseases do not appear immediately after exposure. Cancers, for example, might not appear for many years after exposure. The time between a defined exposure period and the occurrence of disease is often referred to as a latent period (IOM, 2004). Exposures can be brief (sometimes referred to as acute exposures) or protracted (sometimes referred to as chronic exposures). At one extreme, an exposure can be the result of a single event, as in an accidental poisoning. At the other extreme, a person exposed to a chemical that is stored in the body may continue to experience “internal exposure” for years even if exposure from the environment has ceased. The determination of the proper timeframe for duration of exposure constitutes a challenge to exposure scientists.
Exposure misclassification in epidemiologic studies can affect estimates of risk. A typical situation is in a case-control study in which the reported measurement of exposure of either group or both groups can be misclassified. The simplest situation to consider is the binary case in which the exposure is classified into just two levels, for example, “ever exposed” versus “never exposed.” If the probability of exposure misclassification is the same in both cases and controls (that is, nondifferential), then it can be shown that the estimated association
between disease and exposure is biased toward the null value; in other words, one would expect the true association to be stronger than the observed association. However, if the probability of misclassification is different between cases and controls, then a bias in the estimated association can occur in either direction, and the true association might be stronger or weaker than the observed association.
The situation in which exposure is classified into more than two levels is somewhat more complicated. Dosemeci et al. (1990) demonstrated that in that situation the slope of a dose–response trend is not necessarily attenuated toward the null value even if the probability of misclassification is the same in the two groups of subjects being compared; the observed trend in disease risk among the several levels of exposure may be either an overestimate or an underestimate of the true trend. Greenland and Gustafson (2006) discussed the effects of exposure misclassification on the statistical significance of the result and demonstrated that if one adjusts for exposure misclassification when the exposure is represented as a binary variable, the resulting association is not necessarily more significant than in the unadjusted estimate. That result remains true even though the observed magnitude of the association (for example, the relative risk) might be increased.
Even progressing beyond discrete exposure categories, some continuous exposure metrics can be problematic from the perspective of misclassification. It is often noted that continuous measures of an exposure variable carry more information per observation than do those that are partitioned into categories at basically arbitrary cut points. Despite their continuous nature, however, measurements of serum TCDD levels have decreasing utility for epidemiologic research as they are derived from samples drawn longer and longer after the exposure in question occurred. The variance of the underlying exposed and non-exposed groups has increased to the extent that two overlapping populations can no longer be distinguished, effectively leading to an increase in misclassification.
The committee has been concerned about the strong possibility that the degree of misclassification associated with a particular exposure assignment convention employed in several recent publications (Ansbaugh et al., 2013; Li et al., 2013; Qureshi et al., 2013) may be vastly underappreciated by even the researchers using it. Over the past few years, VAO literature searches have identified publications concerning various health outcome authored by researchers affiliated with the VA medical care delivery system. The analyses in question used exposure categories assigned on the basis of a variable in a patient’s electronic medical record indicating whether the individual was “exposed to Agent Orange.” Its presence in VA’s medical records system conveys a degree of authenticity that the committee strongly suspects is unmerited. From the vagueness of what the committee could learn about how this variable is populated, it remains unclear whether the source of this information is deployment status, entry on the Agent Orange Registry, the veteran’s self-report, a physician’s observation that the patient has a condition presumed to be service-related, results of serum TCDD measurements performed on some patients, or some other criteria. At any rate,
none of these approaches represents a reliable method of determining whether an individual was truly exposed to herbicides (above some unspecified level) or for uniform application to all veterans using the VA medical system, who themselves are a self-selected subset of veterans. The committee has no reason to believe VA has access to some previously unrecognized means of definitively establishing whether a given veteran was truly exposed to herbicides in Vietnam. The committee is concerned that these publications misrepresent (perhaps unintentionally) the reliability of the underlying exposure metric.
The incorporation of the findings of studies of persons exposed to components of the herbicides sprayed in Vietnam requires some decisions about their relative contributions to the VAO project’s evidentiary database. Only a few herbicidal chemicals were used as defoliants during the Vietnam conflict: esters and salts of 2,4-D and 2,4,5-T, cacodylic acid, and picloram in various formulations. Many scientific studies reviewed by the committee report exposures to broad categories of chemicals rather than to those specific chemicals. The categories are presented in Tables 3-5 and 3-6 with their relevance to the committee’s charge. The information in these tables has helped to guide the committee’s evaluation of epidemiologic studies. Earlier VAO committees did not address the issue of exposure specificity in exactly this manner. The committee for VAO and the first several updates gave more weight to results that were based on job title (for example, “farmer” with
TABLE 3-5 Current Committee Guidance for the Classification of Exposure Information in Epidemiologic Studies That Focus on the Use of Pesticides or Herbicides, and Relevance of the Information to the Committee’s Charge to Evaluate Exposures to 2,4-D and 2,4,5-T (Phenoxy Herbicides), Cacodylic Acid, and Picloram
|Specificity of Exposure Reported in Study||Additional Information||Relevance to Committee’s Charge|
|Pesticides||Chemicals of interest were not used, or there was no additional information||Not relevant|
|Chemicals of interest were used||Limited relevance|
|Herbicides||Chemicals of interest were not used||Not relevant|
|There was no additional information||Limited relevance|
|Chemicals of interest were used||Relevant|
|Phenoxy herbicides||—||Highly relevant|
|2,4-D or 2,4,5-T||—||Highly relevant|
|Cacodylic acida||—||Highly relevant|
aNone of the epidemiologic studies reviewed by the committee to date has specified exposure to cacodylic acid.
TABLE 3-6 Current Committee Guidance for the Classification of Exposure Information in Epidemiologic Studies That Focus on Exposure to Dioxin-Like Chemicals and Relevance of the Information to the Committee’s Charge
|Specificity of Exposure Reported in Study||Additional Information||Relevance to Committee’s Charge|
|Dioxin-like chemicals||Exposure to PCBs or polychlorinated dibenzofuran (PCDFs)||Limited relevance|
|Dioxin-like chemicals||Results expressed in terms of (total) toxic equivalent (TEQs) or concentrations of individual congeners recognized as having dioxin-like activitya||Highly relevant|
|TCDD or mixture of PCDDs||Established on the basis of environmental sampling or work histories||Highly relevant|
|TCDD or mixture of PCDDs||Concentrations in tissues of a subset of participants (preferably soon after exposure)||Very highly relevant|
|TCDD or mixture of PCDDs||Concentrations in tissues of individual participants (preferably soon after exposure)||Most informative|
aThe values of toxic equivalency factors for individual dioxin-like chemicals, which are weighted by concentration and summed to derive TEQs are presented in Table 4-2.
NOTE: PCB, polychlorinated biphenyl; PCDF, polychlorinated dibenzofuran; TCDD, 2,3,7,8-tetra-chlorodibenzo-p-dioxin; TEQ, (total) toxic equivalent.
no additional information) than have the committees for the past five updates, but entirely excluded findings from the Yusho and Yucheng PCDF and PCB poisonings, whereas recent committees have considered studies that analyzed for dioxin-like PCDF and PCB congeners and expressed the results in terms of TEQs. Studies that report TEQs based only on mono-ortho PCBs (which are PCBs 105, 114, 118, 123, 156, 157, 167, and 189), however, have been given only limited consideration because mono-ortho PCBs typically contribute less than 10 percent to total TEQs, based on the revised WHO Toxicity Equivalency Factor (TEF) scheme of 2005 (La Rocca et al., 2008; van den Berg et al., 2006). A 2013 joint WHO and United Nations Environment Programme evaluation concluded that there was sufficient evidence for the inclusion of brominated analogues of the dioxin-like compounds in the WHO TEF scheme (van den Berg et al., 2013). Classifications schemes for PCB congeners have also been evaluated in terms of gene expression (Warner et al., 2012). Recent studies of dioxin-like compounds have investigated systemic distribution of the congeners in rodents and differences in their relative effect potencies (van Ede et al., 2013a,b, 2014) and relative effect potencies in human thyroid responses (Trnovec, 2014; Trnovec et al., 2013).
Many studies have examined the relationship between exposure to “pesticides” and adverse health outcomes, while others have used the category of “herbicides” without identifying specific chemicals. A careful reading of a scientific report often reveals that none of the chemicals of interest (COIs) (that is, those used in Vietnam, as delineated above) contributed to the exposures of the study population, so such studies could be excluded from consideration. But in many cases, the situation is more ambiguous. For example, reports that define exposure in the broad category of “pesticides” with no further information have little relevance to the committee’s charge to determine associations between exposures to herbicides used in Vietnam and adverse health outcomes. Reports that define exposure in the more restricted category of “herbicides” are of greater relevance but are still of little value unless it is clear from additional information that an exposure to one or more of the herbicides used in Vietnam occurred in the study population. Possibilities include: if a published report indicates that the COIs were among the pesticides or herbicides used by the study population, if the lead author of the report has been contacted and has indicated that the COIs were among the chemicals used, if the COIs are used commonly for the crops identified in the study, or if the COIs are used commonly for a specific purpose, such as the removal of weeds and shrubs along highways.
Among the various chemical classes of herbicides that have been identified in published studies reviewed by the committee, phenoxy herbicides, particularly 2,4-D and 2,4,5-T, are directly relevant to the exposures experienced by US military forces in Vietnam. On the basis of the assumption that compounds with similar chemical structure may have analogous biologic activity, information on the effects of other chemicals in the phenoxy herbicide class—such as 2-(2,4,5-trichlorophenoxy) propionic acid (Silvex), 2-methyl-4-chlorophen-oxyacetic acid, 2-(2-methyl-4-chlorophenoxy) propionic acid (Mecoprop), and 3,6-dichloro-2-methoxybenzoic acid (dicamba)—has been factored into the committee’s deliberations with somewhat less weight. The very few epidemiologic findings on exposure to picloram or cacodylic acid have been regarded as highly relevant. The committee has decided to include many studies that report on unspecified herbicides in the health-effects sections, and the results of these studies have been entered into the health-outcome–specific tables; however, these studies tend to contribute little to the evidence considered by the committee. The many studies that provide chemical-specific exposure information are believed to be far more informative for the committee’s purposes.
A similar issue arises in the evaluation of studies that document exposure to dioxin-like compounds. Most “dioxin” studies reviewed by the committee have focused on TCDD, but TCDD is only one of a number of PCDDs. The committee recognizes that in real-world conditions exposure to TCDD virtually never occurs in isolation and that there are hundreds of similar compounds to which humans might be exposed, including other PCDDs, PCDFs, and PCBs. Human exposure to TCDD is almost always accompanied by an exposure to one or more of the
other compounds. The literature on the other compounds, particularly PCBs, has not been reviewed systematically by the committee except for those reports in which TCDD was identified as an important component of the exposure or the risks of health effects were expressed in terms of TEQs, which are the sums of toxicity equivalence factors for individual dioxin-like compounds as measured by activity with the aryl hydrocarbon receptor (AHR). The committee took that approach for two reasons. First, the exposure of Vietnam veterans to substantial amounts of the other chemicals, relative to exposure to TCDD, has not been documented. Second, the most important mechanism for TCDD toxicity involves its ability to bind to and activate the AHR. Many of the other chemicals act by different or multiple mechanisms, so it is difficult to attribute toxic effects after such exposures specifically to TCDD. Furthermore, people’s environmental exposures to dioxin-like chemicals and their non-dioxin–like counterparts are to mixtures of components that tend to correlate, so it is not surprising that specific chemicals measured in a person’s serum also tend to correlate; this means that it will be difficult for epidemiologic studies to attribute any observed association to a particular chemical configuration (Longnecker and Michalek, 2000). Analyses in terms of TEQs circumvent that problem to some extent.