Draft released September, 2015 –
For full document download, see link at bottom of this post.
This review is intended to assist the VA in making policy decision regarding the relationship between drinking water exposures to chemicals at Camp Lejeune and various health effects. The results of this review represent ATSDR’s assessment of the state of evidence at this time and we recognize that classifications used and the strength of evidence are subject to differing opinions and interpretations.
The drinking water serving the main portion of the base at Camp Lejeune was contaminated with measured levels of trichloroethylene (TCE) as high as 1,400 ppb as well as contaminants such as tetrachloroethylene (also known as perchloroethylene or PCE), vinyl chloride and benzene. Another drinking water supply serving the Tarawa Terrace housing area at Camp Lejeune was contaminated with measured PCE levels as 215 ppb. ATSDR has been tasked to provide the U.S. Department of Veteran Affairs (VA) with an assessment of the strength of the evidence for causal links between these contaminants and specific diseases. This report summarizes the evidence for 18 diseases for which there is at least some epidemiological evidence for an association with one or more of these contaminants.
ATSDR has developed tables for each disease that list the results from any meta-analyses that have been performed as well as epidemiological studies that were not included in meta-analyses because they appeared after the meta-analyses were conducted. These studies are mostly of workers exposed to these chemicals as well as the few drinking water studies that evaluated exposures to these chemicals. In addition, studies that have information on exposure duration, exposure intensity and/or cumulative exposure are included even if they were evaluated in a meta-analysis. Following each table, ATSDR provides its assessment of the evidence provided in the tables as well as assessments made by other agencies mandated to evaluate the health effects of these chemicals: EPA, NCI, NTP and IARC. Although the evidence from epidemiological studies is emphasized, the findings from animal studies, mechanistic studies, and other toxicological information is mentioned if necessary to support our conclusions.
The report provides a summary table listing each disease and ATSDR’s conclusion concerning the evidence in bold type. Our conclusions are heavily influenced by the recent meta-analyses and/or the reviews conducted by IARC (100F, 2012; 106, 2014; Vlaanderen et al 2014), EPA (2011, Scott et al 2011), NCI (Karami et al 2012, 2013, ) and NTP (2015).
Our assessment of the evidence for each disease uses the following classifications: “Sufficient evidence for causation”, “Modest evidence for causation”, “Sufficient evidence of an association”, and “Limited/Suggestive evidence of an association”.
Our classification of “sufficient evidence for causation” is similar to the IARC Group 1 classification: (1) there is sufficient evidence from human studies in which chance and biases (including confounding) can be ruled out with reasonable confidence, or (2) there is less than sufficient evidence from human studies but sufficient evidence in animal studies and strong evidence that the agent acts through a relevant mechanism in humans. Sufficient evidence from human studies can be provided by a meta-analysis or by several well-conducted studies in which biases can be ruled out with reasonable confidence.
Our classification of “modest evidence for causation” is when the degree of evidence from human studies is less than sufficient but there is supplementary evidence from animal studies and/or mechanistic studies that support causality. Modest evidence for causation could also occur if a meta-analysis does not provide convincing evidence (e.g., the summary risk estimate is close to the null value of 1.0, or if the meta-analysis finds no exposure-response relationship) but there are additional well-conducted epidemiological studies (for which biases can be reasonably ruled out), possibly occurring after the meta-analysis has been conducted, that provide supporting evidence for causality.
Our classification of “sufficient evidence of an association” occurs when a positive association has been observed in several epidemiological studies but biases (including confounding) that could explain all or most of the associations cannot be reasonably ruled out. This is similar to the NTP classification of “reasonably anticipated to be a human carcinogen” in which the evidence from human studies “indicates that causal interpretation is credible but that alternative explanations, such as chance, bias or confounding factors, could not adequately be excluded.”
Our classification of “Limited/Suggestive evidence of an association” is when the evidence from epidemiological studies is weak. This could occur if there are conflicting findings among well-conducted studies; or the studies that have positive findings have serious limitations; or the number of studies are too few to support a stronger classification.
Literature Search Methods
Reviews of epidemiological studies involving TCE and PCE exposure have been conducted by EPA (2011), IARC (2014) and NTP (2015). In addition, meta-analyses have recently been conducted by NCI (Karami et al 2012, Karami et al 2013), EPA (Scott 2011), and IARC researchers (Vlaanderen et al 2014) for TCE and kidney cancer, hematopoietic cancers and liver cancer, and PCE and bladder cancer. ATSDR utilized these reviews and meta-analyses to identify relevant epidemiological studies for TCE and PCE. Meta-analyses of benzene and hematopoietic cancers (Khalade et al 2010, Vlaanderen et al 2011, 2012) were used to identify relevant epidemiological studies for benzene. In addition, literature searches using PubMed were conducted to identify epidemiological studies conducted after the meta-analyses and reviews were completed, using the following keywords: trichloroethylene, tetrachloroethylene, perchloroethylene, and benzene. For vinyl chloride, we reviewed the IARC monograph 100F (2012) that evaluated vinyl chloride and conducted a literature search using PubMed with the key word, vinyl chloride.
Assessment of the Evidence
ATSDR reviewed the journal articles (including meta-analyses) identified through the literature search as well as the overall reviews published by EPA (2011), IARC (2012, 2014) and NTP (2015). Tables for each disease were developed listing the results of: (1) any meta-analysis that was conducted, (2) epidemiological studies not included in a meta-analysis, and (3) epidemiological studies that were included in a meta-analysis that had additional information on exposure duration, exposure intensity, and/or cumulative exposure. After each table, any assessments made by EPA, IARC, or NTP were presented along with ATSDR’s assessment of the evidence. Although the evidence from epidemiological studies (in particular meta-analyses, pooled analyses, and those studies considered well-conducted) is emphasized, the findings from animal studies, mechanistic studies, and other toxicological information is mentioned if necessary to support our conclusions. If there is evidence concerning the relationship between risk of the disease and duration or level of exposure, this is presented. Also included is the 5-year survival percent and whether a study evaluated mortality or incidence. For diseases with a high 5-year survival percentage, an incidence study would have a greater capability than a mortality study of evaluating the risk from exposure to these chemicals for several reasons: (1) the exposure may cause a less fatal form of the disease; (2) a cancer that is not an underlying or contributing cause of death will be missed in a mortality study, and in general, there will be many more incident cases than mortality cases so precision should be improved (i.e., width of confidence intervals will be narrower) in an incidence study; and (3) there is greater accuracy of the cancer information provided by cancer registries (e.g., histological information and identification of primary and metastatic sites) compared to the information available from the death certificate, so disease misclassification should be reduced in incidence studies.
In the disease-specific tables, 95% confidence intervals are provided for the key findings in a study to indicate the level of precision or uncertainty in the effect estimates. We discourage the use of confidence intervals as tests of statistical significance. For the information on exposure duration, we generally do not provide the confidence intervals unless the findings are from a meta-analysis or pooled analysis because the focus is on the intervals were risks appear elevated and the concern is not about the precision of the estimates. Similarly for cumulative exposure and exposure intensity, we generally do not provide confidence intervals unless this is the key finding in a study.
The vast majority of the relevant studies are occupational studies. The key limitation of all the studies was exposure misclassification. Most of the occupational studies utilized job-exposure matrices of varying quality. A few studies had information on urinary TCA levels that improved the exposure assessment. The impact of exposure misclassification bias would likely be to bias dichotomous comparisons (e.g., exposed vs unexposed) towards the null if an effect of the exposure is truly present, and to distort exposure-response trends (e.g., the curve may flatten or attenuate at high exposure levels).
Healthy worker/veteran effect bias likely occurred in studies that compared incidence or mortality rates in worker or veteran cohorts with rates in the general population. Such a bias would tend to produce underestimates of the effect of exposure, and in many situations, reduce measures of association (e.g., SIR or SMR) below the null value.
Another issue for most of the studies is confounding due to co-exposures to other workplace chemicals. For example, dry cleaning workers employed before the early 1960s were likely exposed to other solvents besides PCE. Dry cleaning workers also used solvents for spot removal although these exposures would be considerably lower than exposures to the primary solvent. Workers in aircraft manufacturing or maintenance may have been exposed to TCE, PCE and other solvents. In the Camp Lejeune studies and the NJ studies, both TCE and PCE appeared together as drinking water contaminants.
An additional concern raised by IARC and NTP was confounding by other risk factors such as smoking and alcohol consumption. However, for appreciable confounding by smoking or any other risk factor to occur, at least two requirements must be met: (1) the risk factor must have an association with the outcome of interest at least as strong as the exposure of interest, and (2) the risk factor must also have a strong association with the exposure of interest. For the latter requirement to be met, the prevalence of the risk factor must be very different in the compared groups. This might occur for example when a worker (or veteran) cohort is compared to the general population. However, the prevalence of risk factors (other than the exposure of interest) should be similar when comparisons are made either internal to a cohort or between similar cohorts (e.g., similar workforces or similar military personnel), and therefore confounding would be expected to be minimal for these comparisons.
In general, substantial confounding due to smoking or any other risk factor is rare in occupational and environmental epidemiology. Even for studies of an occupational or environmental exposure and lung cancer, a summary measure (e.g., RR, OR) adjusted for smoking rarely differs by more than 20% from the unadjusted summary measure (Blair et al 2007). In any case, the amount of bias due to confounding will not be greater than the weaker of these two associations: (1) between the exposure of interest and the confounder; (2) between the confounder and the disease of interest (Smith and Kriebel 2010).
Many of the studies included in the meta-analyses or listed in the tables did have information on smoking and were able to evaluate whether confounding due to smoking was present and affected the results. Most of the studies that did not have information on smoking were able to indirectly assess whether confounding due to smoking affected the results by evaluating whether a smoking-related disease that was not known to be associated with the exposure of interest was elevated in the study. Another indirect approach to evaluate possible confounding due to smoking would be to evaluate all smoking-related diseases in the study for which the risk from smoking is known (or expected to be) much larger than the risk from the exposure of interest. If appreciable confounding due to smoking were present, one would expect that all these diseases would be elevated.
Many of the studies evaluated, or adjusted for, risk factors in addition to smoking such as alcohol consumption and SES factors. The appendix lists the studies included in the tables, whether or not they evaluate smoking as a possible confounder, and any additional potential confounders. The appendix also provides comments on the quality of many of the studies included in the table from reviews conducted by IARC (2014) and NTP (2015).
Duration of Exposure
There is limited information on the minimum duration of exposure necessary to cause diseases related to the drinking water contaminants at Camp Lejeune. Moreover, even when duration information is provided in a study, it is often categorized into wide ranges (e.g., > 0 to 5 years). An additional difficulty is the possible inverse relationship between duration and exposure intensity, e.g., high exposure intensities may require only a short duration of exposure whereas low exposure intensities may require longer exposure durations. Although cumulative exposure is a useful metric, it obscures this interplay between duration and intensity. Specifying a minimum duration of exposure also presupposes that there is a known threshold amount of exposure below which there is no excess risk. However, there is no compelling evidence that such thresholds exist for these contaminants and specific cancers.
The 2012 Honoring America’s Veterans and Caring for Camp Lejeune Families Act established a minimum duration at Camp Lejeune of 30 days in order to be eligible for health benefits under the Act. The evidence from the epidemiological studies included in the tables is not sufficient to contradict this minimum duration. Moreover the evidence from the Camp Lejeune mortality studies tends to support a 30 day minimum duration with elevated risks for many of the diseases occurring for an exposure duration of 1-3 months.
For cardiac defects, it is possible that durations of exposure to the mother as short as 1 day may be sufficient if the exposure occurs during the relevant vulnerability period for cardiac defects, i.e., 3-9 months gestation. Very short in-utero exposures (i.e., less than a month) may also be sufficient to cause childhood leukemia.
Given the insufficient evidence for a threshold level of exposure to these contaminants, it may be helpful to explore how other programs have resolved this issue when information on exposure duration or evidence for a threshold level are lacking. For example, although the exposures are very different from those considered in this document, the World Trade Center (WTC) Health Program uses site-specific minimum exposure durations ranging from 4 hours to a maximum of 400 hours.
Given that sufficient evidence for a threshold is lacking, ATSDR recognizes that a decision to establish a specific minimum exposure duration for presumption will primarily be based on social, economic and legal factors. It is ATSDR’s position that the minimum exposure duration of one month in the 2012 Honoring America’s Veterans and Caring for Camp Lejeune Families Act is an appropriate minimum exposure duration and should be considered by the VA in developing its program for presumption at Camp Lejeune.
Link to Full Document: ATSDR-Summary-Evidence-for-VA-Presumptive-Lejeune-09-2015-Draft