Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
3 Evaluation of Epidemiologic and Clinical Methods EVALUATION OF EPIDEMIOLOGIC METHODS The HTDS investigators examined 3,441 persons who were born in 1940-1945 to parents living in the study areas near Hanford. Of these, 3,190 lived in the study areas at some time from 1945 to 1957, so thyroid dose estimates for them could be derived with the CIDER program developed by the HEDR investigators. Dose reconstruction used the available data and accepted methods. The study design was appropriate to address the aims of the study. Among the possible designs, a cohort study with accompanying thyroid screening is optimal for minimizing the potential for bias. The investigators also chose the optimal populations to study: children who were young at the time of the greatest ]3~{ releases and who lived in the most highly exposed areas. Those two choices maximized the potential of the study to detect thyroid-disease effects. The interval between exposure and thyroid screening was adequate to allow radiation-induced thyroid diseases to become evident. The epidemiologic methods were exceptionally good. The sample was based on an almost complete census of eligible subjects who were born in selected years and lived in what are believed to be the high-dose regions and subjects born in the same years who lived in regions where the doses from Hanford releases were lower. Efforts to locate subjects and to elicit study 49
50 Review of the HTDS Draft Final Report participation were thorough. The investigators were able to locate 94°/0 of the targeted sample-an excellent result, considering that 40-50 years had elapsed. Since the time of exposure they succeeded in communicating with about 98% of potential participants by telephone and had 15°/0 refusals to participate. Thus, their losses clue to noniocation, noncontact and refusals totaled only about 23%. That is an excellent rate for a study that requires people to come in for examination, especially given their wide geographic dispersal. Furthermore, the losses to the study were comparable among those with high and low estimated doses, so participation rates were similar across the dose range. There was a high level of quality control in the epidemiologic procedures. The interviewers were carefully trained by experienced interviewers, and detailed interview manuals were developed. Care was taken with data entry (double entry was used routinely), and range checks and consistency checks were implemented to help to reveal errors in the codes entered by the interviewers. Callbacks were used when there were clear errors or missing data. Care was taken to maintain "blinding" in the clinical examination and other parts of the study where it could reasonably be done, to ensure that selection bias, interviewer-induced response bias, or clinical-examination bias would not creep in. For example, the residence-milk interviews with parents or the subjects to obtain residence and milk consumption-rate histories were conducted before, clinical examination so that neither subjects nor interviewers would know subjects thyroid status (except in the case of previously diagnosed thyroid disease). in short, the HTDS was designed with great care to eliminate bias due to selection of subjects and due to reporting and detection of disease. This study compares favorably with other epidemiologic studies in these respects. The HTDS considered an appropriate set of potential confounding variables, including sex; age at first exposure to ]3~i from Hanford; age at examination; history of diagnostic,
Evaluation of Epidemiologic and Clinical Methods 51 therapeutic, and occupational radiation exposures; smoking history; ethnicity; and estimated thyroid dose from the NTS. A few other variables could have been considered, such as number of childbirths (for women) and family history of thyroid disease, but it seems improbable that these would have substantially altered the results because it is unlikely that they would be differentially distributed across the dose range. Another potential confounding variable was global fallout, which is discussed in chapter 4. MILK-CONSUMPTION ESTIMATION The consumption of fresh milk is an essential part of the pathway Tom release of }3~{ to thyroid dose, so information on the amount and sources (for example, backyard cow versus commercial dairy) of milk consumed is important in estimating the individual doses. A questionnaire (called the CATI, for computer- assisted telephone interview, in the HTDS Draft Final Report) was developed to obtain information on residence history and milk sources and amounts during 1944-1957 from the mother or a surrogate (such as father or older sibling) who would presumably know about a subject's dietary patterns in childhood. The questionnaire was carefully developed and went through numerous revisions and multiple field tests. Professionals with expertise in questionnaire development commented on ways of eliciting questionnaire information to maximize the completeness and accuracy of recall. The development process was about as good as it could be. Although the body of the HTDS Draft Final Report discusses how the milk-consumption questionnaire was developed and tested and attachments at the end of the report deal with it, there is not enough explanation of how open-ended information was coded. In addition, more needs to be stated about how much probing was used during the interviews, whether the booklet devised to stimulate memories of events that occurred 40 years earlier worked effectively and whether such memories were
52 Review of the HTDS Draft Final Report checked for reliability in a subsample (for example, by comparing reports of the two parents). The accuracy of such information, obtained 40-50 years after the fact, is unknown. Food-consumption reporting that uses questionnaire-based methods is recognized as error-prone. Assessment (validation) of questionnaire efficacy in estimating true dietary intake generally requires careful study as a separate issue (Willett, 1990) before the questionnaire is used to relate food intake to risk of disease. The HTDS investigators were not able to evaluate the questionnaire's validity, nor did they cite any evidence from other studies about the validity of childhood milk- consumption reports decades after childhood. Milk consumption is often found to be among the better-reported elements of recent diet (for example, in the preceding year) when methods typical in epidemiology studies are used (Salvini and others, 1989), but we are aware of no studies that directly investigated the reliability or validity of retrospective milk-consumption reports by surrogates. A few studies have considered self-recall of diet after considerable periods. Dwyer and others (1989), considering retrospective recall of childhood (age 5-9 years) food consumption by 72 middle-aged subjects who had originally been assessed as children, reported a correlation of 0.3 between the retrospective reports of whole-milk consumption and food histories taken during childhood. It seems credible that mothers' reporting of their children's early milk-consumption habits could be as good as the Dwyer group's findings or possibly somewhat better, but reliance on siblings or other relatives for information about early milk consumption seems unlikely to be better than the self-reports studied. Furthermore, a correlation of 0.3 does not indicate good predictiveness of childhood eating habits. The HTDS investigators made a strong effort to obtain information from parents or surrogates. We would expect the mother to be the best source of childhood milk-drinking information. However, for 26% of the interviews, the investigators had to rely on information from some other family member.
Evaluation of Epidemiologic and Clinical Methods 53 Furthermore, because of parent deaths and other factors, they were unable to obtain any interview for 38°/0 of the subjects. For the 38°/0, they obtained from the subjects themselves a residence history and at least the sources (backyard cow, commercial dairy, and so on) of milk drunk during 1944-1957, but not such details as the amount consumed. Thus, they had to use default values for a substantial fraction of the subjects, and this probably introduced measurement error into the data analyses. The subcommittee's evaluation of the milk- consumption assessment is that the investigators did the best they could under the circumstances but that the resulting data have high intrinsic uncertainty. The effect of that uncertainty on the statistical power of the study is discussed in chapter 6 of this report. The HTDS report has a substantial description of the collection of dose-related data from people but relatively little information on how these data were used. The input to the CIDER program is described as "scenarios", but these are not explicitly described, nor is their construction from the data. That is separate from how the CIDER program uses the scenarios to generate doses. There are a number of references to the use of default values in the CIDER program, but there is no discussion of which parameters used default values or of the degree to which default values changed as life circumstances changed for a given person (for example, if a person moved from a farm to a city). EVALUATION OF CLINICAL-DATA COLLECTION The clinical examinations and laboratory studies used the best modern methods of detecting and defining thyroid disease. Subjects were given physical examinations, including thyroid palpation by thyroid specialists, an ultrasonographic examination, and appropriate thyroid-hormone and thyroid-antibody blood tests. Quality control of the laboratory tests and ultrasonographic examinations was good. The clinicians were kept blinded to subjects' residential or milk-consumption history to avoid possible subtle clinical biases.
54 Review of the HTDS Draft Final Report The study could have been strengthened if additional information on the following clinical and pathologic aspects had been gathered and presented: Added confidence is needed in the cytopathology data. The cytopathologic interpretation of thyroid fine-needle aspirations (FNAs) is a key step in distinguishing benign and malignant nodules. It would be useful to have the slides reviewed by two or three cytopathologists who are expert in thyroid disease and to have a consensus diagnosis when differences in interpretation are encountered. Special attention to the category of acellular and hypocellular aspirations that contain colloid is needed. They were categorized as benign, and that is not typical clinical practice. What constitutes an adequate biopsy should be defined in terms of numbers of cells and preparation technique. Some subjects had nodules that were not biopsied during the course of the study. For people with nodules greater than ~ cm in diameter at last observation, a followup examination including an ultrasonography to look for nodule progression could be useful. if progression were detected, an FNA would be useful to document the characteristics of the lesion. A tabular presentation of the pathways to diagnosis would help readers to assess how the final diagnoses were assigned. For each clinical outcome, there is more than one way to make the assignment. For some, such as hyperparathyroidism, this is straightforward (either high calcium with high parathyroid hormone or a confirmed diagnosis before the study). For others, such as thyroid cancer, it is more complex (combinations of palpable nodule smaller than I.5 cm, palpable nodule of at least I.5 cm, nonpalpable nodule of at least I.5 cm, FNA, surgery, prior diagnosis, cancer
Evaluation of Epidemiologic and Clinical Methods 55 in largest nodule, and cancer as an incidental surgery finding). A table for each diagnosis with a list of the methods of diagnosis and the number of each instance should be included in the full report, and these data should be looked at for any indication of unsuspected ascertainment bias. . The public has expressed concern that the HTDS analyses of the clinical data have "missed the forest for the trees"; that is, examining the fine categories of diagnoses might have caused the data analyst to miss trends, that occurred in broader categories of thyroid disease. Inasmuch as autoimmune thyroiditis, Graves disease, otherwise unexplained hypothyroidism, and ultrasonographic texture changes all are associated with autoimmune processes, one could score a person with any or all of these as positive for a new global variable of autoimmune thyroid disease (a broader category than the one by the same name in the HTDS). Similarly, the variable "any thyroid nodule", which was already analyzed in the HTDS report (tables VITI-53 to VTIl-55), is a global variable of nodular thyroid disease. Any evidence of hyperparathyroidism or abnormal mineral metabolism could constitute another global category of disease. These variables would give a broad-brush view of thyroid disease in relation to |3~{ dose, which would help to ensure that the fine diagnostic categories used in the HTDS report did not miss possible variations in broad categories of thyroid disease. However, a still broader category of "any thyroid disease" is not recommended, because combining pathophysiologically unrelated outcomes lacks biologic plausibility. COMPLETENESS OF ASCERTAINMENT OF THYROID DISEASE One possible weakness in the ascertainment of pre- existing thyroid cancers results from the HTDS investigators' inability to locate death certificates for all the decedents. Causes of death were not ascertained for 39 of 541 deaths. Given the small
56 Review of the HTDS Draft Final Report number of thyroid cancers in the study (only 14 had estimated doses and could therefore be used in the dose-response analysis), a thyroid-cancer death among the unlocated cases could affect the results. This possibility cannot be ruled out, but it seems improbable that there would be any thyroid-cancer deaths among the 39 decedents for whom a death certificate or medical records where absent. First, there were no thyroid-cancer deaths among the 502 whose cause of death was known, so finding one among the 39 seems unlikely. Second, the probability of dying from thyroid cancer by the age of 55 is very small: the rate is about 4 per ~ 00,000 persons among both males and females, and thyroid cancer accounts for every 1,000 deaths in the age range 0-54 years. Even if one triples the rate to account for a possible radiation effect, the probability is still very small that a thyroid-cancer death would be among the 39 whose death certificates were absent. Thyroid cancer could also be listed as a contributory condition on the death certificate, but this is rather unlikely unless it was part of the chain of disease leading to death, which again would have a low probability. A consultant to our subcommittee indicated that, of the ~ 19 thyroid-cancer cases he had seen that were diagnosed before the age of 20 and had an average of 20 years of followup after diagnosis, only one led to death from thyroid cancer (Ernest Mazzafern, personal communication); this further confirms the low likelihood of missed thyroid-cancer deaths. Data on other thyroid diseases, but probably not thyroid cancer, can also be lacking because medical records are missing. The investigators reported that 37°/O of the 1,264 medical records they sought could not be obtained and that they were not able to obtain pathology or cytology slides for 10 of the 52 people on whom they sought them. it would be desirable for them to indicate for how many potential thyroid diagnoses they were unable to obtain any medical confirmation, preferably with a breakdown by reported type of thyroid disease.
Evaluation of Epidemiologic ant! Clinical Methods SUSCEPTIBILITY FACTORS FOR THYROID DISEASE 57 At the various public meetings, several people who lived in downwind areas stated that they and their families had experienced more-frequent thyroid diseases than would have been expected in the population at large. They could be right, and their disease could have been the result of unusual fallout or ingestion patterns. However, thyroid disease does tend to run in families, and the particular occurrences could be related to genetic factors within the families, chance occurrences, or even mistaken diagnoses. The findings by several research groups that many of the thyroid cancers being found in Belarus and Ukraine, downwind of Chernobyl, have relatively unique ptc3 mutations is one line of evidence for genetic factors in the disease following ~3~} exposure. An enumeration and study of such clusters could have been undertaken but were not. Thyroid cancer is not a common disease, and it would be reasonable in future epidemiology surveys to identify, document, and investigate clusters with molecular-biology probes to characterize genetic polymorphisms that could make people more sensitive to ionizing radiation or to look for oncogene prevalence in affected subgroups. These methods are developing rapidly, and will probably play a role in future environmental- epidemiology studies. EVALUATION OF MORTALITY DATA As part of the study protocol, death certificates for members of the Hanford cohort who had died were obtained. On the basis of those early deaths, the investigators calculated a standardized mortality ratio (SMR) for each cause of death. Overall mortality in the Hanford cohort was 20% higher than that in Washington state, which served as the reference population. The increase was due largely to deaths from congenital anomalies and "conditions in the perinatal period", although the SMR for cardiovascular disease was also somewhat increased. in an analysis
58 Review of the HTDS Draft Fzr~al: Report by geographic area, the highest SMR (close to 2.0) was observed in Franklin County. Some of the initial findings raise suspicion about the contribution of radioactive releases from Hanford. The particular vulnerability of the fetus and infant to adverse effects of radiation is well known, and the mortality experience of this group in the Hanford cohort was somewhat unusual. Finding the highest mortality in the county that is closest to the Hanford site is also suggestive. To follow up those observations, the investigators reanalyzed the data, breaking the total period into times before and after the peak releases of AT (between March and November 1945) and categorizing people accordingly. In the Draft Final Report, people are categorized by year of birth. A supplemental table later provided to the committee by the investigators categorized people by year of death as well. For congenital anomalies and conditions of the perinatal penod, the year of birth and year of death are almost always the same. Table 3.! compares the results of the analyses by year of birth and by year of death for the two causes of death. Table 3. ~ Standardized Mortality Ratios for Selected Causes of Death by Birth and Death Years SMR (95% Confidence Interval) . . Year of Birth Year of Death Cause of Death 1940-1944 1945-1946 1940-1944 1945+ Congenital ~ .55 ~ .3 ~ .78 ~ . ~ 9 anomalies (1.06-2.19) (0.63- 2.41) (1.15-2.63) (0.69-1.90) Conditions ofthe 1.73 2.86 2.~8 1.93 perinatal (~.33-2.21) (2.09-3.83) (~.68-2.78) (~.41-2.59) period
Evaluation of Epiclemiologic and Clinical Methods 59 In both versions of the analysis whether data are arrayed by year of birth or year of death SMRs in the period both before the Hanford releases (before 1945) and after 1945 are elevated. For congential anomalies, the excess are slightly larger before 1945 than after, and the same holds true for deaths due to conditions in the perinatal period when considered by year of death. On the basis of the analysis, there appears to be no evidence that 13~} iS responsible for the increased mortality from these causes in the Hanford cohort. However, when viewed by birth year, the SMRs for perinatal deaths do appear to be significantly higher in the period after the beginning of Hanford releases than before. The discrepancy between the two views of the same data is puzzling and needs further explication. There remain additional questions that the Hanford investigators could answer to resolve lingering questions: What is the distribution of types of anomalies and of "conditions of the perinatal period" in the two periods? The question of whether neural-tube defects are in excess around the Hanford site has been raised by another study (Sever and others, 1988) and the information available from the current study should be presented. The suggestion of a possibly larger excess in the 1945-1946 birth cohort than in the 1940-1944 birth cohort indicates that a more detailed presentation of the data is warranted. . To what extent is the excess mortality associated with these two causes independent of the geographic excess in Frar~klin County? AMERICAN INDIAN TRIBAL ISSUES The subcommittee had some concerns about the extremely low number of American Indians in the study. The low compliance or cooperation might not be completely the fault of the .
60 Review of the HTDS Draft Final Report American Indian tribes. There are established methods for approaching and establishing rapport and trust with ethnic communities. The subcommittee wonders whether all these methods were tried. Perhaps a review of the methods used would be helpful in a future study. A second concern is that American Indians had additional pathways of exposure to radiation from the Hanford site. For example, some tribes consumed much fish from the Columbia River. The river was contaminated by various radionuclides because it was used to coo! the eight single-pass production reactors, in addition to the periodically increased 13~{ that would have reached the river from rain and washoff from soil and vegetation. However, the thyroid dose resulting from aquatic pathways was likely to have been much smaller than that due to the consumption of fresh milk or leafy vegetables.
