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.
8 Feasibility of the Study of Adverse Reproductive Outcomes in the Families of Veterans Exposed to ionizing Radiation The feasibility of a study of adverse reproductive outcomes among the families of veterans exposed to ionizing radiation hinges largely on the answers to five interrelated questions. First, how is a suitable sample or cohort -(numerators and denominators) to be defined, and can this be done without inad- vertently introducing selection biases that could obscure a true effect or produce a spurious one? Second, what will be the probable size of that sample or cohort? And, as a corollary, will that size be large enough to reveal effects of the magni- tude anticipated on the basis of current knowledge? Third, what is the probable dose distribution among the members of that sample or cohort? Fourth, how reliable are the individual dose estimates? Fifth, what mechanisms are available for identifying adverse reproductive outcomes? Each question will be consid- ered separately. DEFINITION OF A SUITABLE SAMPLE OR COHORT Anecdotal information can be valuable in establishing the need for an epi- demiologic study, but self-volunteered information is unlikely to provide a basis for reliable estimates of risk since experience shows that persons with a personal or even financial interest in an exposure to some hazard will selectively respond. Accordingly, a scientifically defensible and valid study of the effects of ionizing radiation on reproductive outcomes depends on the availability of a representa 62
FEASIBILITY OF THE STUD Y 63 live sample of exposed veterans and their families, and the means to establish these outcomes without reference to whether they were normal or abnormal. The Nuclear Test Personnel Review (NTPR) program of the Defense Nuclear Agency (DNA) has identified some 210,000 veterans who participated in one or more atmospheric tests involving the detonation of a nuclear weapon. Those individuals or a suitably large and representative sample might provide the basis for a study cohort, and it seems probable that deaths among these veterans could be determined through the records of the Department of Veterans Affairs or other sources. However, it is far more difficult to trace an unbiased sample of living persons, given the lack of identifying information in the original records. Furthermore, the available records do not contain information on the reproduc- tive histories of the veterans, that is, their children, estimated to be about 500,000 in number, and grandchildren, and for reasons adduced elsewhere in this report it seems doubtful that such information can be reliably and accurately obtained at this late date. Thus, the committee concludes that, whereas a study of the life status and health problems of the veterans themselves is feasible (and is in fact being done), the means do not exist to obtain information on adverse reproductive outcomes among their children and grandchildren in a suitably complete and unbiased manner to estimate the risk, if any, stemming from expo- sure to ionizing radiation. SIZE OF THE SAMPLE OR COHORT REQUIRED To determine the size of an epidemiologic study seeking to respond to the concerns of the Atomic Veterans, two somewhat different, but related questions can be posed: 1. If current estimates of risk are correct, how large a sample would be needed to demonstrate that risk? 2. Given the sample size that might be available, how large would the risk have to be to be demonstrable? The committee has sought answers to both of these questions. The results follow. Question 1 Table 7 provides a comparison of certain characteristics, relevant to the feasibility of an epidemiologic study, of the Japanese atomic bomb survivors and their children with the Atomic Veterans and their children The main differences" are average dose and dose range, the number of children potentially available for study, and the environmental conditions existing after birth. Although the po- tentially larger sample of children from the Atomic Veterans is favorable to an
64 ADVERSE REPRODUCTIVE OUTCOMES epidemiologic study, the much lower average dose and limited range of doses more than offsets this advantage in numbers. TABLE 7. Comparison of Characteristics of Hiroshima-Nagasaki Atomic Bomb Survivors with Those of Atomic Veterans ~c ~ Hiroshima-Nagasaki Survivors 0.44 (conjoint) 0~.0 (conjoint) 52,000 11,000 both parents 18,000 mother only 6,000 father only 17,000 neither parent At home Adverse . Atomic Veterans Average dose (Sv) Dose range (Sv) Number of children Parent. exposed Births Postnatal environment 0.006 0-0.03~ soo,ooo Father only In hospital Normal conditions " A total of 95% of the Atomic Veterans received doses in this range. In discussing the feasibility of studying the children of the Atomic Veterans in order to estimate the genetic effects of radiation exposure, it may be useful to rewrite Equation 1 (page 31) as follows: (S+I)/S= 1/DDxMCxD+ 1 . (Equation 3) This form gives the expected total number of cases in the exposed group (S + I) divided by the number expected without the added exposure (S). This is the relative risk, RR (see Chapter 21. The genetic risks associated with low doses of ionizing radiation are quite small when compared with the background incidence of genetic disease in the general population. To give the reader a feel for the magnitude of these risks, three quantities are estimated: (1) an upper-bound relative risk for major con- genital abnormalities in the children of all Atomic Veterans, (2) a best-estimate relative risk for genetic disease in general in the children of all Atomic Veterans, and (3) a best-estimate relative risk for genetic disease in general in the children of Atomic Veterans who were exposed to 100 mSv (10 rem) or more. What is the maximum relative risk one could expect to find among the Atomic Veterans' children given what is known about the effects of ionizing radiation from studies in animals and humans? Consider, for example, the ap- pearance of major congenital malformations among the children of exposed Atomic Veterans. From the data on mice and humans reviewed above, the doubling dose is certainly expected to be no less than 250 mSv (25 rem) (100 rem is the usual estimate for a low dose). Assume a mutational component of 1/2 (BEIR V [NRC, 1990] assumed a maximum mutation component of 0.35 for this endpoint) and assume that 1/10 of the effect will appear in the first genera- tion following exposure (again, this is greater than the most probable value). In addition, assume that, on average, the Atomic Veterans were exposed to 20 mSv
FEASIBILITY OF THE STUDY 65 (2 rem) (the current estimate is 6 mSv [0.6 remit. Finally, the fact that only fa- thers were exposed must be taken into account. The estimate of the maximum relative risk (RR) to be expected is then RR = (S + I)/S = (1/25~1/23~1/10~1/2~2) + 1 = 1.002 . This is a very small relative risk. It means that the increase in the incidence of major congenital malformations due to the ionizing radiation exposure is ex- pected to be only 0.002 of the background incidence normally seen in the popu- lation. This is an upper limit based on extreme assumptions that were all chosen to overestimate the risk. Suppose the 210,000 Atomic Veterans had a total of 500,000 children. Without any radiation exposure to the fathers, about 15,000 of these children would be expected to be affected by a major birth defect; with an additional average exposure of 2 rem the expectation, under these extreme assumptions, is 15,025 affected children. In fact, the BEIR V "best" risk esti- mate for major congenital malformations implies a relative risk of only 1.0004 in the first generation, that is, only five additional cases of malformations due to the radiation. BEIR V (NRC, 1990) estimated genetic risks from ionizing radiation for seven kinds of disorders: clinically severe and mild autosomal dominant, X linked, autosomal recessive, unbalanced translocations, trisomies, and congenital abnormalities. For all seven endpoints combined the relative risk from the BEIR V risk estimates in Table 2-1 of that report is less than 1.0006 for first- genera- tion effects (assuming 1 rem of exposure to the male). Again, suppose that the 210,000 Atomic Veterans had a total of 500,000 children. Without any radiation exposure to the fathers, about 21,150 of these children would be expected to be affected by some disorder in these seven categories; with an additional average exposure of 1 rem the expectation is 21,164 affected children an increase of only 14 children. Finally, consider the relative risk for the children of Atomic Veterans who received 10 rem or more. According to present dose information, this 0.07% of the total cohort has an average dose of approximately 20 rem. For the genetic risks in general (seven endpoints combined) the expected relative risk for this subset of Atomic Veterans the most heavily exposed is 1.012. Suppose that these men had 350 children. Without any radiation exposure to the fathers, about 15 of these children would be expected to be affected by some disorder in these seven categories; with the additional 20 rem the expectation is 15.18 af- fected children less than one additional child. A similar calculation can be made using the "direct method" explained in Chapter 6. In Table 1 of Appendix G of UNSCEAR 1993, the incidence of ge- netic or partially genetic disease having serious health consequences before the age of 25 years is estimated to be approximately 79,400 per million live births. UNSCEAR estimates an additional 15-30 seriously affected individuals per mil
66 ADVERSEREPRODUCTIVE OUTCOMES lion born to fathers exposed to 0.01 Gy of low-LET radiation. Taking thirty as an upper estimate, multiplying by 3 to apply to a high dose rate response as a worst case, and multiplying by 20, about 1,800 additional seriously affected off- spring per million children in the first generation following exposure of their fathers are expected. This corresponds to a relative risk of approximately 1.023. Thus of the estimated 350 children born to fathers exposed to an average of 20 rem, 27.79 would be expected to be affected without the radiation and 28.43 with the additional radiation. Again, less than one additional case would be ex- pected due to the radiation. The committee did not count the large genetic disease burden usually re- ferred to as diseases of complex etiology, but these are expected to have relative risks at least as small as the ones considered here. Also, the risks to the grand- children of the Atomic Veterans will be even smaller. It should be clear from the examples given here that effects of this magni- tude are not measurable in any attainable sample, because the induced cases make up a miniscule part of the spontaneous burden of human genetic disease and are indistinguishable from the naturally occurring cases. If this estimate of the probable relative risk is correct, or nearly so, the sample size needed would run into several millions of children (or grandchil- dren), a number that is certainly beyond the limits of feasibility for an epidemi- ologic study. This can be shown more formally. To do so, however, the com- mittee digresses briefly for some background remarks. When a statistical test is performed, it is done in the context of a hypothesis. Commonly, that hypothesis is known to statisticians as the null hypothesis; it postulates that there is no difference between the groups under study. On the basis of the results of the test performed, that hypothesis is either not rejected or rejected. If the null hypothesis is not rejected, it is equivalent to asserting that any difference between the two groups may be due to chance. If the null hy- pothesis is rejected, the difference between the groups is possibly a consequence of the exposure under study. For example, if a group exposed to ionizing radia- tion was compared with one not exposed and the null hypothesis is rejected, it is possible to conclude that radiation may be associated with the health outcome. The likelihood of observing an association between the exposure and the out- come depends upon the sample sizes involved, the magnitude of the difference between the two groups, and the errors of interpretation of the data an investiga- tor is willing to accept. These errors are of two types. First, the investigator may reject the null hypothesis, that is, conclude that the difference is probably not due to chance when, in fact, it is. This is commonly referred to as a type I error, and in computing the sample size needed to demonstrate a particular risk one must set the acceptable probability of such an error (usually 5 or 10%, or more, generally designated or). Second, the investigator may fail to reject the null hypothesis (no difference between the compared groups) when, in fact, there is a difference between the groups. This is termed a type II error, and the prob
FEASIBILITY OF THE STUDY 67 ability of such an error is often set at 20% (and is designated Q). It warrants noting that the complement of this latter error rate (i.e., 1 - p) is called the power of the test, which is 80% in the example. Table 8 sets out the sample sizes required to demonstrate specific relative risks assuming different background frequencies of an event and error rates of 10% (type I) and 20% (type II). There is a functional relationship between the sample size, the difference that is obtained between two (or more) groups under comparison, and the fre- quencies of the two types of errors. This relationship is such that if any three of these four values are known (or the investigator is prepared to assume their val- ues), the fourth is also known. By using this relationship, it is possible to com- pute that fourth value. For example, if one sets the frequencies of the two types of errors and the relative risk (which is the difference between the two groups of interest), one can compute the required sample size. Table 8 sets out the results of such computations in the present context. The specific methodology (Statistics and Epidemiology Research Corporation, 1993) used is based on the detection of a statistically significant trend (dose-response) test assuming relative risks from 1.5~.0 for the highest dose category (>100 mSv t>10 rem]) and for various outcome frequencies. TABLE 8. Sample Sizes Required to Detect a Range of Relative Risks for the Highest Dose Category >100 mSv (>10 rem) in a Test for TrendU Sample Size for Relative Risk of: Frequency of Outcome(%) 1.5 2.0 3.0 4.0 0.1 2,8 1 6,000 868,000 329,000 1 80,000 0.5 580,000 1 82,000 64,000 36,000 2.0 1 48,000 47,000 1 6,000 9,000 3.0 1 00,000 32,000 1 1,000 6,000 " The estimated sample sizes were calculated for those in the highest dose category (>10 rem), with type 1 and type 2 errors of 10 and 20, respectively. The sample sizes have been rounded to the nearest thousand to avoid an undue perception of accuracy. h Frequency of major congenital defects in the general (unexposed) population. From Table 8, assuming a background rate of 3% for major congenital birth defects present at birth, a total study population of 100,000 individuals (unexposed and exposed) would be required to detect a relative risk of 1.5. Therefore, if the relative risk is, in fact, considerably less (i.e., 1.002), the sam- ple size would be in the millions. Question 2 If the circumstances are such that an investigator has little control over the size of the study groups, then he or she might ask what difference could be dem- onstrated with a particular sample size. The computations are very similar to
68 AD VERSE REPRODUCTI VE OUTCOMES those underlying Table 8, but now the sample size and the frequencies of the two types of errors alluded to above are assumed to be known and it is the relative risk that is to be calculated. In Table 9 the committee presents a series of such calculations for various adverse reproductive outcome frequencies (f) and sam- ple sizes (N). TABLE 9. Minimum Detectable Relative Risk for a Range of Sample Sizes for the Highest Dose Category >100 mSv (>10 rem)a Relative Risk for Samole Size of: Frequency of Outcome (%) 80,000 140,000 210,000 500,000 0.1 6.7 4.7 3.6 2.5 0.5 3.7 2.2 1.9 1.5 1.0 2.1 1.7 1.6 1.4 2.0 1.7 1.5 1.4 1.3 3.0 1.5 1.4 1.3 1.3 . " The estimated relative risks were calculated for those in the highest dose category (>10 rem), with type 1 and type 2 errors of 10 and 20, respectively. h Frequency of major congenital defects in the general (unexposed) population. Again, assuming a population of 500,000 and a background rate of 3% for major congenital defects present at birth, the smallest detectable relative risk would be 1.3, which is 150 times greater than that presumed to be likely (1.002~. The total sample size that would be required to demonstrate a maximum relative risk of 1 .002, assuming a 3 % frequency of outcome, is approximately 212,000,000. DOSIMETRY OF ATOMIC VETERANS Radiation dose is generally considered in two parts, external and internal. External dose is that received from a radiation source outside the body such as radioactive materials on the ground or on equipment such as vehicles. For the exposed Atomic Veterans, the biggest components were gamma rays and beta rays, but only a small fraction of the beta radiation could penetrate the clothing and skin to reach the gonads, the specific organ of interest here. The candidate database for radiation doses for the Atomic Veterans is the Nuclear Test Personnel Review (NTPR) program established in 1978 by the Defense Nuclear Agency (DNA). The work of assembling the dose information was accomplished by the military branches and by contractors working for DNA. This work continues with the objective of obtaining the best estimates of dose for as many veterans as possible. Because only limited measurements were made for some test operations, particularly before 1955, the NTPR database in
FEASIBILITY OF THE STUDY 69 eludes doses that were estimated from few data and with extensive use of mod- els. The models are mathematical relationships based on extensive measure- ments made during later weapons tests and laboratory studies. Although a few veterans were exposed to neutrons, the bulk of the recorded doses were from gamma radiation. In some cases there was the potential for materials containing radioactive particles to be inhaled or ingested, and efforts were made to include doses from these particles in the dose estimates. In some cases, the major component of the dose received, gamma radiation, was measured with film badges. In other instances, dose was inferred from the badge worn by a companion performing the same activities and at the same lo- cation. In others, it was reconstructed from measurements made in the radiation field with instruments and from the time spent by the veteran in that radiation field. In many instances, the dose was estimated only for a relatively large group such as a platoon, a crew of a boat, or a work party. The less specificity avail- able for the estimation of dose, the greater the uncertainties in that estimate. Uncertainty in some of the dose estimates, especially for those people who were not badged, is unavoidable. Unfortunately, for a large number of veterans, the doses must be estimated from very little information, and the accuracies of the doses are correspondingly poor. For many people, a cause for concern has been the magnitude of the dose due to internal emitters such as plutonium. Once radioactive materials are inside the body, types of radiation that are essentially harmless when they are on the outside, such as alpha and beta particles, can irradiate cells, tissues, and organs. An accurate assessment of doses to the gonads is more difficult for some of these materials. Internal doses were found in early tests to be a small part of the total dose in general, but the potential for radioactive materials to be taken into the body by inhalation or ingestion may have existed in some cases, adding to the uncertain- ties about the total dose. All of these factors and others have been studied for many years. The Elects of Nuclear Weapons (Glasstone, 1962) presents dis- cussions of internal and external dose. For example, Chapters VIII and XI of Glasstone (1962) discuss this subject as it was understood in 1962, and NRC (1985) discusses this subject from the vantage point of the Atomic Veterans. An NRC report (1985) concluded that uncertainties about the internal dose are large for the Atomic Veterans but that the overall doses were small compared with the external levels of gamma radiation. The highest potential for internal exposure appears to have existed for the 28 men stationed on Rongerik atoll at the time of the Castle Bravo test. The 1985 NRC report gives estimates of the doses that these men received and concludes that the highest doses were to their thyroids. The doses to their gonads were not estimated, but the gonads would have been affected little by internally deposited radionuclides. In another evaluation (IOM, 1995), the Institute of Medicine's Committee to Study the Mortality of Military Personnel Present at Atmospheric Tests of Nu
70 AD VERSE REPRODUCTIVE OUTCOMES clear Weapons reviewed the NTPR database and the methods used by NTPR to estimate doses. The review found the NTPR dose data to be unsuitable for dose- response analysis. However, the committee believed that comprehensive dose reconstructions may be feasible for a limited subset of Atomic Veterans. Although the DNA dose data are unsuitable for dose-response analysis, they may provide a rough estimate of the magnitude of doses received by the Atomic veterans. Of 210,000 participating veterans, about 1,200 received doses that were estimated to exceed 50 mSv (5 rem) (DNA, 1995a), which is the present annual exposure limit set by the U.S. Nuclear Regulatory Commission (lOCFR20) for workers occupationally exposed to radiation. About 20,000 par- ticipants (DNA 1995b) have assigned doses that exceed the more conservative annual occupational limit, 20 mSv (2 rem), proposed by the International Com- mission on Radiological Protection (1991~. A total of 0.07% of the doses ex- ceeded 100 mSv (10 rem), and the average dose for the Atomic Veterans was 6 mSv (0.6 rem). Although the dose assigned to a given veteran might change with further study, the distribution of doses across the cohort is unlikely to change significantly. Two groups of veterans require additional comment: the veterans who en- tered Hiroshima and Nagasaki at the beginning of the occupation of Japan to assist in the cleanup and prisoners of war who may have been taken into the two cities on work parties. Because of the interest of the scientific community in the potential expo- sures that were received by early entrants in Hiroshima and Nagasaki after the atomic bombings, studies were done in the early 1960s to determine whether or not these exposures were significant from a biologic point of view. The term "early entrants" denotes those persons who were not close enough to the detona- tion to receive a dose directly from the bombs, but who walked into the ground zero area within hours to a few days after the detonations and thus got some ex- posure to either fallout or activation products. Because Japanese scientists made radiation measurements within a few days of the bombings and both American and Japanese scientists made measurements about a month later, there is a body of information on which to make comparisons with theoretical measurements and other measurements made at later weapons tests. Arakawa (1962) reported an extensive study of potential doses to individu- als who may have entered the ground zero and fallout areas. His data show that people entering either the ground zero area or areas of maximum fallout at the time that U.S. forces landed in Japan would have been in the millirem-per-hour range as a maximum. In fact, his studies show that residents of Nishiyama (an area about 3 kilometers to the east of the hypocenter where the bulk of the fallout in Nagasaki occurred) would have received the highest doses of any people not directly exposed to the bombs and that their lifetime doses, assuming that they never left the area, were below the level at which biologic effects would be de- tectable.
FEASIBILITY OF THE STUDY 71 More recent dose reconstructions on occupation forces in Hiroshima and Nagasaki estimated upper-bound doses based on worst-case scenarios (DNA, 1980~. The external doses ranged from 0.3 mSv (.03 rem) at ground zero in Hi- roshima to 6.3 mSv (.63 rem) in the Nishiyama area near Nagasaki. Whole-body internal exposures ranged from 0.03 mSv (.003 rem) in Hiroshima to 0.68 mSv (.068 rem) in the Nishiyama area. Therefore, it is not likely that people entering any area of either Hiroshima or Nagasaki in September 1945 would have re- ceived a dose of as much as a 10 mSv (1 rem) and that a casual visit to the area would have caused an exposure in the range of only a few millirem. The final group of veterans to be considered is the one composed of U.S. prisoners of war. There is no record of any prisoners being held in the Hi- roshima area, and no one claims to have been held there, so the main concern is for persons who may have been held captive near Nagasaki and who may have been taken into the city on work details. It is known that a few prisoners were held north of Nagasaki, and some of these say that they were used on work par- ties in the city after the bombing. In this case, Arakawa's early-entrant calcula- tions for Nagasaki would be most appropriate. The biggest contributor to expo- sure rate was sodium-24, which has a half-life of less than 15 hours, meaning that after 15 hours half of it has gone away by decay, after 30 hours only one- fourth remains, and so forth. If a work party entered the ground zero area of the city the day after the bombing, which under the conditions of communications, management structure, and so forth, seems unlikely, and if the work parties re- turned daily for a full work day until their release, their doses would have been very low, probably less than 10 mSv (1 rem). Such doses and the limited num- ber-of people involved would make this an unlikely basis for an epidemiologic study. There has been no statistically significant demonstration in the populations of Hiroshima and Nagasaki of any induced hereditary effects of radiation. With regard to veterans, within the constraints of the uncertainties, it is clear that the average dose as well as the highest measured dose to veterans were small com- pared with the minimum doses at which the effects of concern were possibly observed in other exposed populations such as the survivors of the atomic bombings of Hiroshima and Nagasaki. The organ of concern for this study is the gonads, but within the other uncertainties in doses, the dose to the gonads can be taken to be the same as the estimated dose to the whole person. IDENTIFICATION OF ADVERSE REPRODUCTIVE OUTCOMES Study of reproductive outcomes among Atomic Veterans requires being able to identify both normal and abnormal outcomes in an unbiased manner. Al- though a nonconcurrent cohort approach, which identifies groups of veterans who differ with regard to radiation exposure but are otherwise similar and fol
72 ADVERSE REPRODUCTIVE OUTCOMES lows them forward in time to determine if rates of reproductive outcomes differ by exposure group, would seem to be a logical approach, it is probably not fea- sible. These groups are likely to have completed their families at least 15 years ago, and the records necessary to identify adverse reproductive outcomes during a time period of from 15 to 50 years ago are not likely to be available. During this time there have been major changes in the information recorded in the vital records. For example, information on variables such as birth weight and gesta- tional age has not always been required to be recorded on the birth certificate in every state. In addition, some states do not maintain the medical information reported on the confidential portion of the birth certificate, for example, infor- mation on congenital malformations, with personal identifiers. Reporting of fetal deaths varies from state to state. For example, many states require that only those fetal deaths occurring at 20 weeks of gestational age or greater be reported, whereas other states require that all fetal deaths, re- gardless of age be reported. There is known to be marked underreporting of fetal deaths at early gestational age, the stages at which the rates of loss are the highest. In addition to underreporting of fetal deaths, underreporting of early neona- tal deaths in very low birth weight newborns has been observed. This has been recognized as a problem in some areas during the last decade and is likely to have been even more prevalent earlier. Even for infant deaths by other causes, there is great variability in the accuracy and completeness of recording of causes of deaths. To study the health conditions of greatest concern and health outcomes for which a biological mechanism related to paternal exposure can be postulated, unbiased information is even less likely to be available than for other types of adverse reproductive outcomes. This would include congenital malformations that are known or believed to include a genetic component in their etiology. Although it may be possible to obtain some information from Atomic Veterans or their spouses on their children who may have had birth defects, the medical and vital records necessary to validate this information in an unbiased way are not likely to be available. If a complete cohort of births to the wives of Atomic Veterans could be identified, then, in theory, it might be possible to determine birth defects diagnosed in the newborn period by reviewing the babies' medical records, but a large percentage of such records are likely to be unavailable be- cause of the length of time that has passed since the events occurred. Hospitals will have closed or changed ownership, records will have been purged or de- stroyed, and records will simply have been lost. Because of the size of the co- horts of interest and their geographic dispersal, it would be highly infeasible to determine the existence of medical records, let alone their availability or com- pleteness. The challenges of studying other outcomes, such as spontaneous abortions, learning disabilities, and mental retardation, would be even greater. These latter endpoints are difficult to study epidemiologically in defined con
FEASIBILITY OF THE STUDY 73 temporary populations and would be nearly impossible to study adequately in a historical cohort. Potential for recall bias is a particular concern for some endpoints, such as spontaneous abortions. Spontaneous abortions present a number of methodo- logical problems for study in contemporary populations (Sever, 1989~. It would be extremely difficult to study them in an unbiased way in populations that were at the height of their reproductive lives more than 30 years ago. In groups of women who have been questioned about their history of spontaneous abortion, recall seems to be relatively accurate for the period up to 20 years prior to the interview, however before that time recall is poor on the basis of a comparison of contemporary reports with later recall (Wilcox and Homey, 19841. This is in the absence of any concern about a potential association with an exposure that might lead to reporting or recall bias (White et al., 1989~. For many of the health outcomes of interest there is the potential for marked variability in the diagnostic criteria used in different years and in different areas. As noted earlier, mental retardation represents a wide variety of possible diagno- ses that share the common feature of some decrease in mental ability. Mental retardation is usually defined as an IQ of less than 70. Some definitions also include a functional component (Grossman, 1977~. The prevalence of mental retardation has been shown to be related to age. For example, the relation be- tween age and prevalence of mental retardation was shown in a cohort of 10- year-olds, in accordance with the belief that by age 10, such children would have been identified and could be ascertained through educational facilities (Yeargin- Allsopp et al., 19901. Study of mental retardation in the children of Atomic Veterans would re- quire access to school records that include information on standardized test scores. Such records are unlikely to be available. In addition, there are a num- ber of other risk factors such as alcohol use by the mother during pregnancy or the mother's educational level that would be very difficult to control for in a historical cohort study.