The Health of Former Prisoners of War: Results from the Medical Examination Survey of Former POWs of World War II and the Korean Conflict (1992)

Before this research study was begun, five basic research questions (A-E) were formulated as a priori areas of interest. These questions were developed at the time that the 1984–1985 questionnaire follow-up was being completed and were thus heavily influenced by its preliminary results. In brief, those results showed that depressive symptoms were elevated among former POWs, and this finding, along with similar results from Beebe's earlier follow-up (1975), highlighted the need to focus attention first and foremost on psychiatric morbidity. The 1984–1985 results, however, could not provide any real evidence regarding underlying diagnoses—for example, whether the elevated depressive symptoms were associated with a diagnosis of major depression or with, perhaps, posttraumatic stress disorder (PTSD). The hypothesis that PTSD was a probable candidate for either an underlying diagnosis or comorbid condition led eventually to the inclusion of additional psychological questionnaires in the current examination study. Thus, question A focused the study's primary attention on psychiatric morbidity, noting the different methods—physician interview and psychological evaluation—by which these data would be collected. The obvious concern with possible differences in findings based on data collected using the two methods led naturally to question B.

In contrast to psychiatric morbidity, at the time the study was designed there was, by and large, little current information on physical morbidity. Studies of Australian POWs under the Japanese, for example, discovered only one statistically significant difference—an increased history of ulcer—

between those POWs and a comparable group of WW II veteran controls (Goulston et al., 1985); psychiatric findings were much more prominent (Tennant et al., 1986; Dent et al., 1987). Beebe's earlier (1975) study, however, identified a number of conditions with significantly higher hospitalization rates among POWs; more generally, he also found that PWP illness rates were slightly higher than PWK rates and much higher than PWE rates. Thus, question C focused broadly on levels of illness, generally specified, and the main thrust of the question was directed toward determining whether Beebe's earlier findings (1975) on illness differentials still held.

Question D was in the broadest sense a holdover from the 1984–1985 follow-up. At the time of the follow-up, there were nagging questions about the quality of the self-reported data on illnesses, and it was decided that their analysis should be postponed until the more solid examination data were available for comparison. Thus, question D was formulated to emphasize the comparison of self-reported and examination data, and the data presented on question D generally support the original reluctance to analyze the self-reported data alone.

Question E, like others already discussed, was framed with findings from the 1984–1985 follow-up in mind, which had shown that malnourished WW II prisoners of the European theater had significantly elevated depressive symptoms compared with other European prisoners. The malnourished group, however, had not been included in Beebe's 1967 follow-up; consequently, less was known about their physical health in detail. Question E was thus formulated to study broadly the physical health of malnourished PWE.

Before turning to each of the above questions, some discussion of statistical testing is necessary. As was the case in Chapter 4, a useful examination of the descriptive data on POW and control lifetime prevalence rates requires some knowledge of the stability or, conversely, variability of these rates. Statistical tests are customarily used to compare such rates because a statistical test takes into account not only the magnitude of differences in rates but also the variability of the rates based on the size of the sample. Given the low response rates (see Chapter 3), however, the customary use of statistical tests would be inappropriate here. As in Chapter 4, statistical tests will nonetheless be used as indicators of whether a given difference in prevalence rates is ''noteworthy" or "appreciable." In doing this, it is recognized that a statistical test done in this setting has no valid inferential use but is instead merely a way of marking such noteworthy or appreciable differences. The additional comments made in Chapter 4 on the limitations of these tests all apply here as well; that is, the tests only mark a difference as noteworthy or not, without gradation; they may conclude that a difference is "not noteworthy" simply because sample sizes are small ("low pow-

er"); when a large number of such tests are done, the probability increases that some differences will be wrongly designated as noteworthy by chance.

The sections that follow examine each of the original five questions in turn.

Will rates of psychiatric illness, as ascertained by interview and psychological evaluation, be higher among former WW H PWP (prisoners of war, Pacific) than among their nonprisoner controls? Will this also hold true for WW II PWE (prisoners of war, European) and PWK (prisoners of war, Korean conflict) when compared with their respective controls? Will rates of psychiatric illness be higher among PWP and PWK than among PWE, as observed in earlier studies?

Based on previous studies of this cohort, it was expected that psychiatric sequelae would continue to persist. The preliminary findings in Chapter 4 clearly support that expectation: all POW groups had appreciably higher lifetime prevalence rates of depression and PTSD, and there is even the suggestion that schizophrenia might be found at a higher rate among POWs than among corresponding controls. In addition to these examination data, however, there are other data on psychiatric sequelae.

The 1984–1985 follow-up found that depressive symptoms were significantly elevated among former POWs, and a hypothesis was formed that these elevated symptoms might be due to or associated with a higher prevalence of PTSD in former POWs. Therefore, this section begins with an examination of the supplemental data on PTSD (see Chapter 2 for details), starting with the PTSD portion of the SCID (structured clinical interview for DSM-III-R); it then moves to one of the four self-administered psychological questionnaires, the Mississippi PTSD scale.

The SCID was designed to be administered by a VA psychiatrist or psychologist, and instructions for its administration directed that it was "to be embedded into the usual POW protocol psychiatric consultation, which should be conducted in the psychiatrist's own practiced manner and style." During data validation checks for this report, however, it was discovered that some of the SCID forms had apparently been given directly to subjects and completed by them. This discovery led to the checking of all completed SCID forms and the removal of those that were apparently self-administered (these amounted to roughly 7% of the total). Data from the remaining SCIDs, presumably administered during the psychiatric interview, are shown in Table 5.A.1.

The SCID measures both current and lifetime rates of PTSD; both are displayed in the table. Although the data show higher lifetime rates of PTSD for PWP, PWE, PWEM, and PWK compared with their respective

controls, only the PWP difference is noteworthy. Similarly, there are higher rates of current PTSD for all POW groups (comparing PWEM to WE) but appreciable differences only for PWP and PWK. Overall, the lifetime rates are approximately the same as those found during examination (see Chapter 4), but they are notably lower than those shown for Eberly and Engdahl's study (again, in Chapter 4). Also, in general, lifetime rates of PTSD are all quite higher than current rates, except for the PWK and WE groups. The apparently high rate of current PTSD among WE is somewhat unexpected and may be due to the quite small sample size on which the estimate is based. A more detailed comparison of psychological and interview results is discussed in the section on Question B, later in this chapter.

Another measurement of (current) PTSD, the Mississippi PTSD scale, was also administered as part of the examination, and Table 5.A.2 displays the prevalence rates of current PTSD based on data from that scale. In contrast to Table 5.A.1, rates are appreciably higher for all POW groups, relative to their corresponding controls. It is noteworthy that the rates in Table 5.A.2 are somewhat higher than those for current PTSD ascertained by the SCID; again, this discrepancy will be analyzed in more detail in the discussion of Question B.

The preliminary data in Chapter 4 highlighted depressive symptoms as another psychological condition found at an apparently higher rate among former POWs. Table 5.A.3 displays data on depressive symptoms as ascertained by two independent psychological instruments, the Beck and Center for Epidemiologic Studies depression (CES-D) scales. The Beck scale is designed to characterize depressive symptoms as mild, moderate, or high; however, the data in Table 5.A.3 tabulate the presence only of moderate and high levels of symptoms. These data show that PWP, PWEM, and PWK all

have appreciably higher rates of high or moderate depressive symptoms than their respective controls (both PWE and PWEM are compared with PWE); the PWE rate is elevated, but not appreciably so.

As noted earlier, the CES-D scale was administered by mail in the 1984–1985 survey whose results showed depressive symptom prevalence rates roughly three to five times higher among POWs than in other studies of comparable populations. The data in Table 5.A.3 are in line with those of the earlier survey, showing apparently similar rates of depressive symptoms, and all POW groups show appreciably higher levels of depressive symptoms than their respective controls. The WK control group apparently has a much higher rate of depressive symptoms than the WE control group and a slightly higher rate than the WP group; as noted earlier, this may be because this control group consists of combat veterans who were wounded and returned to action.

Table 5.A.4 displays data for the 90-item Hopkins Symptom Check List (SCL-90), scored in the customary manner to yield nine symptom scores plus an ''additional items" score, a global severity index (basically, the sum of the individual symptom scores), and a positive symptom distress index (the total score divided by the number of positive items). Comparing PWP, PWEM, and PWK with their control groups, one sees appreciably higher scores for all symptom indices (except the positive symptom index in PWP). Comparing PWE with WE, one sees elevations in scores for somatization, depression, anxiety, interpersonal sensitivity, hostility, and global symptom index.

Looking across POW groups, the PWK group generally has the highest level for every score, with most of their scores strikingly higher than all the others. Next are the PWEM scores, followed closely by PWP scores, with

PWE and WK scores fairly close together, trailing somewhat the PWEM and PWP scores. WP and WE scores follow, with WP levels a good deal above those of WE. In brief, looking across all the SCL-90 data, POWs show appreciably higher symptom scores than their respective controls, except for four PWE scores. In addition, PWK and PWEM are notable for their generally raised levels of psychological symptoms across all the subscales.

In summary, question A must be answered in the affirmative: psychiatric morbidity, whether assessed by examination, structured interview, or psychological questionnaire, is apparently higher in former POWs than in their respective controls. Particularly striking are the apparently high rates of PTSD and depressive symptoms. Psychiatric morbidity is appreciably higher in WW II prisoners of the Pacific theater and those of the Korean conflict compared with former WW II POWs of the European theater, as observed in earlier follow-ups of these same groups.

What differences, if any, will there be between psychiatric morbidity assessed by interviewer versus questionnaire? If there are differences, how will they influence the interpretation of results from question A?

As for question A, this section begins with an examination of the data on PTSD. Table 5.B.1 exhibits the comparative data on PTSD from the medical examination and from the SCID structured interview, limiting this comparison to those who had data for both. In the aggregate, the two give roughly equivalent rates of PTSD prevalence—25.9% for the medical exam and 30.7% for the SCID—but it is imperative to compare the two measures on a person-by-person basis. In Table 5.B.1, each subject is categorized into one of four columns based on his paired response to the two methods of measuring PTSD. Thus, each individual is either positive/positive by both measures, positive by exam and negative by SCID, positive by SCID and negative by exam, or negative by both. The resulting categorization results in what is called matched pair data, in which each subject provides his own matched responses for the pair of ratings. The matched pair data of Table 5.B.1 show that there are 21 cases in which the examination diagnosed PTSD and the SCID did not, versus 36 in the opposite direction; the matched pair chi-square (McNemar) test denotes this difference as noteworthy.

Looking more closely, however, one notes that almost all of this difference is concentrated among the controls, specifically, the WK controls; for POWs, the exam and SCID data are nearly identical. There are, of course, two competing interpretations for this difference. If the exam is taken as the standard measurement against which the SCID is compared, then the

SCID estimates are too high in WK; alternatively, if one considers the SCID the standard measure, this says that the physician examiners may have had a bias against diagnosing PTSD in WK. Based on the data in hand, either interpretation is defensible.

Moving from lifetime to current PTSD, Table 5.B.2 shows a similar comparison between the examination and the SCID. To estimate the prevalence of current PTSD from exam data, however, it is first necessary to exclude past, inactive cases of PTSD. This was done by making use of some additional data collected for each medical condition, namely, a judgment by the nosologist as to whether that condition had been resolved. For purposes of the following analysis, current PTSD was determined by eliminating PTSD diagnoses for which no resolution could be coded as well as those that had been resolved. Only those unresolved PTSD cases, arguably current cases, are included in the examination data tabulated in Table 5.B.2.

In contrast to the comparison of lifetime data, there are no noteworthy differences, for either POWs or controls, between the estimates of current PTSD based on the SCID and on examination. Also in contrast to the lifetime estimates, the overall estimates of current PTSD from the SCID data are slightly lower than the physician-based estimates: 22.1% for all

subjects using the SCID and 24.1% using the exam data. Thus, while there was a noteworthy difference between the SCID and medical exam estimates for lifetime PTSD—due solely to the WK group, in which SCID estimates were higher—there are no appreciable differences between SCID and exam estimates for current PTSD.

Table 5.B.3 shows a similar comparison of examination estimates of PTSD (again, unresolved cases only) and estimates of PTSD using the Mississippi scale. In contrast to the SCID data on current PTSD (Table 5.B.2), overall Mississippi scale estimates of current PTSD are appreciably higher than the examination estimates: 26.4% prevalence by examination versus 32.4% by the Mississippi scale. Comparison of the matched pair data, however, shows no noteworthy difference between the two estimates for POWs; it does show a noteworthy difference between exam and Mississippi estimates for controls.

The final comparison, the SCID data versus the Mississippi scale data, is shown in Table 5.B.4. Overall, the SCID estimate of current PTSD is appreciably lower than the Mississippi scale estimate: 20.6% for the SCID and 27.2% for the Mississippi scale. Comparison of the matched pair data shows an appreciable difference for POWs only, and nearly all of this dif-

ference is concentrated in the WW II groups. Although the POW and control SCID versus Mississippi comparisons give different results from the exam-versus-Mississippi comparisons, in both cases the overall Mississippi estimates are appreciably higher.

Given the relatively good agreement of estimates of lifetime and current PTSD derived from examination and SCID data (except for WK), but the less favorable agreement of estimates of current PTSD for the exam and SCID versus the Mississippi scale, can any firm conclusions be drawn? There are, unhappily, too many problems with nonresponse bias and measurement issues to permit sound conclusions. Nevertheless, it can be stated that the Mississippi estimates are somewhat higher than the exam and SCID estimates, and that appreciable differences, when they occur, seem concentrated in specific war eras: Mississippi estimates are higher than exam estimates in PWK and WK and higher than SCID estimates in WW II POWs. More research is clearly warranted, especially because the Mississippi scale was developed in groups of Vietnam-era veterans and its items may not properly capture PTSD symptoms in older veterans of earlier wars. There may also be a need simply to redefine scale cut-points and to recalibrate this instrument in older veteran populations.

Analyses parallel to those for PTSD were also conducted for depressive symptoms. Here, the psychological tests were the Beck and the CES-D scales, and the comparison examination data were unresolved diagnoses of depressive disorder, including major and minor depression. The data for the Beck scale, shown in Table 5.B.5, are scored in such a way as to characterize depression as none, mild, moderate, or high; for this analysis, only moderate and high levels (combined) are shown. Overall, the examination estimate of depressive disorder was 28.2% versus 22.6% for the Beck scale, and the matched pair data show a noteworthy difference in the two measures only among POWs (33% for the exam versus 26% for the Beck scale). There were no noteworthy matched pair differences for controls.

Table 5.B.6 displays similar data for the CES-D, using the customary cut-off point whereby scores 16 and above are considered positive for depression. Here, in contrast to the Beck scale data, there are large differences in overall prevalence rates (27.7% using exam data and 39.7% using CES-D data) and noteworthy matched pair differences between the two measures for POWs and controls. In all, the CES-D agrees much less closely with the medical exam than does the Beck scale.

In summary, the answer to question B is that examination and psychological questionnaire estimates of the prevalence of psychiatric disorders do differ somewhat and in complicated ways. Taking PTSD first, there was fairly good agreement for lifetime and current PTSD estimates between the exam and the SCID, except for the WK group. The overall Mississippi scale estimates of current PTSD were a bit higher than the SCID or exam rates, but such differences appeared to be dependent on war era.

Turning to depression, the CES-D data yield appreciably higher estimates of the prevalence of depressive disorder—more for POWs than for controls—compared with examination data. This may merely reflect the fact that the CES-D was designed to measure depressive symptoms rather than provide a diagnosis of depression. In contrast, the Beck scale data show appreciably lower estimates of the prevalence of depressive disorders for POWs but not controls, compared with the exam data. The exclusion of the ''mild" category cases from the analysis may have affected this comparison. Overall, the Beck scale results parallel the physicians' examination diagnoses of depression more closely than do the CES-D scale results.

Notwithstanding the above discussion, POWs show appreciably more

psychological problems than their respective controls by all measures. The differences among the various measurements, while of methodological interest and of potential importance when making comparisons with other studies, do not in any material way influence the conclusions drawn earlier in question A.

How have differences in illness levels changed over time? In particular, have the earlier differentials between the PWP and PWK groups, on the one hand, and the PWP and PWE groups, on the other, decreased with time?

In Beebe's earlier (1975) follow-up, illness levels were higher among PWP than among PWK, in line with repatriation examination data suggesting that PWP sustained greater impairment to their health during captivity than did PWK. On the other hand, PWP illness levels were notably higher than those for PWE, among whom only psychiatric sequelae were apparent. The passage of an additional two decades raises the obvious question: Have these earlier relationships changed?

To answer this question of changes over time requires comparable data from the two time periods. The 1967 follow-up collected data by questionnaire and record review, and such data are not directly comparable to the current medical examination information. However, the 1967 data on hospitalization are fundamentally comparable to similar data collected as part of the 1984–1985 questionnaire follow-up, and it is these data that will be used in the following analysis. Although these hospitalization data provide material for comparison between the two time periods, such data are relatively gross measures of morbidity; for example, many people suffering from mental illness are not hospitalized. Thus, there may be differences between groups that would not be readily apparent in hospital data. In addition, the 1967 data combine self-reported hospitalization data for nonfederal hospitalizations with data from VA and Army files, whereas the 1984–1985 data are strictly self-reported. Also, the earlier data were coded to the seventh revision of the International Classification of Diseases and Injuries, which necessitates the use of broad diagnostic categories when making comparisons with the 1984–1985 data, which were coded to the ninth revision of the International Classification of Diseases, Clinical Modification (ICD-9-CM).

Table 5.C.1 shows the earlier hospitalization data for PWP, PWE, and PWK. (The PWEM group, which was sampled independently of PWE, was not a part of the 1967 follow-up.) Hospital admission rates have been combined for the earlier follow-up periods (1946–1965 for PWP and PWE and 1954–1965 for PWK) to compute an average admission rate, adjusted for length of follow-up, of admissions per 1,000 per year. The data show some quite striking (fivefold or more) differences between PWP and PWE admission rates for infectious disease; allergic, endocrine, and nutritional disorders; and mental diseases; they also show sizable (around twofold) differences in almost every other organ system. Differences between PWP and PWK are on the whole much smaller, with sizable PWP elevations only for allergic and endocrine diseases and mental diseases, and a sizable elevation for PWK in hospitalizations for accidents, injury, and poisonings.

Table 5.C.2 shows later data on hospitalization, during the period 1984–1985, for PWP, PWE, and PWK, plus an additional group, PWEM, again calculated as admissions per 1,000 per year. These data tell a different story. There are no longer any striking differences among the three groups of Table 5.B.1, not even when PWEM is included, and sizable differences persist only for mental disease. PWK hospitalization rates are lower for circulatory disease, a finding that is probably age related, but they are not otherwise uniformly lower than rates for WW II POWs.

Although no definitive interpretation of the data in Tables 5.C.1 and 5.C.2 is possible, three pertinent points should be kept in mind. First, some of the differences seen between PWP and PWK illness levels two decades

ago could be attributed to age, with the typical PWP being 45–55 years old and the typical PWK 35–45 years old; as both groups age, this chronological difference should become less important. Second, differences in hospital admission rates over time may be due, at least in part, to deaths of the sicker men over time. Third, as one moves further from the time of captivity, one also moves closer to old age, when regardless of earlier health, background levels of illness start to rise dramatically.

Another, less suitable way to look at changes over time is to compare lifetime prevalence rates from the examination with current prevalence rates from the same source, again considering current conditions to be those that are still unresolved. Earlier, Table 4.2 displayed lifetime prevalence rates for all medical conditions in broad ICD rubrics. Table 5.C.3 exhibits cur-

rent prevalence rates in a similar fashion. Categories for which there are noteworthy differences in rates between POWs and controls are so designated.

Table 5.C.3 shows noteworthy differences in the current prevalence of diseases of the blood (the PWP and PWK groups have lower levels), other endocrine diseases (the PWP group is lower), psychoses (all but PWE), neuroses (all groups), nervous system (PWP and PWK), digestive diseases (PWP and PWK), symptoms (PWK), and V-codes, which are not strictly medical diagnoses but are factors that influence health status and contact with health services (PWP). Because they are ill-defined, the latter two nondisease categories will not be studied further; however, the remaining findings will be reconsidered in Chapter 6. In all, there were 13 noteworthy differences, 5 fewer than seen in Table 4.2; these are summarized in Table 5.C.4. These results support the findings in Tables 5.C.1 and 5.C.2 of lessening health differences over time.

In summary, based on hospitalization data, it appears that differences in illness levels among POW groups have become smaller over time. In the 1967 follow-up, PWP hospitalization rates were somewhat greater than those for PWK and notably larger than those for PWE; by 1984–1985, nearly all readily apparent differences had disappeared. Comparing lifetime and current prevalence data from the exam (a much more indirect measure of change over time) generally supported the hospitalization findings and showed that

there were fewer noteworthy current POW-versus-control differences than lifetime differences.

How do the physician-reported physical examination findings compare with the self-reported diagnoses, symptoms, and complaints from the 1984 questionnaire? Which physical findings are under-or overreported, and how do nonmedical factors influence this reporting?

Although the 1984–1985 questionnaire follow-up was focused on psychiatric morbidity and contained as its centerpiece the Center for Epidemiologic Studies depression scale, it also contained two general items on current health, asking about medical conditions that were and were not being treated. The first item was worded as follows: ''Are you currently now under medical care?" It was followed with a "By whom?" Space was then left to list up to five such medical conditions; the subject's responses were coded using ICD-9-CM. The second item asked, "Do you need medical care that you are not receiving?" Again, it was followed by blank spaces for up to five conditions. These data, however, were relatively sparse and are not tabulated here.

Because the levels of reported medical conditions under treatment are roughly comparable across all four POW groups, the data in Table 5.D.1

show all POW groups combined. To facilitate later comparisons, each of the four POW groups has been given equal weighting. Self-reported rates are, for the most part, under 10% prevalence, with the exceptions of heart disease, hypertension and other circulatory disease, digestive disease, arthritis, other musculoskeletal disease, symptoms and ill-defined conditions, and V-codes.

Before attempting to compare the self-reported data in Table 5.D.1 with examination data, some further discussion is in order. The self-reported data were responses to a question about current medical care, but the examination contains historical as well as current problems. Question B used one potential method of separating current from historical medical exam conditions: selecting only unresolved conditions. In this section, another method in addition to that is used. This second method considers the source of the examination information, coded for every medical condition, and selects information only from certain sources. Because summary diagnosis data, perhaps the single best source of data, were unavailable for many subjects, data from all findings and laboratory sections of the medical examination were combined to produce the exam data. (Both the findings and laboratory sections of the exam were thought to be freer of historical information.) Prevalence rates based on these two combined sources are displayed in Table 5.D.2.

Comparing these two sets of rates, one sees that in general the prevalence rates for unresolved conditions are equal to or greater than the rates for medical conditions in the findings and laboratory sections of the exam. This apparently higher rate is most striking for psychoses, neuroses, and nervous system conditions; only the unresolved acute respiratory conditions have a notably lower rate.

Despite their differences, both the unresolved medical conditions and the medical conditions from the findings and laboratory sections show rates that are strikingly higher than the rates of self-reported medical conditions in Table 5.D.1. To facilitate a rough comparison between self-reported and exam data, prevalence rate ratios were calculated by dividing the rates for unresolved and for findings/laboratory medical conditions by the self-reported ones. For ease of calculation and comparison, each of the four POW groups was weighted equally. Both of these ratios are displayed in Table 5.D.3.

Table 5.D.3 shows that, with only two exceptions, every ratio is well above 1, meaning that the examination-based rates are notably higher than the self-reported rates. The ratios, by and large, range between 2 and 10 and are mostly greater for unresolved conditions than for findings/laboratory conditions. Part of the explanation for the lower rates of self-reporting lies in the reporting process itself. First, questionnaire space limited the maximum number of self-reported conditions to five per person, whereas

the number of medical examination responses, for both unresolved and findings/laboratory conditions, was essentially unbounded. Although the average number of questionnaire responses was around 3, ranging from 2.9 for PWP to 3.5 for PWE, the critical question—for which data are unavailable—is, how many of those who reported five conditions might have reported six or more? Second, self-reported medical conditions were necessarily reported in less precise layman's language, leading in some cases to artificial differences in rate ratios. For example, V-codes are factors that influence health status; they may include such items as a personal history of

selected conditions (e.g., cancer). When they are self-reported, such medical conditions would be assigned V-codes, but they probably would not be recorded by a physician as an unresolved condition nor included in either the findings or laboratory sections of the examination. It is not surprising, therefore, that V-codes were reported more frequently on a questionnaire than in the findings/laboratory section of the medical examination. Nor is it surprising that self-reporting was most complete (i.e., comparability ratios

were lowest) for chronic conditions that are easily described in less technical language, such as cancer, diabetes, heart disease, and hypertension.

Although comparability ratios tended to be higher for unresolved medical conditions than for medical conditions derived from the findings and laboratory sections of the examination, some of the artifactual problems above make ratio comparisons difficult. To facilitate such comparisons Table 5.D.4 displays the ranking as well as the value of the ratios. The

ranked conditions show fairly good agreement between the two columns with the exception of psychoses, for which the ratio of self-reported to findings/laboratory conditions is substantially higher than the ratio of self-reported to unresolved medical conditions; presumably, psychoses are less often resolved. In addition, several of the high-ratio conditions, like psychoses and benign neoplasms, are based on very low self-reported prevalence, which makes their ratios less stable and more prone to error.

In summary, the simple answer to question D is that physician-reported findings appear to be much more frequent than self-reported findings on medical conditions across almost all broad diagnostic categories. Some of that difference can be attributed to the reporting process, in particular, the limitation of five self-reported medical conditions; some differences are accentuated or possibly created by the use of medical versus nonmedical terminology. In more general terms, underreporting may be due to the subjects' lack of understanding of medical conditions or even denial of them. Nonetheless, the differences measured here were sizable and deserve additional future study, along with the question of how nonmedical factors influence differences between physician and self-reported rates.

Can any distinctive signs of abnormal physical findings be seen in the subset of PWE veterans who were seriously malnourished at repatriation?

The group of WW II European theater POWs designated as PWEM are an independent sample of POWs added by Nefzger (1970) for the second follow-up. The group was derived from a 20% sample of Army hospital admissions with diagnoses of malnutrition; those chosen had no diagnosis other than malnutrition and had remained in the hospital at least 10 days. The comparison of these PWEM to the larger group of all WW II European theater prisoners, PWE, affords an opportunity to study the long-term effects of a relatively short (compared with PWP) period of captivity that nonetheless produced severe malnutrition in its survivors.

The tables pertaining to question A showed that rates of psychological sequelae are clearly higher in PWEM, compared with PWE; the focus of question E, however, is on physical findings. In Chapter 4, lifetime prevalence data for medical conditions were displayed in Table 4.2. These data showed few striking differences between PWEM and PWE, although in general, lifetime prevalence rates across all ICD rubrics were higher for PWEM than for PWE; the only noteworthy difference in PWEM and PWE rates is for ''other endocrine disease" (which includes nutritional deficiencies).

Because there were no precise medical hypotheses specified before-hand, the study of broad disease categories in Chapter 4 was defensible as a protection against overlooking unanticipated findings. This high level of

diagnostic aggregation also produces larger samples and yields more stable prevalence estimates. Yet such high-level aggregation is not without its problems. In particular, the higher the level of diagnostic aggregation, the greater the likelihood that distinct medical conditions will be put together, thereby increasing the risk of not observing an elevated rate for a narrowly defined condition because it has been combined with other lower rate conditions into a broader category.

One solution to this dilemma, which is discussed again in the next chapter, is to settle on some small, fixed set of narrower diagnostic categories for study. Although there were insufficient data to formulate precise a priori medical hypotheses for question E, a relatively small set of diagnostic categories can be chosen for further study on the basis of recently published POW examination studies. In Chapter 4, the diagnostic categories reported by Eberly and Engdahl (1991) were studied. To these are added other diagnostic categories from the work of Steven Oboler and colleagues of the Denver VA Medical Center, whose findings are based on the examination of some 200 former POWs, mostly veterans of WW II (Oboler, 1987, and subsequent personal communication). Table 5.E.1 shows lifetime prevalence data for the combined Eberly/Engdahl and Oboler categories.

There are five diagnostic categories in which PWEM lifetime prevalence rates appreciably exceed those of PWE. Aside from dysentery, beriberi, and frozen feet—these three are presumably elevated as a result of prison camp medical history—the rates of peripheral nerve disease and gastroenteritis are appreciably elevated. To determine whether some specific diagnoses might account for the above aggregate differences, detailed diagnostic data were examined. Two narrowly defined diagnostic conditions make up the gastroenteritis category: other and unspecified noninfectious gastroenteritis and colitis (ICD-9-CM code 558.9) and irritable colon (ICD9-CM code 564.1, which includes irritable bowel syndrome). The rates of the latter condition were nearly identical in PWEM and PWE, so that the apparent difference in lifetime prevalence of gastroenteritis is almost all accounted for by a difference in the prevalence of noninfectious gastroenteritis and colitis.

The data for peripheral nerve disease are more complicated. Table 5.E.2 shows the detailed three-digit subcategories and their ICD codes. Although the numbers of cases, and therefore the prevalence rates, are quite low when reported by detailed diagnostic category, it remains clear that most of the excess PWEM peripheral neuropathy is confined to three subcategories. Lifetime prevalence rates of mononeuritis of the upper and lower limbs and hereditary and idiopathic neuropathy together appear to account for most of the PWEM excess. The very sparse numbers of diagnoses here, however, yield rates that must be even more cautiously interpreted than earlier prevalence rates.

In summary, the answer to question E must be that there is some evidence of differences in physical findings between the malnourished WW II European theater prisoners and all other European theater POWs. Looking first at the broad ICD disease groups, one sees an appreciably higher lifetime prevalence rate of other endocrine diseases (which include the nutritional disorders). Moving to the selected set of medical conditions noted in similar studies of VA examinations of POWs, one sees appreciably higher lifetime prevalence rates of dysentery, beriberi, and frozen feet, all presumably the earlier sequelae of captivity. There are, in addition, appreciably higher rates of gastroenteritis—noninfectious gastroenteritis and colitis, but not irritable bowel syndrome—and of peripheral nerve disease, concentrated in the three categories of mononeuritis of the upper and lower limbs and hereditary and idiopathic peripheral neuropathy.

This chapter portrays most clearly the persistence and predominance of psychological aftereffects, now four-and-a-half decades after repatriation in

some cases. Surely, posttraumatic stress disorder and depressive symptoms, both of particular interest and the focus of special data collection, are the most striking sequelae, and the findings from the symptom checklist suggest that many other psychological conditions are also much more prevalent among POWs than among their comparable controls. Physical findings are not nearly as striking, and earlier differences among POW groups may have lessened over the past 25 years. The special group of malnourished WW II European theater prisoners, however, exhibits not only elevated prevalence rates of conditions clearly linked with their earlier captivity, such as dysentery and beriberi, but also some apparently longer-term aftereffects, such as gastroenteritis and peripheral nerve disease.

The relative deficiency of apparent physical findings across the whole of the examined POW population, however, may be somewhat attributed to the relative paucity of specific a priori medical hypotheses. The analyses of Chapter 6 attempt to remedy this deficiency and focus more sharply on the source of current POW health problems by examining their associations with experiences in military captivity.

Beebe GW. 1975. Follow-up studies of World War II and Korean war prisoners: II. Morbidity, disability, and maladjustments. Am. J. Epidemiol. 101:400–422.

Dent OF, Tennant CC, Goulston KJ. 1987. Precursors of depression in World War II veterans 40 years after the war. J. Nerv. Ment. Dis. 175:486–490.

Eberly RE, Engdahl BE. 1991. Prevalence of somatic and psychiatric disorders among former prisoners of war. Hosp. Commun. Psychiatry 42:807–813.

Goulston KJ, Dent OF, Chapuis PH, et al. 1985. Gastrointestinal morbidity among World War II prisoners of war: 40 years on. Med. J. Aust. 143:6–10.

Nefzger MD. 1970. Follow-up studies of World War II and Korean war prisoners: I. Study plan and mortality findings. Am. J. Epidemiol. 91:123–138.

Oboler S. 1987. American prisoners of war—an overview. In Williams T (ed.), Post-traumatic Stress Disorders: A Handbook for Clinicians. 1987. Disabled American Veterans, Cincinnati, pp. 131–143.

Tennant C, Goulston K, Dent O. 1986. The psychological effects of being a prisoner of war: forty years after release. Am. J. Psychiatry 143:618–621.