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)

The analyses undertaken in Chapter 5 were all directed at those issues that had been identified as important before the examination study actually began. By and large, they were simple, straightforward comparisons of POW and control disease prevalence rates. Unfortunately, while these are the most usual and orthodox types of analyses, they have their problems, both medical and statistical. In the hope that some of the inherent shortcomings of those analyses might be overcome, this chapter presents other analyses of the examination and psychological data, all of which lie outside the scope of the narrowly defined, formal assessments of Chapter 5. As noted earlier, the disappointingly low response rates for the exam render the customary statistical testing and inferences inappropriate. Thus, as in earlier chapters, statistical tests, when used, will merely provide an indication of ''noteworthy" or "appreciable" differences or associations and not of statistically significant differences.

Aside from the difficulties caused by low response rates, the chief problem in Chapter 5 was the lack of definite, specific hypotheses about medical conditions, a lack easily seen when such conditions were contrasted with mental disorders. As explained earlier, similar research on former Australian prisoners of the Japanese, conducted on an admittedly smaller scale, found no excess prevalence of medical conditions among former POWs except for ulcer—specifically, a history of and not current ulcers—and increased use of analgesics, suggesting an increased prevalence of arthritis that was, however, not directly observable. The reasonable choice then was

to look at broad disease categories, so as not to miss any important findings that might manifest themselves in the examination data. The danger of doing the same kind of analyses on a detailed level arises from the fact that doing literally thousands of statistical tests has the real risk of marking as a medical problem that which is merely statistical noise. This chapter attempts to gain a middle ground by revisiting and expanding the analyses in Chapter 4. Only a limited number of medical conditions will be considered in more detail, and the medical conditions to be studied will be chosen based on published epidemiological research and clinical material. To avoid a bias from historical reporting, only unresolved medical conditions will be discussed.

The core set of conditions included in Table 6.1 come from Eberly and Engdahl's study (see Chapter 4). All but one of the remaining conditions were chosen for analysis based on the published work of Oboler and subsequent personal communications (discussed under question E in Chapter 5). Thus, the conditions to be studied further were all suggested by epidemiologic or clinical considerations, and their selection was not based on examination findings, with one exception—fewer blood dyscrasias (discussed below).

In Chapter 4, an appreciably lower lifetime prevalence of diseases of the blood was noted, and in question C in Chapter 5 an examination of current prevalence data produced a similar finding. There were additional International Classification of Diseases rubrics with higher POW current prevalence rates as well, but in almost all of those cases, a specific disease had already been singled out for analysis—for example, ulcer in the digestive disease group and peripheral nerve disease in the nervous system group. The apparent deficit of other endocrine disease among PWP, which was primarily due to an excess number of WP cases of other metabolic and immunity disorders, is not analyzed further, given the earlier data on the possible nonrepresentativeness of the WP group and the failure of other control groups to show this excess.

The appreciably lower prevalence of blood diseases, however, is both persistent, appearing in both the lifetime and current prevalences derived from exam data, and pervasive, occurring in both the PWP and PWK groups as well as the PWE and PWEM, although in the latter the differences were not noteworthy. Moreover, the reason for this lower prevalence appears to be fairly specific, in that most of the deficit can be attributed to anemia, by and large "other and unspecified." All of these factors together argue for the inclusion of anemia in subsequent analyses; thus, although this is a condition suggested from examination of the data and less frequently found in POWs, it is nevertheless included in Table 6.1.

Table 6.1 presents data on the selected conditions for all study groups. As seen in Chapter 5, psychiatric illnesses predominate, with appreciably

higher rates of depressive symptoms and generalized anxiety in POWs in all study groups; posttraumatic stress disorder (PTSD) rates are appreciably higher for all but PWE. Among the medical conditions, there are appreciably higher prevalence rates of ischemic heart disease for PWEM, peripheral nerve disease for PWP and PWK, and gastroenteritis for PWEM and PWK;

WP had an appreciably higher rate of malaria than PWP, but, again, doubts about the soundness of the findings from the WP group motivated the decision to conduct no further analyses. In addition, there were somewhat higher rates for hypertension in PWP, ischemic heart disease in PWE, intermittent claudication in PWK, gastroenteritis in PWP, and osteoarthritis in PWK.

Besides the problem of diagnostic categorization, there are other, statistical problems with the analyses of Chapter 5. First and foremost is the problem of control groups. The small samples in the control groups directly reflect their lower response rates, and although the demographic and VA hospitalization data in Chapter 3 were somewhat reassuring, it should be noted that the members of the WP group with completed exams, in particular, appear unusual. Among the three control groups, after eliminating the four medical conditions with zero prevalence rates, WP have the highest prevalence rates for 11 of the 15 remaining conditions. The WP group also has the only control prevalence rate—that for malaria—that is higher than a corresponding POW rate.

Second, there is the question of whether there are more statistically sensitive analyses that could be undertaken than those that simply compare POW-versus-control prevalence rates. Comparing POWs with themselves, using those less harshly treated as a basis for comparison with those more harshly treated, should not only provide a more sensitive analysis but should also avoid the problems that arise from analyzing data from a small and potentially unrepresentative control group, such as the WP.

This chapter continues with a set of analyses that use POWs as internal controls to provide independent information about possible associations between selected medical conditions and prior military captivity. The underlying idea is straightforward. Although the widest differences among subjects are thought to be between POWs and controls, there is sufficient variation among POWs to compare those who reported less harsh treatment in captivity with those reporting harsher treatment; this essentially uses the first group as an internal control for the second.

The overwhelming advantage of internal controls, although it may not be immediately obvious, is that one may include in the analyses any supplemental data collected only from POWs. This is particularly important in instances in which captivity data are present in such detail that a gradient for harshness of treatment can be established; if the prevalence of a particular medical condition then increases in a regular fashion with the increase in the harshness of treatment gradient, causality arguments are strengthened. The disadvantages of using internal controls include the possibility of reporting bias in the supplemental data collected only among POWs and lack of a true baseline, which would tend to underestimate the adverse effects of military captivity. In addition, the careful demographic balances between

POWs and controls that were established when the samples were taken are lost (although it is unclear how great a disadvantage this might be). Finally, because the PWEM group was not included in the 1967 follow-up and thus lacks the more detailed data on harshness of captivity, their examination data are not included in this chapter; there are, therefore, three study groups that are considered: PWP, PWE, and PWK.

Characterizing the harshness of treatment is the sine qua non of analyses that use internal controls. Clearly, a complete, simple characterization is impossible, and so one approximates. Even if there were available data characterizing the harshness of treatment in individual prison camps, their use would not be straightforward; for example, tracking an individual's movements in and out of various camps might prove to be fairly complicated. Furthermore, even if this exercise were successful, it is not clear whether aggregate prison camp data would adequately characterize the treatment of each and every individual POW. Instead, two kinds of military captivity data, each collected on an individual basis, are examined in this chapter: medical symptoms reported while in captivity and self-reported weight loss during captivity (both kinds of data were collected as part of Beebe's [1975] follow-up study). Weight loss during military captivity, in particular, is one of the strongest overall markers of severity of trauma and has been strongly associated with severity of aftereffects in published reports.

Medical symptom data regarding infectious and nutritional diseases during captivity were collected in 1965–1967 by a questionnaire sent to all POWs (not including PWEM). The POWs were given a checklist of physical conditions and asked to indicate, condition by condition, whether they suffered from the effects of a particular condition during captivity. These prison camp symptoms are listed in Table 6.2. In collecting symptom data by questionnaire, one encounters the possibility of recall bias; that is, persons with health problems at the time that the questionnaire is being filled out may be more likely to recall earlier harsh treatment than persons who were not ill at the time of data collection. The fact that these symptom data were collected some 25 years ago, however, removes the possibility of current recall bias—any recall bias must be associated with symptoms and medical conditions reported at the time of the earlier questionnaire.

The symptoms shown in Table 6.2 include many that are associated with nutritional deprivation. Later in this chapter, these prison camp symptoms will be aggregated into three separate measures, two of which—edema and visual symptoms (such as night blindness)—have obvious associations with specific nutritional deprivations (thiamin and vitamin A, respectively). Unfortunately, roughly one-quarter of the queried POWs did not provide data on weight loss or prison camp symptoms, and for the remainder of the chapter, those with missing data for these variables are excluded from the analyses.

Table 6.3 shows prevalence rates of selected conditions for all PWP, PWE, and PWK combined and subdivided into two categories: those whose prison camp weight loss was 35% or less and those with more than 35% weight loss (this categorization was chosen to correspond with Engdahl's analyses). Because the prevalence rates in this chapter are all based on unresolved conditions, the rates for dysentery, malaria, beriberi, and frozen feet, all presumably short-term sequelae of captivity, were very low; therefore, these conditions, as well as bipolar disorder, were removed from further consideration.

The data in Table 6.3 show a number of noteworthy differences in prevalence rates between low-and high-weight-loss groups, including intermittent claudication, arterial vascular disease, peripheral nerve disease, and osteoarthritis. In the light of the findings discussed in earlier chapters, this appreciable association of weight loss and intermittent claudication and peripheral nerve disease is probably not surprising, nor are the appreciably higher rates of depressive disorder and PTSD among the high-weight-loss group; Eberly and Engdahl's study showed significant associations with depression, schizophrenia, PTSD, and generalized anxiety disorder. However, the inverse association of osteoarthritis and weight loss—less osteoar

thritis in the higher weight loss group—is the first such finding of its type in this report.

Table 6.4 contains data on the same medical conditions that are noted in Table 6.3, but this table shows associations with the number of reported prison camp symptoms. In this analysis, self-reported conditions were simply added together, and all POW respondents were arranged into three categories: those reporting 0 to 3 conditions, those reporting 4 to 9 conditions, and those reporting 10 or more conditions. There are a number of conditions in which prevalence rates rise from the lowest symptom category to the middle category, and then to the high category. Rather than merely look

for differences among the three categories, the prevalence rates in Table 6.4 were subjected to a chi-square test for linear trend, which indicates increasing or decreasing linear relationships between symptom category and prevalence rate. Although the results of these tests, like the others in this report, cannot be used as the basis for valid statistical inferences, they are more powerful than the usual chi-square test and may thus generate new hypotheses regarding associations of disease and prison camp treatment.

Among the medical conditions, there is a noteworthy linear increase in prevalence as symptoms increase; see, for example, intermittent claudication, arterial vascular disease, asthma, and peripheral nerve disease. Simi-

lar noteworthy associations hold for depressive disorder, schizophrenia, PTSD, and generalized anxiety. There are also fairly strong relationships between symptom category and ischemic heart disease, myocardial infarction, and osteoarthritis, although the prevalence of unresolved osteoarthritis falls with increasing past reporting of prison camp symptoms. This last finding is consistent with the earlier finding of an inverse association with weight loss.

Although the data on weight loss and prison camp symptoms are more easily viewed when categorized as in Tables 6.3 and 6.4, their actual values can be used to calculate product-moment correlations, treating each medical condition as a ''zero-one" value (i.e., either absent [0] or present [1] for each examined subject). The results of this analysis are shown in Table 6.5, with separate columns for weight loss and symptoms correlations for each of the three study groups.

Many of the appreciable correlations in Table 6.5 reflect earlier noteworthy differences in the combined POW data in Tables 6.3 and 6.4. The noteworthy association of prison camp symptoms with peripheral nerve disease, for example, is seen in both PWP and PWE, and all three groups show appreciable correlations of prison camp symptoms with depressive disorder. Some conditions, however, have noteworthy associations in only one POW group: intermittent claudication (PWK), arterial vascular disease (PWK), asthma (PWP), PTSD (PWK), and generalized anxiety (PWP). One condition, schizophrenia, showed no appreciable association in any of the three individual groups. Part of the explanation for the sparser findings among the individual groups is no doubt a result of their smaller sample sizes. Similar comments apply to the correlations with weight loss, except that only arterial vascular disease among the medical conditions shows a noteworthy correlation in more than one group.

This last observation points up one of the shortcomings in the data in Tables 6.3 and 6.4. Both tables were generated by pooling the data from all three POW groups (PWP, PWE, and PWK); therefore, some of the apparent associations between weight loss and symptom categories might actually be due to underlying differences among POW groups in harshness of treatment (and thus amount of weight loss and number of prison camp symptoms). Moreover, if weight loss and symptoms are markers for the POW group, the potential exists for some confounding between them and age, since PWK are notably younger than PWP and PWE.

Looking at each POW group separately, as in Table 6.5, removes the possibility of confounding group and weight loss or prison camp symptom effects, but it has its own drawbacks. Chief among these is the reduced sample size, which may mean that possibly noteworthy associations are lost in statistical "noise." One way to examine the effects of weight loss and prison camp symptoms while controlling for the separate effect of POW

group is to perform a Cochran-Mantel-Haenszel (CMH) analysis, which estimates the common association between a given medical condition and a weight or symptom score, using separate data for each POW group. This kind of analysis requires an assumption that the data in the separate POW groups are homogeneous, which can be tested, and provides a single summary estimate of the strength of association.

A CMH analysis was conducted for all of the selected medical conditions used in previously reported analyses in this chapter; only the noteworthy results are displayed in Table 6.6. Shown in the table are results based on an analysis of the risk ratio (which is akin to a ratio of prevalence rates) for the weight loss data (which have two categories—see Table 6.3) and a measure of general association for the prison camp symptom data (which have three categories—see Table 6.4). All noteworthy associations are so designated; a few less marked associations are also shown.

The data in Table 6.6 present a picture similar to earlier findings, even with the statistical adjustment for the effect of POW group differences. There are noteworthy associations between weight loss and arterial vascular disease, depressive disorder, and PTSD, and important but less strong associations between weight loss and intermittent claudication, peripheral nerve disease, and osteoarthritis. (Again, the latter is an association between higher weight loss and a lower prevalence of unresolved osteoarthritis.) The data for symptom scores show noteworthy associations with ischemic heart disease, ulcer, asthma, depressive disorder, and generalized anxiety, as well as somewhat weaker associations with PTSD and cerebrovascular disease.

That there are a number of common medical conditions among different analyses is a comforting finding; it suggests that the analyses undertaken thus far in this chapter have unearthed a number of strong candidates for current medical conditions that might be associated with earlier military captivity. As a final refinement, statistical analyses were undertaken to determine the joint, simultaneous effects of weight loss and prison camp symptoms, taking into account the differences among POW groups.

For this analysis, the prison camp symptom data were further refined along the lines of Beebe's previous work (1975). Specifically, reported symptoms on edema (two items: feet or ankle and leg swelling) were combined into a single marker of edema; reported visual symptoms (five items) were combined into a single variable that counts the number of these responses (from zero to five); and the remaining 13 symptoms were added together for a third symptom marker. These three symptom markers plus percent weight loss and POW group were used as independent variables in logistic regression analyses of the medical conditions reported in Table 6.6, plus schizophrenia (from Tables 6.4 and 6.5) and gastroenteritis (from Table 6.1). The use of logistic regression not only enables independent estimates of the effects of weight loss and the three symptom scores to be made, but also permits the analysis of all the available data at one time while controlling for differences among the three POW groups. The latter is accomplished by creating two variables, one comparing PWP with PWE and one comparing PWK with PWE, and including them in the regression. Backward stepwise regressions were fit to eliminate those variables without appreciable associations, although in some cases in which no independent variables were appreciably associated with a condition, additional regressions were fit using variables with less strong associations.

Table 6.7 shows the results of these logistic regressions for all of the medical conditions noted above. Estimates of odds ratios are shown together with 95% confidence limits; confidence limits that do not include the value 1.0 would be considered to have a noteworthy association with a particular medical condition. Odds ratios greater than 1.0 indicate that a higher prevalence is associated with the factor, while odds ratios of less than 1.0 indicate a lower prevalence. For example, ischemic heart disease is 1.737 times more likely to be found among POWs who reported symptoms of edema than among POWs who did not, and it is 0.405 times less likely to be found among PWK, compared with PWE.

In the cases of visual and other symptoms, the estimates of risk are for a single reported symptom, and the effects of more than one symptom are multiplicative. For example, each additional reported other symptom raises the estimate of the odds of having intermittent claudication by 1.11; 2 reported symptoms raise the odds to 1.23, 3 to 1.37, and so forth, up to a maximum of 13, for which the risk is 3.93 (1.1109¹³) times as high as for those with no reported symptoms. The situation is similar for percent weight loss, and here the odds ratio associated with a 10% weight loss is reported in Table 6.7. Odds ratios for other values of weight loss can be easily calculated from these figures.

Among the nonpsychiatric conditions, all but cerebrovascular disease and osteoarthritis have a noteworthy association with one or more markers of harshness of treatment during captivity. For cerebrovascular disease,

there is the suggestion of an association with visual symptoms, and for osteoarthritis, a similar association with percent weight loss: both of these give evidence of the presence of potentially important but weaker associations. No one symptom marker predominates in its associations with unresolved medical conditions; rather, each shows only one or two noteworthy associations.

All psychiatric conditions have some noteworthy association with a prison camp symptom marker except for schizophrenia, which is appreciably higher in PWK and has a weaker strong association with other prison camp symptoms. Except for PTSD, which shows appreciable differences between PWP and PWE as well as PWE and PWK, the only noteworthy group differences involve the PWE versus PWK comparison.

To summarize, this chapter's analyses were directed at more specific medical conditions and, using POWs as internal controls, have suggested a number of new findings. Appreciably higher prevalences of current ischemic heart disease, intermittent claudication, arterial vascular disease, peripheral nerve disease, ulcer, and asthma appear to be associated with nutritional deprivation and other measures of treatment during imprisonment; there are less strong associations of these factors with current prevalence of gastroenteritis as well as an apparently lower prevalence of osteoarthritis in those with higher weight loss.

The results of these analyses, however, are far from uniform in their findings of association, and the meaning of the different prison camp symptom markers is far from evident. Clearly, the customary POW-versus-control analyses and analyses that use internal POW controls yield complementary types of findings. In the next chapter, all of these findings are brought together in a comprehensive discussion.

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.