Persistence of Symptoms in Veterans of the First Gulf War: 5-Year Follow-up

Background During the 1990–1991 Gulf War, approximately 700,000 U.S. troops were deployed to the Persian Gulf theater of operations. Of that number, approximately 100,000 have presented medical complaints through various registry and examination programs. Objectives Widespread symptomatic illness without defining physical features has been reported among veterans of the 1991 Gulf War. We ascertained changes in symptom status between an initial 1995 symptom evaluation and a follow-up in 2000. Methods We assessed mailed symptom survey questionnaires for 390 previously surveyed members of the U.S. Department of Veterans Affairs Gulf War Registry for changes over the 5-year interval in terms of number and severity of symptoms. Results For the cohort as a whole, we found no significant changes in symptom number or severity. Those initially more symptomatic in 1995 showed some improvement over time, but remained much more highly symptomatic than those who had lesser initial symptomatology. Conclusions The symptom outbreak following the 1991 Gulf War has not abated over time in registry veterans, suggesting substantial need for better understanding and care for these veterans.

During the 1990-1991 Gulf War (GW), approximately 700,000 U.S. troops were deployed to the Persian Gulf theater of operations. Large-scale involvement of U.S. forces in combat in this region largely ended by the winter of 1991. Nevertheless, significant numbers of the 700,000 U.S. service personnel deployed to the region in 1990-1991 presented with medical complaints in the years following operations. Of that number, approximately 100,000 have presented medical complaints through various registry and examination programs (Brown et al. 2002;Stuart et al. 2002). The symptomatology that characterizes their complaints has been described in a number of studies, including investigations from the United Kingdom and other countries (Iowa Persian Gulf Study Group 1997;Lashof and Cassells 1998;Perisan Gulf Veterans Coordinating Board 1995;Unwin et al. 1999). These studies of both randomly sampled and selected populations show that GW veterans clearly report both increased numbers and severity of virtually all symptoms queried when compared with personnel not deployed to the region. Significantly, to date, these symptoms are not reliably associated with characteristic signs or diagnoses of pathological conditions (Eisen et al. 2005;Fukuda et al. 1998;Hodgson and Kipen 1999;Wessely 2004).
Only one study has made an attempt to longitudinally follow GW veterans to determine how their symptomatic presentations have evolved-whether there has been improvement, stability, or decline in the selfreported health of individuals and groups of GW veterans (Hotopf et al. 2003(Hotopf et al. , 2004. This information is critical in trying to understand not just the etiology of such complaints but also how to plan for continued care and rehabilitation. We have studied, both clinically and by survey, various groups of GW veterans since 1995 (Fiedler et al. 2006;Hallman et al. 2003;Kipen et al. 1999;Peckerman et al. 1999;Pollet et al. 1998). Our original investigations were from a random sample of the Department of Veterans Affairs (VA) GW registry members. We found that physician diagnoses, including those without a generally accepted etiology such as chronic fatigue syndrome, multiple chemical sensitivities, and fibromyalgia, could not account for much of the symptomatology presented and that, based on symptom endorsement and severity, veterans on the registry could be robustly classified as either highly symptomatic (40% of subjects) or mildly symptomatic (60% of subjects) . The large number of highly symptomatic "cases" (i.e., self-reported sick individuals), as well as representation from all four service branches, including National Guard and Reserve units, makes the VA registry a robust population from which to investigate the characteristics, although not the prevalence, of GW illnesses. We are currently studying more representative random populations of GW veterans (Fiedler et al. 2006). As part of this random sample of deployed and nondeployed veterans surveyed in 2000, we took advantage of the opportunity to do a 5-year follow-up of our original cohort of individuals from the registry. We hypothesized that the symptom reporting would remain relatively constant over the 5-year frame and that the highly symptomatic cluster would show a greater tendency toward decline in their health status (more symptomatology) after controlling for demographic variables.

Materials and Methods
Participants. Participants were U.S. military veterans from the Gulf War Health Registry maintained by the VA. In 1995 (time 1), veterans residing in seven states-Delaware, Illinois, New Jersey, New York, North Carolina, Ohio, and Pennsylvania-were selected by VA's Environmental Epidemiology Service using a simple random sampling procedure. The veterans were originally selected as part of recruitment for clinical studies of GW illness Nelson et al. 2001;Pollet et al. 1998). A random subset of veterans who completed that survey formed the sampling frame for the present study conducted in 2000 (time 2). Procedure. At time 1, we mailed introductory letters and questionnaires with postagepaid response envelopes to 2,011 veterans. We followed this with reminder postcards, a second (identical) letter and questionnaire, and a maximum of three follow-up phone calls at intervals of approximately 2 weeks until a response was received, ultimately yielding 1,161 responses (60%).
At time 2, The U.S. Department of Defense's Defense Manpower Data Center (DMDC) provided updated addresses and demographic information on those who had responded at time 1. These veterans were sent a letter describing the project and requesting their consent to be interviewed. Records from the Internal Revenue Service and national directory assistance databases (e.g., TeleMatch, Environmental Health Perspectives • VOLUME 114 | NUMBER 10 | October 2006 Springfield, VA) were used to locate veterans whose letters were returned without forwarding information. These databases did not include cellular telephone numbers.
This 5-year follow-up of the registry cohort was included as a part of a larger investigation of a national random sample of deployed GW veterans (Fiedler et al. 2006). As part of this study, after informed consent was obtained, all subjects were interviewed by a computer-assisted telephone interview (CATI) method. However, to allow direct follow-up comparisons to the data collected by mail at time 1 and to compare data collection modalities, 60% of those in our original registry cohort were randomly mailed a one-page questionnaire, comprising only symptom questions, 2 weeks before their telephone interview (Brewer et al. 2004). The veterans reported an average of five more symptoms via mail than via telephone, mainly as the result of mild symptoms reported by mail that were not reported at all during the telephone interview (Brewer et al. 2004). As a result, we used this one-page written symptom questionnaire as the basis of our comparisons. Additional demographic and other health information was collected as part of the subsequent CATI interview.
Measures. Participants were asked to examine a list of 48 symptoms identical to that used at time 1 (Table 1). For each symptom, the respondents indicated whether they had experienced "persistent or recurring" problems within the last 6 months, and if so, whether the problems they had experienced were "mild," "moderate," or "severe." For each symptom, a severity index was calculated by coding "no" as 0, "mild" as 1, "moderate" as 2, and "severe" as 3.
Based on factor and cluster analyses of their time 1 symptom data, each respondent had been classified as belonging to one of two clusters: a) veterans reporting good health and few moderate/severe symptoms (n = 242; ~60%), and b) veterans reporting fair/poor health and endorsing an average of 37 symptoms, 75% as moderate/severe (n = 148; ~40%). Thus, there was a total of 390 respondents who completed both the time 1 and time 2 symptomatology assessments ( Table 2) .
Statistical analyses. To assess changes in symptoms over time, we conducted paired samples t-tests on the average number of symptoms and average severity of symptoms endorsed at time 1 and time 2. We performed repeated measures MANCOVA (multiple analysis of covariance) to examine the main effect of time on the severity of 48 symptoms by controlling for age (mean centered), rank, race, marital status, sex, education, branch of service, duty (active or National Guard/Reserve), smoker status, and cluster membership ("mildly" or "highly" symptomatic as classified at time 1). In addition, we performed an ANCOVA (analysis of covariance) to examine the main effect of cluster membership at time 1 on change in number of symptoms over the 5-year period. The change score was calculated by subtracting time 1 scores from time 2 scores. We used the following variables as controls: age (mean centered), sex, rank, race, marital status, education, branch, duty, and smoking status. For tests of main effects and interactions, we adopted p < 0.05 as the critical value. To adjust for multiple comparisons when comparing changes in the 48 individual symptoms, we employed a Bonferroni correction, adopting as a critical value p < 0.001 (0.05/48).

Results
Response rates. Time 1. The results for time 1 have been fully described by Hallman et al. (2003). Of the 1,935 deliverable questionnaires, 1,161 were completed and returned by the respondents, yielding a response rate of 60%. Using chi-square analyses, we tested for potential selection biases and found no significant differences in the distribution of branch of service, duty status (active, Reserve, National Guard), or sex between those randomly selected to be in the sample and those in the registry as a whole.
Logistic regression analyses suggested no significant differences in response rates attributable to sex, date of entry into the registry, branch of service, type of duty (active, Reserve, or National Guard), or rank [enlisted, noncommissioned officer (NCO), warrant officer, officer]. Multivariate analysis of variance also revealed no significant differences in response rates attributable to either specific symptoms or specific diagnoses as determined by VA examining physicians and independent questionnaires completed at the time of registry enrollment (generally, months to years prior to time 1 study). Time 2. Updated DMDC data and data collected by telephone indicated that 70 of the original 1,161 registry veterans were ineligible to participate in the follow-up study. These included veterans who were deceased (n = 20), not deployed to the Gulf (n = 39), incarcerated (n = 3), or incapacitated (e.g., loss of hearing; n = 8). Of the remaining 1,091 eligible to participate, 640 (60%) were mailed a questionnaire, of which, 453 (71%) were completed and returned. Forty-five (7%) could not be delivered, and no forwarding address could be found. Of those who completed the questionnaire, 426 were contacted by telephone and by mail, and 398 of these completed both time 1 and time 2 assessments. Thus, 62% of those mailed a questionnaire at time 2 completed it and the subsequent telephone interview.
Eight cases were excluded because of missing data or because the demographic categories to which they belonged were too small to permit meaningful analyses. These included three veterans who served in the Coast Guard, one widowed veteran, three who indicated the "other" category for race, and one for whom rank was unknown, leaving 390 subjects for further analysis.
These 390 respondents did not differ significantly from the 1,161 respondents who participated at time 1 in terms of sex (8.5% females vs. 9.6% of total); rank (12.6% officers vs. 10.1% of total); duty status at mobilization (57.5% active vs. 59.0% of total); or branch (70.0% Army vs. 70.2% of total). They were significantly different from the 1,161 originally studied in terms of age at time 2 (42.7 years vs. 41.7 years for total), racial composition (16.9% African American vs. 23.2% of total; p < 0.006, two-sided), and education (22.8% college graduates vs. 19.1% of total; p < 0.02). In addition, t-tests with Bonferroni correction revealed no significant differences in the number of symptoms or their average severity reported at time 1 by the 390 current respondents and the remainder of the time 1 cohort.
Change in number of reported symptoms over time. At time 1, the mean number of symptoms (± SD) reported by the sample of 390 was 22.07 ± 12.92. At time 2, the mean number of symptoms increased slightly (0.67 symptoms) to 22.74 ± 12.89; however, this change was not significant [t(389) = -1.48; p = 0.14]. Average symptom severity also increased slightly from 0.83 ± 0.62 at time 1 to 0.84 ± 0.62 at time 2. Again, this change was not significant [t(398) = -0.52; p = 0.61]. Results of ANCOVA on change in number of symptoms controlling for age (mean centered), sex, rank, race, marital status, education, branch, duty, and smoking status revealed that the effect of time 1 cluster membership was significant [F(1, 379) = 31.84; p < 0.001] ( Table 3). Those veterans classified as mildly symptomatic at time 1 showed an increase in number of symptoms over time (mean, 2.33), and those veterans who were highly symptomatic showed a decrease in symptoms at time 2 (mean, -2.04). Also, there were significant effects of race [F(1, 379)  Next, we looked at tests of between-subject effects for the variables that yielded significant multivariate tests to examine the severity changes in individual symptoms. The main effect of age was significant for "persistent or recurring problems" with "arms, hands, and shoulders," "back problems," "frequent or painful urination," "sexual or genital problems," "pain in arms or legs," and "pain in more than one joint," with the severity of all these symptoms increasing more in older individuals. The main effect of race was significant for "persistent or recurring problems" with "constipation," "hair," and "sweating not due to exercise," with all three symptoms increasing more in severity among African-American veterans (Table 4). The main effect of cluster membership was significant for "persistent or