The Postwar Hospitalization Experience of Gulf War Veterans Possibly Exposed to Chemical Munitions Destruction at Khamisiyah, Iraq

American Journal of Epidemiology Copyright 01999 by The Johns Hopkins University School of Hygiene and Public Health All rights reserved vol. 150, No. 5 Printed In U.SA. The Postwar Hospitalization Experience of Gulf War Veterans Possibly Exposed to Chemical Munitions Destruction at Khamisiyah, Iraq Gregory C. Gray, 1 Tyler C. Smith, 1 James D. Knoke, 1 and Jack M. Heller 2 Using Department of Defense hospital data, the authors examined the postwar hospitalization experience from March 1991 through September 1995 of US Guff War veterans who were near Khamisiyah, Iraq, during nerve agent munition destruction in March 1991. Multiple sources of meteorologic, munition, and toxicology data were used to circumscribe geographic areas of low level, vaporized nerve agent for 4 days after the destruction. Plume estimates were overlaid on military unit positions, and exposure was estimated for the 349,291 US Army Gulf War veterans. Exposure was classified as (n = 224,804), uncertain low dose exposure (n = 75,717), and specific estimated subclinical exposure (n = 48,770) categorized into three groups for doseresponse evaluation. Using Cox proportional hazard modeling, the authors compared the postwar experiences of these exposure groups for hospitalization due to any cause, for diagnoses in 15 unique categories, and for specific diagnoses an expert panel proposed as most likely to reflect latent disease from such subclinical exposure. There was little evidence that veterans possibly exposed to the nerve agent plumes experienced unusual postwar morbidity. While there were several differences in hospitalization risk, none of the models suggested a dose-response relation or neurologic sequelae. These data, having a number of limitations, do not support the hypothesis that Gulf War veterans are suffering postwar morbidity from subclinical nerve agent exposure. Am J Epidemiol 1999; 150:532-40. environmental exposure; hospitalization; military medicine; military personnel; morbidity; Persian Gulf syndrome; veterans During the Persian Gulf War, Iraq was known to possess chemical weapons (1, 2). Shortly after the 1991 fighting ended, expert panels evaluated available exposure and medical data and concluded that there was no evidence that Iraq used these weapons of mass destruction during the conflict (3). In June 1996 (4), the US Department of Defense announced United Nations' findings that chemical agents had been destroyed by US forces near Khamisiyah, Iraq, during March 1991. In a series of follow-up reports, the Department of Defense announced that US soldiers who were near Khamisiyah may have been exposed to the plume from munition destruction (5). Subsequently, there has been speculation that Gulf War veterans located under the Received for publication June 16, 1998, and accepted for publication January 22, 1999. Abbreviations: ICD-9, International Classification of Diseases, Ninth Revision; RR, risk ratio. 1 Emerging Illness Division, Department of Health Sciences and Epidemiology, Naval Health Research Center, San Diego, CA. 2 Deployment Environmental Surveillance Program, US Army Center for Health Promotion and Preventive Medicine, Aberdeen Proving Ground, MD. Reprint requests to Dr. Gregory C. Gray, Naval Health Research Center, Emerging Illness Division, P.O. Box 85122, San Diego, CA 92186-5122. plume may be suffering increased postwar morbidity as a result of this exposure (6-8). This report documents our investigation into the postwar hospitalization experience of these veterans. MATERIALS AND METHODS Objective The objective for this investigation was to compare the postwar hospitalization experience of Gulf War veterans who were possibly exposed to a plume from munition destruction at Khamisiyah, Iraq, in March 1991, with that of other Gulf War veterans who were not likely to have been exposed. Data sources This study was conducted in accordance with Protection of Human Subjects' guidelines from the US Department of the Navy. As previously described (9-11), demographic, hospitalization, and deployment data for the Gulf War veterans were obtained from the Defense Manpower Data Center, Monterey Bay, California. Geographic information system data were 532

Gulf War Veterans and Khamisiyah 533 compiled by the US Army Center for Health Promotion and Preventive Medicine, Aberdeen Proving Ground, Maryland. Descriptive data Descriptive data available for analysis included the social security number, sex, date of birth, pay grade, race, ethnicity, home of record, service branch, service type, marital status (during deployment period), Department of Defense primary occupational specialty, active-duty service entry date, beginning and ending date of Gulf War service, and date of separation from military service. The service branch (Navy/Coast Guard, Army, Air Force, Marine Corps) was coded to reflect military service on February 21, 1991. Age, as of September 30, 1990, was categorized so as to form approximate quartile age groups: 17-21 years, 22-25 years, 26-31 years, and ^32 years. Occupation (226 unique codes) was divided into 10 major groups, as defined by the Department of Defense Occupational Conversion Manual (12). In an effort to be consistent with previous military categorizations and yet maintain group size for robust modeling, we combined race and ethnicity into one variable with four categories: White, Black, Hispanic, and other. Length of service, calculated from the active-duty service entry date and date of separation variables, was categorized into 0-2 years, 3-4 years, 5-9 years, and 10-42 years. Pay grade was categorized into enlisted, warrant officer, and commissioned officer. Length of service and pay grade were then used to define a five-category monthly salary variable: less than $1,000, $1,000-$ 1,399, $1,400-$2,099, $2,100-$3,199, and >$3,199. Marital status was categorized into married or not married. Hospitalizatlon data Hospitalization data were captured from all Department of Defense hospitals from March 10, 1991, through September 30, 1995, and included date of admission, up to eight individual discharge diagnoses, and disposition after hospital discharge. Additionally, as per previous research (9,10), a prewar hospitalization covariate (coded as yes or no) was used to reflect a hospital admission during the 12 months prior to August 1,1990. Diagnoses were coded according to the International Classification of Diseases, Ninth Revision (ICD-9) (13). For these analyses, we ignored the decimal component of ICD-9 diagnoses and considered diagnoses with the same whole number (up to three digits) to be the same. No hospitalization data were available from hospitals outside the Department of Defense system. No outpatient data were available. Geographic information system data Troop locations on specific days were recorded using an Oracle Relational Database Management System (Oracle Corporation, Redwood Shores, California) and Intergraph Modular GIS Environment software (Intergraph Corporation, Huntsville, Alabama) on a Windows NT-based Intergraph TD Series computer (Intergraph Corporation). Geographic information system data were compiled from troop movement information (US Armed Services Center for Research of Unit Records, Springfield, Virginia), linked by Unit Identification Codes to the roster of Persian Gulf War veterans. As individual veteran location data were not available on a daily basis, troop movement was normally tracked by companies of from 50 to 120 individuals. Generally, these companies or units traveled together with a geographic spread not exceeding 400 m. However, some individuals and small groups of veterans did occasionally separate from their unit. The geographic information system contained all records of daily unit locations (latitude and longitude) that could be found following the war. These data had limitations. Not all units reported daily location data in their journals, logs, and messages. Units were widely dispersed throughout the region, and constructing the geographic information system from thousands of paper records was difficult and not without possible error. During 1997 and 1998, the Office of the Special Assistant to the Deputy Secretary of Defense for Gulf War Illnesses interviewed approximately 150 military operational officers to optimize troop movement data for March 10-13, 1991, the time frame of the Khamisiyah munitions destruction. Nerve agent exposure estimates Intelligence information (14) and demolitions (5) concerning Khamisiyah have been previously reported in detail. In summary, Khamisiyah was a large ammunition storage facility, located in southern Iraq, approximately 25 km southeast of the city of An Nasiriyah. The site contained numerous ammunition bunkers, storage buildings, pits, and sand mounds to protect stored weapons. During March 1991, Army engineers operating from remote sites destroyed many of the bunkers, warehouses, and stored weapons at the facility. The chief focus of concern regarding Khamisiyah is the March 10, 1991, destruction of a cache of 1,250 rockets, which were stored in an open pit. At that time, Army engineers did not have evidence that any of the munitions stored at Khamisiyah contained nerve agents. However, in May 1996, the

534 Gray et al. United Nations Special Commission inspectors examined demolition debris and determined that some of the destroyed rockets contained the nerve agents sarin and cyclosarin. Rocket destruction modeling An estimated 1,250 rockets were present at Khamisiyah, each containing 6.3 kg of 50 percent sarin and cyclosarin at a 3:1 ratio. Lacking empiric data on the nerve agent release caused by open-air detonation of these rockets, the Department of Defense and Central Intelligence Agency jointly conducted destruction testing using simulated rockets containing simulated nerve agent. The rocket destruction modeling was used to determine how much nerve agent was released instantaneously and how much was released over time by evaporation. This simulation, combined with intelligence data, led to the estimate that 342 gallons (1,294.57 liters) of nerve agent were released on March 10, 1990, from 500 damaged rockets (15). It was further determined, from interviews with the engineers that placed the charges at Khamisiyah and from the multiple simulations, that 2 percent of the agent was released instantaneously (1 percent as droplets and 1 percent as vapor), and that 16 percent was released over time by evaporation from the soil (6 percent) and from wooden rocket packing crates (10 percent). Meteorologic transport and diffusion modeling All Khamisiyah modeling was guided by an expert panel composed of numerous nonfederal and federal experts in meteorology and atmospheric modeling. Meteorologic modeling was conducted to simulate the weather conditions in the vicinity of Khamisiyah on March 10, 1991, and succeeding days. Results from three global (synoptic) weather models and three regional (mesoscale) high resolution weather models were used (15). The meteorologic reconstruction was based upon a number of data sources including the following: global observation data from the Persian Gulf region during March 1991, Saudi Arabian surface and radiosonde observations, declassified surface data collected by the US Air Force and Special Forces in the Khamisiyah region, and Navy ship and satellite data. In addition, soot vector patterns from the Khamisiyah bunker explosions and fires and smoke dispersion from the Kuwait oil fires were used to validate the modeling result. The meteorologic models used in this effort were a combination of civilian university models and military models, many of which have been peer reviewed and published. Three Department of Defense transport and diffusion models were selected by an expert committee for exposure modeling. Two of the diffusion models have been validated, one in a series of field tests in the midwestern United States, and one by field trial data from at least 60 reports on chemical and biologic agent simulation releases. These methods have historically been used by the Department of Defense to model chemical agent dispersion for military tactical analysis. The three dispersion models were combined with the meteorologic models to generate five estimates or simulations of daily plume coverage (15). Exposure plumes and exposure estimates Since no US personnel were known to have died or to have been incapacitated during the time period of concern, the Department of Defense defined two lesser nerve agent concentrations of interest (15). The first noticeable effects concentration, 1 mg-minute/m 3, was defined as the dosage expected to cause mild symptoms, such as rhinorrhea, muscle twitching, chest tightness, and headache. The general population limit (hereafter termed low level exposure) concentration, 0.0126 mg-minute/m 3, was the dosage below which the general population, including children and the elderly, could endure for at least 72 hours without symptoms. Two different types of chemical agent plumes were used in these analyses. The union of five different meteorologic/dispersion model simulations, termed the notification plume, was used to inform Gulf War veterans in early 1997 of potential chemical agent exposure. These model simulation contours represent a 99 percent probability that persons exposed to the general population limit dosage would fall within that perimeter (15). This union methodology was selected upon advice from an independent expert review panel. After much deliberation in November 1997, a similar independent expert panel recommended that a smaller epidemiologic plume model be constructed from the combination of the "best" meteorologic and dispersion models for unit-specific dose estimation. This epidemiologic plume or footprint enabled epidemiologists to estimate individual exposure by quantifying the nerve agent concentration at specific troop locations over time. Because an updated mesoscale meteorologic model was used with one of the three dispersion models, the epidemiologic plume was not completely contained in the notification plume. In the future, the Department of Defense plans to redefine the notification plume to include the new exposure estimate (epidemiology plume). Plumes were estimated for each day for the period March 10-13, 1991. The daily locations of the vapor

Gulf War Veterans and Khamisiyah plumes were overlaid on a geographic information system troop unit location base map (figure 1). symptoms, signs, and ill-defined conditions; and injury and poisoning. The aggregation of many diagnoses into large ICD9 categories might mask population risk differences due to individual diagnoses. We therefore also chose to examine specific ICD-9 diagnoses, suggested by an expert panel as possible manifestations of subtle, nerve agent-induced, neurophysiologic effects: mononeuritis, peripheral neuropathy, toxic neuropathy, and myoneural disorders and myopathies (6). For each veteran, hospitalizations (if any) were scanned in chronologic order, and diagnostic fields were scanned in numeric order for the ICD-9 codes of interest. Only the first hospitalization meeting the outcome criteria was counted for each veteran. Only women were included in the analyses for the complications of pregnancy, childbirth, and the puerperium category. Study outcomes With a focus on vapor plume exposure, risks of hospitalization for "any cause" and hospitalization with a diagnosis in each of 15 broad ICD-9 diagnostic categories (13) were examined: infection and parasitic diseases; neoplasms; endocrine, nutritional, and metabolic diseases and immunity disorders; diseases of the blood and blood-forming organs; mental disorders; diseases of the nervous system and sense organs; diseases of the circulatory system; diseases of the respiratory system; diseases of the digestive system; diseases of the genitourinary system; complications of pregnancy, childbirth, and the puerperium; diseases of the skin and subcutaneous tissue; diseases of the musculoskeletal system and connective tissue; Day 1 - March 10,1991 Iraq Day 2 - March 11,1991 -Khamisiyah Saudi Arabia Notification Khamisiyah Iraq Iran Epidemiologic 535 Iran Epidemiologic Notification Saudi Arabia V Aii% ), Persian * Gulf 0 50 100 200 Km 100 150 Miles I00 0 Day 3 - March 12,1991 50 100 150 Day 4 - March 13,1991 -Khamisiyah Khamisiyah Iran Epidemiologic (See Inset- Right) Saudi Arabia 0 0 50 100 200 Km 100 150 Miles FIGURE 1. Geographic information system plots of US Army units with plume estimate overlays after the March 10, 1991, munition destruction at Khamisiyah, Iraq. Am J Epidemiol Vol. 150, No. 5, 1999

536 Gray et al. Statistical analyses Using SAS software (Version 6.12; SAS Institute, Inc., Cary, North Carolina), Cox proportional hazard modeling was used to obtain risk ratios, 95 percent confidence intervals, and probabilities of hospitalization as a function of time (at the mean value of the included covariates). Follow-up time was calculated from March 10, 1991, until hospitalization, separation from service, or September 30, 1995, whichever occurred first. Because no separation data were available for reserve personnel, they were assumed to have remained in active service until June 10, 1991 (at which time most veterans had returned home), and then were removed from further modeling consideration (censored). RESULTS No units were identified as having been exposed to the first noticeable effects concentration or higher in the vapor plume modeling. However, considering the union of the notification and epidemiologic plumes, complete data were available for 124,487 Army Gulf War veterans (95,402 (76.6 percent) regular active-duty and 29,085 (23.4 percent) reserve) who were determined to have the possibility of at least low level exposure during the period March 10-13, 1991 (figure 1). Possibly exposed Gulf War veterans were categorized into four groups: uncertain low dose exposure (n = 75,717); 0.0-0.01256 mg-minute/m 3 (exposure 1, n = 18,952); 0.01257-0.09656 mg-minute/m 3 (exposure 2, n = 23,061); and 0.09657-0.51436 mg-minute/m 3 (exposure 3, n = 6,757). As all of these possibly exposed individuals were US Army personnel, they were compared with 224,804 other Army Gulf War veterans (177,343 (78.9 percent) regular active-duty and 47,461 (21.1 percent) reserve) who had complete data and were deployed to the Gulf War theater (9) at the same time but whose units were not under the circumscribed vapor plumes (). More than 99.5 percent of all Army Gulf War veterans had complete data. During the 54 months of observation of regular active-duty personnel, hospitalizations occurred in 21.6 percent of the group; 21.5 percent of the uncertain low dose group; 23.0 percent of the exposure 1 group; 21.0 percent of the exposure 2 group; and 21.5 percent of the exposure 3 group. Similarly, considering only regular active-duty personnel, 48.4 percent of the group, 49.1 percent of the uncertain low dose group, 49.6 percent of the exposure 1 group, 47.0 percent of the exposure 2 group, and 49.9 percent of the exposure 3 group remained on active duty during the entire 54 months of observation. Based upon univariate comparisons for hospitalizations occurring during the study period and preliminary Cox proportional hazard modeling with all covariates, the following covariates were selected for the further multivariate modeling: sex, service type, age group, marital status, race/ethnicity, occupational category, pay grade, prewar hospitalization, and vapor plume exposure. Home of record and length of service covariates were not important to the model. Salary was dropped from the analyses because of colinearity with age group. Race/ethnicity was reduced to two categories: White and other. The occupational category was reduced to nine categories by combining undesignated with missing. The records missing marital status were aggregated with those in the single category. The four groups of possibly exposed veterans had similar adjusted risks of "any cause" hospitalization during the March 10, 1991, to September 30, 1995, time period compared with veterans (table 1), and the corresponding probability plots were nearly parallel for the 54 months of follow-up (figure 2). However, the multivariate model (table 1) revealed other better predictors of postwar hospitalization, which included female sex (risk ratio (RR) = 2.63), prewar hospitalization (RR = 1.65), enlisted pay grade (RR = 1.50), and reserve service type (RR = 1.33). Cox proportional hazard modeling for each of the 15 diagnostic categories over the 54 months was performed (table 2). There was no evidence that possible nerve agent exposure was associated with postwar hospitalizations. In a similar fashion, Cox proportional hazard modeling was also conducted for diagnoses thought most likely to be associated with subtle, nerve agent-induced, neurophysiologic effects (6). These analyses revealed no increased risk for personnel possibly exposed to the vapor plume (table 3). Additionally, when several such diagnoses were combined, there was no indication of an increase in risk among the possibly exposed groups. Similarly, Cox proportional hazard models using only the notification plume data (yes or no exposure) and only the epidemiologic plume data (yes or no exposure) were performed for "any cause" and the 15 diagnostic category hospitalizations (data not shown). Only one of these 32 additional models suggested an association of possible exposure and postwar morbidity. Possibly exposed veterans, as defined by the notification plume alone, were slightly at increased risk (RR = 1.07) of hospitalizations due to mental disorder diagnoses (data not shown). Further modeling for the 10 most frequent diagnoses in this category revealed that possibly exposed veterans were more likely to be hospitalized because of the mental disorder diagnoses' adjustment reaction (RR =

Gulf War Veterans and Khamisiyah 537 TABLE 1. Adjusted risk ratios for "any cause" postwar hospttallzation among regular active-duty US Army Gulf War veterans In Department of Defense hospitals, March 10,1991, to September 30,1995 Descriptive characteristic Under plume* Uncertain low dose Exposure 1 Exposure 2 Exposure 3 Prewar hospltalization Yes no Reserve Yes no Sex Female male Age (years) 17-21 26-31 22-25 26-31 32-83 26-31 Married Yes no Race White other Pay grade Enlisted officer Warrant officer officer Occupationt Infantry, gun crews Communications and intelligence Health care Other technical Administration Electrical and mechanical repair Construction and related trades Supply handlers Undesignated and missing Risk ratio 0.95 0.96 0.93 1.02 1.65 1.33 2.63 1.10 1.07 1.18 1.07 0.93 1.50 1.22 1.07 1.01 1.25 1.06 1.02 1.04 1.16 1.12 1.18 95% confidence interval 0.93, 0.97 0.92, 0.99 0.90, 0.96 0.96, 1.08 1.61,1.69 1.27,1.40 2.57, 2.70 1.07, 1.13 1.04, 1.09 1.15, 1.20 1.04, 1.09 0.91,0.94 1.45, 1.55 1.15, 1.30 1.03, 1.12 0.97, 1.06 1.19,.31 0.99, 1.13 0.98,.07 0.99,.09 1.08,.24 1.07,.17 1.12,.23 Reflects nerve agent exposure estimates: uncertain low dose; exposure 1 (0.0-0.01256 mg-minute/m 3 ); exposure 2 (0.01257-0.09656 mg-mlnute/m 1 ); exposure 3 (0.09657-0.51436 mg-minute/m 3 ). t Referenced to electronic equipment repair personnel. 1.11) and nondependent use of drugs (RR = 1.27) (data not shown). DISCUSSION The acute and long-term effects of sarin and cyclosarin, as well as other chemical agents, have been previously reviewed (3, 6, 16, 17). Both nerve agents are organophosphate compounds and acutely cause miosis, a runny nose, increased salivation, muscle contractions, headache, convulsions, and respiratory paralysis. Expert panels have reviewed intelligence and military data and concluded that there was no evidence that Iraq used such chemical agents during the Gulf War (3, 6). Although the probability is arguably low (16), the lack of evidence of acute poisoning among US military personnel does not rule out the possibility of subclinical, low-level exposure to these agents after weapon destruction. Data are sparse regarding the long-term effects of low-dose sarin or cyclosarin exposure. Organophosphates cause subtle electroencephalographic changes in some subjects, but the general health implications of these changes are uncertain (16). Limited data from these nerve and other organophosphates suggest that, if chronic manifestations were associated with low-level nerve agent exposure, they would likely be manifested as neurophysiologic (especially sensory neuropathies) or neuropsychologic effects (6). However, other specific diagnostic associations cannot easily be ruled out Hence, we sought to broadly and specifically compare the postwar hospitalization morbidity of GuLf War veterans who were possibly exposed to the vapor plume from the munition destructions with that of other similar Gulf War veterans. We examined three exposure modeling strategies and performed Cox proportional hazard models for "any cause" and large diagnostic category postwar hospitalizations' outcomes. Consistent with other hospitalization analyses (9), women, married personnel, persons who were hospitalized before the war, health care personnel, and enlisted personnel were at increased risk of hospitalization for "any cause" after the war. However, only one of 48 models suggested an increased postwar risk. This finding (using the notification plume exposure estimate alone) was small in magnitude and due to adjustment reaction and nondependent drug abuse diagnoses known to be associated with combat (9, 18, 19). It is quite probable that Army personnel possibly exposed to the nerve agent plumes in Iraq experienced more combat stress than many of their Army Gulf War veteran counterparts who remained in Saudi Arabia. The benefits (20) and limitations (21) of epidemiologic studies of hospitalization data have been previously described. These data permit the comparison of morbidity severe enough to warrant hospital admission. Milder morbidity could not be compared. Often Gulf War veterans have complained of symptoms not requiring hospital admission (22, 23). However, symptom studies are easily confounded by other factors, such as recall bias (9, 24), psychologic disorders (25), stress reactions (17, 26), desires for compensation (26), and media influence (27). More severe morbidity, such as that causing hospitalization, requires validation by a health care delivery team and is a more objective comparison of health differences. Additionally, where symptom studies may suffer from low participation (23, 28, 29) and missing data, hospitalization data are much more complete, because active-duty personnel have ready access to medical care and are rarely hospitalized

538 Gray et al. 0.4 n 0.3 0.2 0.1- Not exposed (A) Uncertain, low dose (B) Exposure 1 (Q Exposure 2 (D) 00.0 Exposure 3 (E) Years from March 10, 1991 FIGURE 2. Probability of hospitaiization for "any cause" in US Department of Defense hospitals during the period March 10, 1991, to September 30, 1995, by possible nerve agent plume exposure in March 1991. TABLE 2. Adjusted exposure risk ratios for postwar hospitaiization due to diagnoses In large International Classification of Diseases, Ninth Revision (ICD-9), categories among regular active-duty US Army Gulf War veterans In Department of Defense hospitals, March 10,1991, to September 30,1995 Major ICD-9 diagnostic categories Infection and parasitic diseases Neoplasms Endocrine, nutritional, and metabolic diseases Blood diseases Mental disorders Nervous system diseases Circulatory system diseases Respiratory system diseases Digestive system diseases Genitourinary system diseases Complications of pregnancy Skin diseases Musculoskeletal system diseases Symptoms, signs, and ill-defined conditions Injury and poisoning Uncertain low closet 0.86 (0.78, 0.94)$ 1.00(0.90,1.12) 0.75 (0.62, 0.92) 1.06(0.78, 1.45) 1.07(1.01, 1.15) 0.94(0.85,1.04) 0.94(0.85, 1.05) 0.92 (0.85, 0.99) 0.98(0.93,1.03) 0.93 (0.86, 1.00) 0.96(0.89, 1.02) 1.06(0.95,1.19) 0.90 (0.86, 0.94) 1.04(0.95,1.13) 0.95 (0.90, 0.99) Exposure 1f 0.90(0.75,1.07) 1.24(1.04, 1.47) 0.75(0.52, 1.07) 0.52(0.25, 1.12) 1.00(0.88, 1.13) 0.96(0.80, 1.15) 1.12 (0.93, 1.33) 0.90(0.77, 1.04) 1.03(0.94, 1.12) 1.07(0.95,1.21) 0.80 (0.72, 0.89) 0.77 (0.60, 0.98) 0.90 (0.83, 0.98) 1.07(0.92,1.24) 1.00(0.91,1.09) Risk ratio* Exposure 2t 0.79 (0.67, 0.94) 1.04(0.87, 1.24) 0.63 (0.44, 0.90) 0.71 (0.38, 1.30) 0.95(0.85, 1.07) 0.85(0.72,1.02) 0.98(0.83, 1.17) 0.89(0.79,1.02) 0.99(0.92,1.08) 0.97(0.86,1.10) 0.82 (0.73, 0.91) 0.92(0.76, 1.13) 0.90 (0.83, 0.96) 1.03(0.90, 1.19) 0.98(0.90, 1.06) Exposure 3t 1.20(0.95, 1.51) 0.74(0.52, 1.07) 0.88(0.52, 1.50) 1.33(062,2.83) 0.99(0.82,1.19) 1.04(0.79,1.36) 1.11 (0.84,1.45) 1.26(1.05,1.51) 1.01 (0.88,1.15) 0.89(0.70,1.11) 0.73 (0.57, 0.94) 0.93(0.67,1.30) 0.98(0.87,1.09) 1.08(0.85,1.36) 1.12(0.99, 1.27) * Risk ratios were adjusted for sex, age group, prewar hospitaiization, race, service type, marital status, pay grade, and occupation using Cox proportional hazard modeling. t Reflects nerve agent exposure estimates: uncertain low dose; exposure 1 (0.0-0.01256 mg-minute/m*); exposure 2 (0.01257-0.09656 mg-minute/m 3 ); exposure 3 (0.09657-0.51436 mg-minute/m 1 ). t Numbers In parentheses, 95% confidence interval. Only women were considered.

Gulf War Veterans and Khamisiyah 539 TABLE 3. Adjusted risk ratios for hospltallzatlon in US Department of Defense facilities with select neurologic conditions thought likely to be associated with subtle, nerve agent-induced, neurophyslologlc effects, March 10,1991, to September 30,1995 ICD-9* code 354 355 356 357 358 359 AN Specific diagnosis Mononeuritis of upper limb Mononeuritis of lower limb Hereditary and idiopathic peripheral neuropathy Inflammatory and toxic neuropathy Myoneural disorders Muscular dystrophies and other myopathies All of the above diagnoses Total 347 258 23 9 5 9 637 Cases among not exposed 225 169 16 6 4 7 421 Uncertain low doset Exposure ^t Riskratiot Exposure^ Exposure 3t 0.90(0.68, 1.18) 1.41 (0.95, 2.09) 0.80 (0.50, 1.28) 1.28 (0.66, 2.50) 0.95 (0.70, 1.29) 1.03 (0.60, 1.79) 1.05 (0.65, 1.71) 0.69 (0.25, 1.85) D 0.88(0.72, 1.08) 1.24(0.91,1.69) 0.85(0.61,1.19) 0.94(0.54,1.63) * ICD-9, International Classification of Diseases, Ninth Revision. t Risk ratios were adjusted for sex, age group, prewar hospitalization, race, service type, marital status, pay grade, and occupation using Cox proportional hazard modeling. $ Reflects nerve agent exposure estimates: uncertain low dose; exposure 1 (0.0-0.01256 mg-minute/m 3 ); exposure 2 (0.01257-0.09656 mg-minute/m 3 ); exposure 3 (0.09657-0.51436 mg-minute/m 3 ). Numbers In parentheses, 95% confidence interval. H Empty cells reflect inability to compare risk due to sparse outcome data. outside the Department of Defense system (9, 30). Routine periodic physical screening and frequent physical fitness testing of military personnel reinforce the detection of severe morbidity and the likelihood of hospitalization when appropriate. Even so, it is possible that some Gulf War veterans, fearing separation for medical disability, may have not sought medical care for symptoms related to Gulf War exposures. It is also possible that hospital admission and perhaps medical evaluation may have been somehow related to geographic exposure and mask important exposure-hospitalization associations. However, such biases are unlikely, because potential plume exposures at Khamisiyah were not recognized until 1996, after the end of the study period. These data permit only the comparison of Gulf War veterans who remained on active duty after the conflict or who retired with medical benefits. Hospital data are not available to evaluate personnel who left military service or military medical care. However, a recent study of the nonfederal hospitalization experience of Gulf War veterans did not demonstrate an association of neurologic conditions and Gulf War service (31). Additionally, military service data from the present study do not support the hypothesis that possibly exposed veterans were leaving the military faster because of illnesses. Finally, there was no difference in cohort participation in the Department of Defense's Gulf War veteran health registry (27) (data not shown). Study data are also time limited in that 54 months of study are a brief time to detect chronic disease development. More definitive evaluation of exposure and subsequent chronic disease development would require future studies of alternative design. Our analyses have numerous unique characteristics. They represent the first combined use of operational data (temporal and geographic) with dispersion and meteorologic data to estimate an exposure during the Gulf War. These analyses represent one of the first epidemiologic studies of Gulf War veterans to use other Gulf War veterans as a comparison group, thereby reducing potential confounding such as that due to a transient healthy worker selection effect (9, 30). These data also are important in that they include reserve personnel in the analysis. This study's large populations and numerous covariates permitted much statistical power to detect differences in hospitalization risk due to the estimated exposure. If plume-induced disease manifestations meriting hospitalization were common among possibly exposed veterans, it is very likely that these analyses would have detected this effect. Additionally, the adjusted risk ratio results for demographic and operational covariates were consistent with those of other postwar hospitalization studies (9, 10), suggesting model validation. In summary, we compared the postwar hospitalization experience of veterans with possible exposure to low-level, subclinical nerve agent vapor concentrations with that, of other Gulf War veterans. Our analyses included broad outcomes (any hospitalization and diagnoses from 15 large diagnostic categories) as well as specific diagnoses suggested by expert panels most likely to reflect new, chronic manifestations of subclinical acute nerve agent microexposure. None of these models suggested an increased risk among the possibly exposed. We next repeated the "any cause" and large diagnostic category models using each of two unique

540 Gray et al. plume component exposure estimates and found that only one of these 32 models suggested an increase in risk among the possibly exposed. This slight increase was due to adjustment reaction and nondependent drug abuse diagnoses and consistent wim medical literature regarding combat exposure. These data do not support the hypothesis that Gulf War veterans who were possibly exposed to nerve agent plumes from the March 1991 munition demolitions at Khamisiyah, Iraq, experienced unusual postwar morbidity. ACKNOWLEDGMENTS This represents report no. 97-35, supported by the Office of Naval Research, Arlington, Virginia, under Naval Medical Research and Development Command reimbursable work unit 6423. The authors thank the following individuals for their most helpful assistance and recommendations in conducting this research: Dr. Gary D. Gackstetter, Uniformed Services University of Health Sciences, Bethesda, Maryland; and Jeff Kirkpatrick and Chris Weir, Deployment Environmental Surveillance Program, US Army Center for Health Promotion and Preventive Medicine, Aberdeen Proving Ground, Maryland. A preliminary analysis of these data was presented in session 1017 at the 125th Annual Meeting of the American Public Health Association, San Diego, California, November 10, 1997. REFERENCES 1. Department of Defense. Conduct of the Persian Gulf War final report to Congress. Washington, DC: Department of Defense, 1992. 2. Zilinskas R. Iraq's biological weapons. JAMA 1997;278:418-24. 3. Defense Science Board. Final report: Defense Science Board Task Force on Persian Gulf War Health Effects. Washington, DC: Office of the Under Secretary of Defense for Acquisition and Technology, 1994. 4. Graham B, Brown D. US troops were near toxic gas blast in gulf. Washington Post 1996 June 22;Sect. A:l. 5. Rostker B. US demolition operations at the Khamisiyah ammunition storage point Washington, DC: Department of Defense, April 14, 1997. (http://www.gulflink.osd.mil/khamisiyah/). 6. Institute of Medicine Committee on the Evaluation of the Department of Defense Comprehensive Clinical Evaluation Program. Adequacy of the Comprehensive Clinical Evaluation Program: nerve agents. Washington, DC: National Academy Press, 1997. 7. United States General Accounting Office. Gulf War illnesses: improved monitoring of clinical progress and reexamination of research emphasis are needed. Washington, DC: United States General Accounting Office, 1997. 8. Priest D. Gas exposure in Gulf War revisited. Washington Post 1997 July 24;SecL A:28. 9. Gray GC, Coate BC, Anderson CM, et al. The postwar hospitalization experience of US veterans of the Persian Gulf War. N Engl J Med 1996;335:1505-13. 10. Knoke JD, Gray GC. Hospitalizations for unexplained illnesses among US Persian Gulf War veterans. Emerg Infect Dis 1998;4:211-19. 11. Knoke JD, Gray GC, Garland FC. Testicular cancer and Persian Gulf War service. Epidemiology 1998;9:648-53. 12. Occupational conversion manual: enlisted/officer/civilian. Washington, DC: Department of Defense, Office of the Assistant Secretary of Defense, Force Management, and Personnel, 1991. 13. International classification of diseases. Ninth Revision, clinical modification. 3rd ed. Washington, DC: US Department of Health and Human Services, 1991. 14. Persian Gulf War Illnesses Task Force. Khamisiyah: a historical perspective on related intelligence. Washington, DC: Central Intelligence Agency, 1997. 15. Walpole RD, Rostker B. Modeling the chemical warfare agent release at the Khamisiyah pit. Washington, DC: Department of Defense, September 4, 1997. (http://www.gulflink.osd.mil/ cia_092297/). 16. Perrotta DM. Long-term health effects associated with subclinical exposures to GB and mustard. A review conducted by the Environmental Committee of the Armed Forces Epidemiological Board. Falls Church, VA: Armed Forces Epidemiological Board, 1996. 17. Presidential Advisory Committee on Gulf War Veterans' Illnesses. Final report, December 1996. Washington, DC: US Government Printing Office, 1996. 18. Hobfoll S, Spielberger C, Breznitz S, et al. War-related stress: addressing the stress of war and other traumatic events. Am Psychol 1991;46:848-55. 19. Sutker P, Uddo M, Brailey K, et al. Psychological symptoms and psychiatric diagnoses in Operation Desert Storm troops serving graves registration duty. J Trauma Stress 1994;7:159-71. 20. Cowan DN, DeFraites RF, Gray GC, et al. The risk of birth defects among children of Persian Gulf War veterans. N Engl J Med 1997;336:1650-6. 21. Kuller L. The use of existing databases in morbidity and mortality studies. Am J Public Health 1995;85:1198-200. 22. The Iowa Persian Gulf Study Group. Self-reported illness and health status among Persian Gulf War veterans: a populationbased study. JAMA 1997;277:238-^5. 23. Unexplained illness among Persian Gulf veterans in an Air National Guard Unit: preliminary report August 1990-March 1995. MMWR Morb Mortal Wkly Rep 1995;44:443-7. 24. McCauley L, Joos S, Bourdette D, et al. Persian Gulf War illnesses: validation of self-reported symptoms by clinical examination. Presented at the 125th Annual Meeting and Exposition of the American Public Health Association, Indianapolis, Indiana, November 9-13, 1997. 25. Malone JD, Paige-Dobson B, Ohl C, et al. Possibilities for unexplained chronic illnesses among reserve units deployed in Operation Desert Shield/Desert Storm. South Med J 1996; 89:1147-55. 26. Kalinowsky LB. Problems of war neuroses in the light of experiences in other countries. Am J Psychiatry 1950;107:340-6. 27. Gray GC, Hawksworth AW, Smith TC, et al. Gulf War veterans' health registries. Who is most likely to seek evaluation? Am J Epidemiol 1998;148:343-9. 28. Haley R, Kurt T, Horn J. Is there a Gulf War syndrome? Searching for syndromes by factor analysis of symptoms. JAMA 1997;277:215-22. 29. Gray GC, Kaiser KS, Hawksworth AW, et al. Increased postwar symptoms and psychological morbidity among U.S. Navy Gulf War veterans. Am J Trap Med Hyg 1999;60:758-66. 30. Gray GC, Knoke JD, Berg SW, et al. Counterpoint: Responding to suppositions and misunderstandings. Am J Epidemiol 1998; 148:328-33. 31. Gray GC, Smith TC, Kang HK, et al. Are Gulf War veterans suffering war-related illnesses? Federal and civilian hospitalizations examined, June 1991 to December 1994. Am J Epidemiol (in press).