Testicular cancer surveillance

The possible relationship between military deployment and the subsequent increase in cancer rates has been prominent since the Vietnam War. The objective of this study was to investigate whether any form of cancer was increased among veterans deployed to the Persian Gulf in the 1991 conflict. This study matched data from central cancer registries in the District of Columbia and New Jersey with the records for 1.4 million Gulf War era veterans, i.e., 621,902 veterans who arrived in the Persian Gulf before March 1, 1991, and 746,248 non-Gulf veterans. Using a proportional incidence ratio, testicular cancer was found to be the only significantly increased malignancy among deployed Persian Gulf War veterans. The increase became apparent 2 to 3 years after the Persian Gulf War and peaked 4 to 5 years afterward. Our data and those of investigators studying Vietnam veterans suggest that testicular cancer may be related to military deployment.

Introduction

Interest in the possible effects of deployment on subsequent increases in cancer rates has been prominent since the Vietnam War. In 1994, after conflicting studies were evaluated, the Institute of Medicine published a report linking certain forms of cancer to Agent Orange exposure.1 In 1998, the decision was made to add Hodgkin's disease to the 1994 list, which already considered non-Hodgkin's lymphoma, chloracne, and soft tissue sarcoma as having sufficient evidence to indicate a positive association with Agent Orange exposure.2 In addition, the Institute of Medicine noted some suggestive evidence for links with respiratory cancer, prostate cancer, and myeloma.2 The more recent development of the North American Association of Central Cancer Registries, which has facilitated the use of standardized procedures in cancer registries in each of the 50 states and the District of Columbia,3 has allowed the possible determination of national cancer rates in various study groups. The impetus from the Presidential Advisory Committee on Gulf War Veterans' Illnesses4 to develop long-term studies investigating cancer rates among Persian Gulf War veterans and the success of the acquired immunodeficiency syndrome (AIDS)-cancer records matching program in investigating cancer patterns among AIDS patients5,6 led us to investigate whether cancer patterns differed among those who were deployed compared to those who were not.

Methods

Files obtained from the Defense Manpower Data Center included data for 697,000 veterans who arrived in the Persian Gulf during the study period (Aug. 2, 1990 to March 1, 1991) and 746,248 non-Gulf region veteran control subjects. The former group represents the entire population deployed, whereas the latter consists of a stratified random sample of all military personnel who served during the conflict but were not deployed to the Gulf region. Both groups included personnel on active duty, in the reserves, and in the National Guard. The details of selection for these two study groups and their demographic and military characteristics are described in detail elsewhere.7,8 The database contains names, Social Security numbers, demographic data, and military service information for the 1.4 million veterans. The study group was selected to exclude individuals who arrived in the Persian Gulf after the end of hostilities to allow a focus on those with potentially greater exposure to harmful environmental agents. In the control group, equal numbers of subjects were selected from units that were activated and not sent to the Persian Gulf and units that were not activated at all. Race/ethnicity, gender, and age distributions for the deployed and nondeployed subjects are shown in Table I.

The matching procedure involved the central cancer registries of New Jersey and the District of Columbia. The District of Columbia Cancer Registry was chosen because of proximity to the study investigators, and the New Jersey State Cancer Registry was chosen because it was the closest registry with experience with the AIDS-Cancer Match.5,6

Identification of military personnel who received a diagnosis of cancer between 1991 and 1999 was accomplished through record linkage of the veterans database with files supplied by the central cancer registries. Before participation, each central cancer registry conducted ethical reviews. In addition, Institutional Review Board approval was obtained from the George Washington University and the State of New Jersey. For confidentiality reasons, the matching procedures were performed at the registry offices, either by registry personnel or under their supervision. All personal identifiers were deleted from the matched records before the records were removed from the registry offices for data analysis.

The New Jersey registry matching was performed with matching software (Automatch; Match Ware Technologies, Burtonsville, Maryland) used by that registry. Automatch9-11 implements a probabilistic matching algorithm that weighs the likelihood of subject identity on the basis of identical or nearidentical information in the veterans and cancer registries. The matching variables included Social security number, name, race, gender, and dates of birth and death.

The District of Columbia registry did not have available matching software, such as Automatch. Instead, our George Washington University-based team wrote a SAS program (SAS Institute, Gary, North Carolina) to identify potential matching records on the basis of the same fields as the New Jersey registry matching. Social security numbers were available for all records in the veterans database but not all cancer registry records. For the first pass, exact Social security number matches were a necessary but not sufficient criterion for judging records as potential matches. A second match pass was conducted for registry records with missing Social security numbers. There an exact match of birth dates was necessary but not sufficient for judgment as a potential match. Data from each potential match were viewed and discussed by at least four individuals (including cancer registry personnel) to classify the match quality as "exact," "uncertain," or "unlikely." In all cases, a consensus on the match quality was obtained without difficulty.

For each of the linked records, the type of cancer was determined on the basis of International Classification of Diseases for Oncology, second Edition, primary site codes and histology codes for Hodgkin's and non-Hodgkin's lymphoma. The number of cancers at each site, relative to the total number of cancers (or the total number of cancers among male patients, if the cancer was gender specific), and the histology codes for the lymphomas were examined descriptively to determine whether any particular form of cancer was occurring in higher numbers in the Gulf-deployed group compared with the nondeployed group. For each cancer type meeting this criterion, a crude proportional incidence ratio (PIR; referred to as proportionate mortality rate or proportional mortality ratio in mortality studies) and 95% confidence interval (CI) were calculated for the proportional incidence of that specific cancer type among all cancers in the Gulf-deployed group compared with the proportional incidence in the nondeployed group.

To control for potential confounding, a logistic regression analysis was conducted to predict the specific cancer (against all other cancers) from deployment status and central registry and demographic variables. In addition, age-specific PIRs were calculated by dividing, within each age group, the observed number of cases of the specific cancer by the expected number. The expected number was based on the relative proportion of that specific cancer found in the Surveillance, Epidemiology, and End Results (SEER) incidence database for the years 1991-1995. Data were extracted with SEER Stat 4.0 software.12 An overall standardized incidence ratio was also calculated.

Results