An investigation of bias in a study of nuclear shipyard workers.
Greenberg-ER; Rosner-B; Hennekens-C; Rinsky-R; Colton-T
Am J Epidemiol 1985 Feb; 121(2):301-308
Discrepancies in two studies of cancer mortality among nuclear workers at the Portsmouth Naval Shipyard were investigated. Lists of subjects from a 1978 proportional mortality study and a 1981 NIOSH cohort study were compared. The cohort was composed of workers monitored for radiation exposure. Standardized mortality ratios (SMRs) (compared to the United States population) and age adjusted proportional mortality ratios (PMRs) of workers appearing on both lists for whom cause of death information was available were determined. The effect of selection bias was assessed by comparing cancer PMRs for nuclear and non nuclear workers enrolled in the initial study with those whose next of kin had been contacted. Effects of measurement bias were assessed by examining cancer PMRs for nuclear and non nuclear workers classified correctly and incorrectly by next of kin. For nuclear workers, calculations of SMRs showed a small deficit in cancer mortality (SMR 90) and a relatively large deficit in all cause mortality (SMR 75). For all workers both cancer and all cause mortality SMRs were very close to 1.0. Thus the cancer PMR for nuclear workers was about 20 percent higher because of their low all cause mortality. For all workers identified in the original PMR study, the PMRs for cancer were 0.268 in nuclear and 0.198 in non nuclear workers. For workers whose next of kin had been contacted, PMRs were 0.207 in non nuclear and 0.282 in nuclear workers. Misclassification of exposure status resulting in measurement bias was also found, due to inaccurate information from next of kin regarding work exposure histories of nuclear workers dying from cancer. The authors conclude that discrepancies between these two studies result largely from the healthy worker effect and measurement bias, the latter being the more substantial effect. Bias occurring in the selection of subjects for the initial study accounts for a smaller portion of the discrepancies.
NIOSH-Author; Epidemiology; Quantitative-analysis; Work-performance; Carcinogenesis; Carcinogens; Workers; Nuclear-power-plants; Carcinogenicity; Nuclear-engineering; Workplace-studies; Cancer;
Author Keywords: epidemiologic methods; mortality; neoplasms; radiation effect
Dr. E. Robert Greenberg, Norris Cotton Cancer Center, Hanover, NH 03756
American Journal of Epidemiology