An investigation of bias in a study of nuclear shipyard workers.

The authors examined discrepant findings between a 1978 proportional mortality study and a 1981 cohort study of workers at the Portsmouth, New Hampshire, Naval Shipyard to determine whether the healthy worker effect, selection bias, or measurement bias could explain why only the proportional mortality study found excess cancer deaths among nuclear workers. Lower mortality from noncancer causes in nuclear workers (the healthy worker effect) partly accounted for the observed elevated cancer proportional mortality. More important, however, was measurement bias which occurred in the proportional mortality study when nuclear workers who had not died of cancer were misclassified as not being nuclear workers based on information from their next of kin, thereby creating a spurious association. Although the proportional mortality study was based on a small sample of all deaths occurring in the cohort, selection bias did not contribute materially to the discrepant results for total cancer deaths. With regard to leukemia, misclassification of occupation in the proportional mortality study and disagreement about cause of death accounted for some of the reported excess deaths.