Skip to main content

Using an alternative data source to examine randomization in the Canadian National Breast Screening Study

Cohen MM, Kaufert PA, MacWilliam L, Tate RB. Clin Epidemiol. 1996; 49(9):1039-44.

Criticisms of the Canadian National Breast Screening Study (NBSS) assert that reported findings (more deaths in women aged 40-49 allocated to mammography versus no mammography and no difference in mortality between women who had mammography and physical examination versus physical examination alone for those aged 50-59) may be due to more women with prior breast disease being allocated to the mammography arms of the study. The possibility that allocation was not random was examined for NBSS participants in Manitoba, Canada, using health insurance data that were external to and independent of the NBSS. The study design consisted of a retrospective observation study using health insurance (claims) data to construct health histories of breast disease prior to NBSS entry. Self-reported breast disease from the NBSS entry questionnaires was compared to breast disease histories on the basis of health insurance claims. The setting consisted of one NBSS screening center in Winnipeg, Canada. The patient population consisted of 9477 women with at least one health insurance claim in the 24 months prior to NBSS entry. We determined the proportions of women within each study arm who had claims for breast disease, breast disease investigation, or xeromammograms in the 24 months prior to NBSS entry, and compared the proportion of women's self-reports of breast disease or xeromammography with the presence/absence of a claim prior to NBSS entry for breast disease or xeromammography. No significant differences in the proportion of women with prior histories of benign breast disease, investigation, or xeromammograms (p > 0.05) were found across the study arms. Nine women in the mammography group versus one in the no mammography group had one prior health insurance claim for breast cancer, but eight of these women had no subsequent claims for breast cancer. There were no differences across the study arms in the proportion of women who reported a prior history of breast disease or a prior xeromammogram for whom an insurance claim was found. Using data external to the NBSS for Manitoba participants, the study found no definitive evidence to support a nonrandom allocation of women with prior breast disease to the mammography arms of the study. However, generalizability to the other NBSS centers cannot be assured.

Keywords: Breast cancer Data collection