Validating a method that deals with missing drug information in the Saskatchewan Drug Plan database

Abstract

Introduction: An important limitation of Saskatchewan Drug Plan is the incomplete prescription data during an 18-month period (1987-1988), referred to here as the "black-hole". Objective: To assess the impact of assuming drug non-exposure during the "black-hole" on measures of effect. Methods: We used data from a matched case-control study carried out to assess the association between warfarin use and risk of prostate cancer. All subjects diagnosed with prostate cancer between 1981 and 2002 were matched to six controls. In order to avoid the "black-hole", we included subjects whose cancer was diagnosed after 1994. Conditional logistic regression was used to calculate adjusted incidence rates and 95% confidence intervals (CIs) of prostate cancer in relation to warfarin. Two analyses were carried out: (a) without a "black-hole" and (b) with a random "black-hole" of 18 months imposed in the drug history, during which time subjects were assumed to be unexposed. Results: Compared to non-use, the OR of prostate cancer for ever-use of warfarin in the preceding 5 years was 0.91 (95%CI: 0.81-1.02) (without "black-hole") and 0.89 (95%CI: 0.79-1.01) (with "black-hole"). Compared to non-use, cumulative use of 1, 2, 3, and 4 years were associated with ORs of 1.02, 0.94, 0.77, and 0.76, respectively in the analyses without a "black-hole", and 1.02, 0.88, 0.75, and 0.76, respectively in the analyses with a "black-hole" imposed. Conclusion: When using Saskatchewan Drug Plan data, assuming non-exposure to warfarin during the "black-hole" has a minor effect on the measures of association. Copyright © 2008 John Wiley & Sons, Ltd.