A cohort study of mammography screening finds that comorbidity measures are insufficient for controlling selection bias

Abstract

ObjectiveTo examine the potential of claims-based comorbidity measures for controlling selection bias in observational studies of mammography screening. Study Design and SettingBased on claims data of a large German Statutory Health Insurance fund, the single comorbidities considered by the Charlson, Elixhauser, Multipurpose Australian Comorbidity Scoring System, and M3 comorbidity measures were identified for mammography screening participants and nonparticipants. Total death rates within 4 years after screening invitation were compared. Cox proportional hazards regressions were performed unadjusted and adjusted for age, federal state of residence, and comorbidity. ResultsAmong 1,247,919 insured women aged 50–68 years (56.2% participants), 10,311 participants (death rate 375.8/100,000 person-years) and 18,113 nonparticipants (death rate 854.8/100,000 person-years) died from any cause during the follow-up. The unadjusted hazard ratio (HR) for death from any cause for participants vs. nonparticipants was 0.44 (99.9% confidence interval 0.42–0.46). Adjustments attenuated the HR to a maximum of 0.52 (0.50–0.54). ConclusionThe lower short-term all-cause mortality among participants cannot be explained by mammography screening effects and should be interpreted as selection bias. Adjusting for comorbidities only slightly attenuated this bias. Future studies should examine whether claims data include further information that is beneficial to adequately control selection bias in observational studies of mammography screening.