Collections > Electronic Theses and Dissertations > Breast Cancer Risk after Metformin Initiation in Older Women: the Role of Study Design, Potential Confounding by Body Mass Index, and Differential Detection
pdf

Several observational studies reported that metformin may reduce breast cancer risk; however, many of these studies were affected by time-related biases. Additionally, confounding by unmeasured body mass index (BMI) and differential detection for breast cancer have not been examined in metformin-breast cancer studies. The dissertation aimed to examine the relative risk of breast cancer for older women initiating metformin versus sulfonylureas, avoiding time-related bias and accounting for potential bias due to unmeasured confounding and differential screening mammography. Using 2007-2011 US Medicare claims data, we identified cohorts of cancer-free women aged 65+ who initiated monotherapy with metformin or sulfonylureas. Hazard ratios of breast cancer were estimated comparing metformin to sulfonylureas initiators, using weighted Cox models. Unmeasured confounding by BMI and smoking was adjusted by propensity score calibration using external information from Medicare Current Beneficiary Survey 2006-2009 panels. Among new users of Medicare claims, we compared the risks of screening mammograms and screen-detected breast cancer in 12 months pre- and post-initiation between metformin and sulfonylureas initiators. Metformin initiators did not have reduced risks of breast cancer compared with sulfonylureas initiators (Hazard Ratio: 1.08; 95% Confidence Interval: 0.81 to 1.44). Externally controlling for BMI and smoking did not affect the estimate, indicating a little independent effect of BMI and smoking on metformin relative to sulfonylureas prescribing. Metformin initiators were not only more frequently screened for breast cancer than sulfonylureas initiators, but they also had higher probabilities of screen-detected breast cancer both in 12 months before and after initiation. The results indicate possible detection bias due to differential screening mammography, but the absolute difference in screen-detected breast cancer is too small to explain observing no metformin-breast cancer association assuming a real protective effect of metformin. This study provides no support for reduced risks of breast cancer after initiation of metformin compared with a clinical alternative, sulfonylureas, in older women. Our findings support the notion that reduced breast cancer risks in metformin users observed in previous studies is likely due to time-related biases, and emphasize the importance of conducting observational studies with rigorous, state-of-the art design to avoid observing spurious effects or missing real ones.