St. James Ballroom
Survival Associated with Pharmaceutical Drug Exposures Following Cancer Diagnoses: An Analysis of Pharmacoepidemiologic Evidence and Bias (303909)
*Scott W. Keith, Thomas Jefferson University
Keywords: Cox modeling, proportional hazards, time-varying covariates, time-dependent coefficients, immortal time bias, cancer, pharmacoepidemiology
Objectives: To evaluate survival associated with metformin in head and neck cancer (HNC) patients and ACE inhibitors in pancreatic cancer (PC) patients and to investigate the immortal time bias caused by applying time-fixed methodology to analyze these time-varying exposures. Data: Italian Health Ministries administrative health care database from the Emilia-Romagna Region of Italy from 2003-2012. Methods: All-cause mortality analyzed using both time-fixed and time-varying covariates, respectively, in Cox models, to estimate unbiased associations and, by comparison, immortal time biases. Results: Our modeling revealed that immortal time bias had a strong influence on the associations of interest and that the hazard ratios were time-dependent. Discussion: The magnitude of immortal time bias depended on the lethality of the cancer and prevalence of the exposure. Proper accounting for the timing of exposures by using time-varying covariates, in contexts such as these pharmacoepidemiologic studies, is critically important for avoiding bias. Careful evaluation of the proportional hazards assumption in Cox modeling can reveal the most valuable findings in a study.