Comparing Immortal Time Biases in Pharmacoepidemiologic Survival Analyses of Antihypertensives After Pancreatic Cancer Versus Antidiabetics After Head and Neck Cancer

• Activity Number: 468
• Type: Contributed
• Date/Time: Wednesday, August 6, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Scott Keith*+ and Thomas Karagiannis and Vittorio Maio and Daniel Louis and Carol Rabinowitz and Mengdan Liu
• Companies: Thomas Jefferson University and Thomas Jefferson University and Thomas Jefferson University and Thomas Jefferson University and Thomas Jefferson University and Thomas Jefferson University
• Keywords: immortal time bias ; time-varying covariates ; time-dependent covariates ; pancreatic cancer ; head and neck cancer ; pharmacoepidemiology

Abstract

Objective: To evaluate the biases caused by applying time-fixed methodology to analyze mortality associations of time-varying exposures to antihypertensive versus oral antidiabetic drugs in patients hospitalized with a head and neck cancer (HNC) versus a pancreatic cancer (PC) diagnosis, respectively, while residing in the Emilia-Romagna Region of Italy from 2003-2012. Methods: Time-fixed and time-varying covariates were respectively constructed for the two drug class exposures and used to predict all-cause mortality following HNC or PC diagnosis, respectively, by Cox proportional hazards modeling. Parameter estimates from these two modeling strategies are used to identify the respective immortal time biases. Results: Preliminary analyses show that, by comparing time-fixed and time-varying drug exposures models, immortal time bias has a strong influence on mortality estimates associated with antidiabetics after HNC, but not on those of antihypertensives following PC. Conclusions: The magnitudes of immortal time biases appear to be cancer site-dependent. It is unclear what properties of these cancers, drug classes, or patients might account for disparities in immortal time biases.