Abstract:
|
Massive longitudinal healthcare databases enable development of surveillance solutions to identify and evaluate drug risk at unprecedented scale. Recent comparative drug safety analyses using administrative claims data continue to rely on unadjusted incidence rate ratios. We develop a large-scale regularized regression framework to control for drug exposure-assignment and estimate adjusted incidence rate ratios at scale. Our framework uses advancing computing technology for Big Data to fit statistical models involving 1,000,000s of patients. In our framework, we include all clinical information available about patients up to their time of indication diagnosis and treatment exposure, such as all possible drug prescriptions, medical conditions, procedures and other demographics. The number of covariates stands in the 10,000s, regularization helps us avoid overfitting and algorithmic optimization provides estimates in real-time. We apply our method to examine incidence rates of in-patient gastrointestinal bleeding among atrial fibrillation patients taking dabigatran or warfarin in a database that covers over 227M patient-years.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.