A penalized Cox proportional hazards model with multiple time-varying exposures

Chenkun Wang, Hai Liu, Sujuan Gao

Research output: Contribution to journalArticle

Abstract

In recent pharmacoepidemiology research, the increasing use of electronic medication dispensing data provides an unprecedented opportunity to examine various health outcomes associated with long-term medication usage. Often, patients may take multiple types of medications intended for the same medical condition and the medication exposure status and intensity may vary over time, posing challenges to the statistical modeling of such data. In this article, we propose a penalized Cox proportional hazards (PH) model with multiple functional covariates and potential interaction effects. We also consider constrained coefficient functions to ensure a diminishing medication effect over time. Hypothesis testing of interaction effect and main effect was discussed under the penalized Cox PH model setting. Our simulation studies demonstrate the adequate performance of the proposed methods for both parameter estimation and hypothesis testing. Application to a primary care depression cohort study was also illustrated to examine the effects of two common types of antidepressants on the risk of coronary artery disease.

Original languageEnglish (US)
Pages (from-to)185-201
Number of pages17
JournalAnnals of Applied Statistics
Volume11
Issue number1
DOIs
StatePublished - Mar 2017

    Fingerprint

Keywords

  • Interaction
  • Penalized spline
  • Pharmacoepidemiology
  • Time-varying exposure

ASJC Scopus subject areas

  • Statistics and Probability
  • Modeling and Simulation
  • Statistics, Probability and Uncertainty

Cite this