CAUSAL PROPORTIONAL HAZARDS ESTIMATION WITH A BINARY INSTRUMENTAL VARIABLE

• Type of Material: Article
• Authors: Behzad Kianian, Emory University; Jung In Kim, University of North Carolina; Jason P Fine, University of North Carolina; Limin Peng, Emory University
• Language: English
• Date: 2021-01-01
• Publisher: STATISTICA SINICA
• Publication Version: Final Published Version
• Copyright Statement: © 2021 Institute of Statistical Science, Academia Sinica
• Final Published Version (URL): http://dx.doi.org/10.5705/ss.202019.0096
• Title of Journal or Parent Work: STATISTICA SINICA
• Volume: 31
• Issue: 2
• Start Page: 673
• End Page: 699
• Grant/Funding Information: The authors gratefully acknowledge the support from the National Institutes of Health grant R01 HL113548.
• Keywords: MODEL; Mathematics; Causal treatment effect; instrumental variable; Science & Technology; NONCOMPLIANCE; Cox proportional hazards model; Statistics & Probability; RANDOMIZED-TRIAL; Physical Sciences; DESIGN; IDENTIFICATION; TIME-TO-EVENT; noncompliance

Abstract

Instrumental variables (IVs) are useful for estimating causal effects in the presence of unmeasured confounding. IV methods are well developed for un-censored outcomes, particularly for structural linear equation models, where simple two-stage estimation schemes are available. Extending these methods to survival settings is challenging, partly because of the nonlinearity of the popular survival regression models, and partly because of the complexity of right censoring and other survival features. Motivated by the Prostate, Lung, Colorectal and Ovarian (PLCO) Cancer Screening Trial, we develop a simple causal hazard ratio estimator in a proportional hazards model with right-censored data. The method exploits a special characterization of IVs that enables the use of an intuitive inverse weighting scheme that is generally applicable to more complex survival settings with left truncation, competing risks, or recurrent events. We rigorously establish the asymptotic properties of the estimators, and provide plug-in variance estimators. The proposed method can be implemented in standard software, and is evaluated through extensive simulation studies. We apply the proposed IV method to a data set from the PLCO Cancer Screening Trial to identify the causal effect of flexible sigmoidoscopy screening on colorectal cancer survival, which may be confounded by informative noncompliance with the assigned screening regimen.