Estimating Uncertainty in Weighted Competing Risk Analyzes

• Activity Number: 88 - SPEED: Causal Inference and Related Methodology Part 2
• Type: Contributed
• Date/Time: Sunday, July 28, 2019 : 5:05 PM to 5:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Amber Hackstadt* and Jonathan Chipman and Christianne L. Roumie and Adriana M. Hung and Jea Young Min and Carlos G Grijalva and Marie R Griffin and Robert Greevy
• Companies: Vanderbilt University Medical Center and Vanderbilt University and Vanderbilt University Medical Center, Veteran Administration Tennessee Valley VA Health and Vanderbilt University Medical Center and Vanderbilt University Medical Center and Vanderbilt University Medical Center and Vanderbilt University Medical Center and Vanderbilt University
• Keywords: Competing Risks; Aalen Johansen Estimator; Propensity Score Weighting; Matching Weights; Bootstrap; Greenwood Estimator

Abstract

When estimating cumulative incidence probabilities for a time to event outcome of interest, such as a major adverse cardiovascular event (MACE), competing risks, such as a non-cardiovascular death or a renal event, are often present. Multistate models and Aalen Johansen plots are useful tools for addressing competing risks in time to event analyses. However, the performance of methods to estimate confidence bands for the cumulative incidence probabilities is an open question, particularly when used with existing and modern inverse propensity score weighting methods (IPW). We test the performance of various methods over a range of event rates, samples sizes, and degrees of baseline covariate imbalance, which impact the IPW. This research is motivated by an electronic health record derived study of nearly one-hundred thousand US veterans comparing the effects of the anti-diabetic medications metformin and sulfonylureas after reaching a renal function threshold. Patients reaching the renal threshold have a high risk of experiencing MACE and non-cardiovascular death.