ANALYSIS OF REGRESSION DISCONTINUITY DESIGNS USING CENSORED DATA

Area of Science:

• Biostatistics
• Epidemiology
• Econometrics

Background:

• Treatment decisions often rely on covariate thresholds (e.g., prostate-specific antigen levels).
• Estimating causal effects is crucial when randomized trials are unavailable.
• Regression discontinuity (RD) designs offer a method to estimate causal effects in threshold-based scenarios.

Purpose of the Study:

• To develop and evaluate a method for estimating causal effects in regression discontinuity designs with censored data.
• To provide a robust framework for analyzing interventions determined by a continuous covariate threshold.

Main Methods:

• The study proposes an estimation procedure using censoring-unbiased transformations.
• Techniques include inverse probability censored weighting and doubly robust transformation schemes.
• Finite-sample properties are assessed through simulation studies.

Main Results:

• The proposed method effectively estimates causal effects under regression discontinuity with censored data.
• Simulation results demonstrate the estimator's finite-sample performance.
• The method is illustrated using prostate-specific antigen (PSA) dependent screening strategies.

Conclusions:

• The developed method extends regression discontinuity designs to handle censored outcomes.
• This approach enables causal inference for threshold-based treatments in observational studies.
• The findings are applicable to medical and scientific settings where covariate thresholds guide interventions.