Bias Adjustment Methods for Analysis of a Non-randomized Controlled Trials of Right Heart Catheterization for Patients in ICU

Abstract

Kaplan-Meier estimate or proportional hazards regression is commonly used directly to estimate the effect of treatment on survival time in randomized clinical studies. However, such methods usually lead to biased estimate of treatment effect in non-randomized or observational studies because the treated and untreated groups cannot be compared directly due to potential systematical difference in baseline characteristics. Researchers have developed various methods for adjusting biased estimates by balancing out confounding covariates such as matching or stratification on propensity score, inverse probability treatment weighting. However, very few studies have compared the performance of these methods. In this paper, we conducted an intensive case study to compare the performance of various bias correction methods for non-randomized studies and applied these methods to the right-heart catheterization (RHC) study to investigate the impact of RHC on the survival time of critically ill patients in the intensive care unit. Our findings suggest that, after bias adjustment procedures, RHC was associated with increased mortality. The inverse probability treatment weighting outperforms other bias adjustment methods in terms of bias, mean-squared error of the hazard ratio estimators, type I error and power. In general, a combination of these bias adjustment methods could be applied to make the estimation of the treatment effect more efficient.