Adjusting for Selection Bias in Single-Blinded Randomized Controlled Clinical Trials

• Activity Number: 518
• Type: Topic Contributed
• Date/Time: Wednesday, August 7, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Lieven Kennes*+
• Companies: RWTH Aachen University
• Keywords: Selection Bias ; Randomized controlled clinical trial ; Blinding ; Maximum Likelihood

Abstract

Selection bias affects the evaluation of clinical trials. In absence of double-blinding, the investigator is able to predict future allocation based on past assignments yielding the possibility to cause inhomogeneous treatment groups. To quantify the impact of selection bias on a single-blinded randomized controlled clinical trial, type one error rate elevation is used. Based on the principle of maximum likelihood (ML), a test adjusting for selection bias as well as confidence intervals for the true treatment difference and the selection effect are derived. The established ML-estimators are consistent and asymptotic normal. Formulas for confidence intervals for all parameters are given. The confidence interval for the selection effect can be used to detect selection bias. Results of simulation studies illustrate that while the t-test exceeds the nominal significance level in presence of selection bias, the ML-test corrects for selection bias and asymptotically holds the nominal significance level.