Impact of Sample Size on Prognostic Imbalance and Covariate Adjustment in Randomized Controlled Trials (RCTs)

• Activity Number: 569
• Type: Contributed
• Date/Time: Wednesday, August 3, 2011 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Rong Chu*+ and Stephen Walter and Lehana Thabane and Philip James Devereaux and Gordon Guyatt
• Companies: McMaster University and McMaster University and McMaster University and McMaster University and McMaster University
• Keywords: Randomized Controlled Trials ; Prognostic Imbalance ; Covariate Adjustment ; Simulation Study ; Sample size

Abstract

Randomization does not guarantee balance in all baseline covariates particularly in trials with small sample sizes. Study objectives include: 1) To assess the probability of imbalance in a rare yet strong prognostic factor (PF) between two arms; 2) To investigate the impact of prognostic imbalance on statistical modeling of treatment effect; 3) To examine the effect of sample size in relation to the first two objectives.

We simulated RCTs with a binary outcome by varying the risk of outcome in the control group, effect of treatment, effect of PF, prevalence of PF, and sample size. Logistic regression models with and without adjustment for PF were compared in terms of bias, standard error, coverage of confidence interval and statistical power.

The probability of imbalance increase as sample size decreases. For a PF with prevalence of 0.5, the probability of an imbalance = 5% can reach 0.4 with 125 per arm. Ignoring a strong PF (relative risk = 5) leads to bias in assessing moderate treatment effect; the bias is essentially independent of sample size. Adjusting for PF reduces precision but increases statistical power. Results conditional on 5% imbalance will be presented.