Adjustment for baseline measurement error in randomized controlled trials induces bias

Area of Science:

• Biostatistics
• Clinical Trials
• Epidemiology

Background:

• Continuous outcomes in randomized controlled trials (RCTs) are often measured at baseline.
• Baseline measurements can be subject to significant measurement error (e.g., within-patient variability).
• The necessity of adjusting for baseline measurement error in RCTs is a debated topic in statistical literature.

Purpose of the Study:

• To compare the bias in treatment effect estimation with and without adjusting for baseline measurement error in RCTs.
• To evaluate the impact of varying levels of baseline imbalance on bias.
• To assess the influence of sample size and measurement error reliability on bias.

Main Methods:

• Computer simulations were employed to model RCT scenarios.
• Simulations compared ordinary least squares (OLS) estimators with and without measurement error adjustment.
• Key parameters varied included sample size (30 vs. 300 per group) and reliability coefficient (0.6, 0.8, 1).

Main Results:

• The OLS estimator, without adjusting for measurement error, demonstrated unbiasedness in RCTs, irrespective of baseline imbalance.
• Adjusting for measurement error introduced bias, particularly pronounced with smaller sample sizes and higher measurement error.
• Overestimation of treatment effect occurred when the control group's baseline mean exceeded the treated group's; underestimation occurred otherwise.

Conclusions:

• Adjusting for baseline measurement error is not recommended in RCTs as it can introduce bias.
• The standard OLS approach without adjustment provides an unbiased estimate of the treatment effect.
• Careful consideration of measurement error properties is crucial when analyzing RCT data.