This Guide to Statistics and Methods provides an overview of the use of adjustment for baseline characteristics in the analysis of randomized clinical trials and emphasizes several important considerations.
The purpose of randomization in clinical trials is to ensure that there are no systematic differences between treatment groups with respect to measured and unmeasured baseline characteristics that could influence the outcome of interest.1 In a randomized clinical trial (RCT) without selection or information bias, an unadjusted analysis (ie, an analysis that does not take baseline characteristics into account) will provide unbiased estimates of treatment effects.
However, adjusting for baseline characteristics in the analysis of RCTs is advised by both the European Medicines Agency and the US Food and Drug Administration because it may improve statistical efficiency, enhancing the ability to draw a reliable conclusion from the available data.2,3 Nevertheless, investigators may be unaware of the benefits of adjusting for baseline characteristics or may misinterpret the purpose of adjustment as a correction for chance imbalances between groups to obtain valid results.4
In an RCT, Zampieri et al5 investigated the effect of fluid therapy with a balanced solution compared with a 0.9% sodium chloride solution on 90-day survival among critically ill patients across 75 intensive care units in Brazil. In their primary analysis, the investigators adjusted for the enrollment site and specific baseline patient characteristics.
USE OF BASELINE ADJUSTMENT IN THE ANALYSIS OF RCTS
Description of Adjustment for Baseline Patient Characteristics in RCTs
When considering adjusting for baseline characteristics in the analysis of an RCT, the investigator team must decide when to adjust, which variables to adjust for, what statistical method to use, how to handle missing data, and what to report.2,3 Adjustment for baseline variables should generally be considered when stratified randomization is used or when there is a known or anticipated strong association between baseline characteristics and the primary outcome (eg, strong prognostic factors).2,3
Variables included in the adjustment should be selected prospectively, should not be plausibly affected by the treatment (ie, they should generally be characteristics measured prior to randomization), and should be prespecified in the trial protocol. Improved statistical efficiency is achieved only if the included baseline characteristics are strong prognostic characteristics (ie, they are strongly associated with the outcome). This often occurs when the end point is a later measurement of a characteristic measured at baseline.6
There is no formal rule for specifying the number of baseline variables that should be adjusted for in the statistical analysis, but the model should generally include a limited number of variables relative to the sample size.2,3 Including too many variables may lead to complex and unstable statistical models.
Adjustment is typically achieved by including the variables of interest in a regression model, ...