Covariate adjustment increases statistical power in randomized controlled trials
Article Outline
To the Editor:
In the January issue of this journal, Peter Austin et al. [1] presented a systematic review on how baseline covariates are handled in randomized controlled trials (RCTs) published in four leading medical journals. In this well-performed study, the authors reviewed 114 RCTs published in the BMJ, the JAMA, The Lancet, and the NEJM, between January 1, 2007 and June 30, 2007.
One of their findings was that 34% of the RCTs presented a covariate-adjusted estimate of the treatment effect, whereas 66% presented an unadjusted estimate. The authors conclude that there is need for a debate about the relative merits of adjusted vs. unadjusted treatment effects. They include a discussion about the arguments pro and contra covariate adjustment. However, in this discussion, the authors disregard a major issue: the beneficial effect of covariate adjustment on statistical power.
Many studies have shown that covariate adjustment of a treatment effect results in increased statistical power [2], [3]. This gain in statistical power is not because of correction for (statistically significant) imbalances in baseline covariates between the randomized groups. More important is the magnitude of the prognostic effect of the covariates [4], [5]. When the imbalance is marginal, but the prognostic effect of a covariate is strong, covariate adjustment will nevertheless result in a substantial power increase. The relation between imbalances, prognostic strength of covariates, and the estimation of the treatment effect was illustrated before for the GUSTO-I trial [4].
In our work within the IMPACT1 project on the design and analysis of RCTs in traumatic brain injury, we found that application of covariate adjustment using seven strong predictors resulted in a 25% reduction of required sample size. In other words, only 75% of the patients would be needed to have the same power to detect a particular treatment effect [6], [7].
In conclusion, the review of Austin et al. nicely shows the current practice of handling of baseline covariates in RCTs and the room for improvement. But there is an additional argument in favor of covariate-adjusted analysis of the treatment effect: the opportunity to substantially increase statistical power at no additional costs.
References
- . A substantial and confusing variation exists in handling of baseline covariates in randomized controlled trials: a review of trials published in leading medical journals. J Clin Epidemiol. 2010;63:142–153
- . Should we adjust for covariates in nonlinear regression analyses of randomized trials?. Control Clin Trials. 1998;19:249–256
- . Covariate adjustment in randomized controlled trials with dichotomous outcomes increases statistical power and reduces sample size requirements. J Clin Epidemiol. 2004;57:454–460
- . Clinical trials in acute myocardial infarction: should we adjust for baseline characteristics?. Am Heart J. 2000;139:745–751
- . Subgroup analysis, covariate adjustment and baseline comparisons in clinical trial reporting: current practice and problems. Stat Med. 2002;21:2917–2930
- Adjustment for strong predictors of outcome in traumatic brain injury trials: 25% reduction in sample size requirements in the IMPACT study. J Neurotrauma. 2006;23:1295–1303
- Baseline characteristics and statistical power in randomized controlled trials: selection, prognostic targeting, or covariate adjustment?. Crit Care Med. 2009;37:2683–2690
- 1 IMPACT: International Mission on Prognosis and Analysis of Clinical Trials in Traumatic Brain Injury.
Supported by NIH grant NS-042691.
PII: S0895-4356(10)00188-5
doi:10.1016/j.jclinepi.2010.05.003
© 2010 Elsevier Inc. All rights reserved.
