To Be or Not To Be Covariate Adjusted

Stephane Bandeira

The FDA’s Guidance On Adjusting For Covariates In The Statistical Analysis of RCTs

Clinical trials are relevant to demonstrate the safety and performance of Medical Devices, and the design of a clinical study can be [statistical] art on its own. Not only is it essential that the trial will be viable, but there is also often the uncertainty of it being accepted as sufficient by the regulatory authorities.

What does the FDA say about Covariates in Clinical Trials?

The FDA has revised its draft guidance on covariate adjustment in randomized controlled trials (RCTs). An RTC includes statistical adjustments for observed covariates (treatment effect parameter estimates typically acknowledging observed baseline characteristics such as age, weight, or BMI). In many cases, baseline covariates have no impact on treatment response, leading to inflated false-positive rates versus non-adjusted data. In this Guidance, FDA recommends approaches for addressing this problem and provides recommendations by study type (superiority vs. non-inferiority) and primary endpoint type (binary vs. continuous).

This draft guidance is intended to update and clarify statistical methods for incorporating covariates in clinical trials for drugs and biologics subject to FDA regulation. It should not be interpreted as a fundamental departure from current thinking but, instead, as an effort to make some aspects of that thinking more explicit and to provide practical advice on designating important demographic and baseline characteristics while maintaining the statistical validity of the randomized clinical trial.

Traditionally, randomization is assumed to result in an equal distribution of prognostic and non-prognostic covariates; therefore, unadjusted analysis is quite common. However, suppose there is clarity regarding the prognostic covariates. In that case, such covariates should be identified at the baseline, and the mathematical model for considering the covariate should be included in the SAP.

When should sponsors adjust for covariates during drug development?

Covariate adjustments in RCTs can protect against chance imbalances for important baselines covariates lead to narrower confidence intervals, thereby increasing the power of the study. Covariates should be carefully selected to avoid the risk of reducing the precision in a study due to non-prognostic covariates.

However, this guide does not address the use of covariates to control for confounding variables in non-randomized trials or the use of covariate adjustment for analyzing longitudinal repeated measures data.

While this Guidance focuses solely on the question of covariate adjustment in RCTs seeking FDA approval, statisticians involved in the conduct of such RCTs may find this guide helpful when considering issues related to their design and analysis. While considerations specific to RCTs seeking approval from other regulatory bodies may warrant modifications to this analysis, we believe such modifications should be made on a case-by-case basis.

Caution should be observed in trials with small sample size, for instance, rare disease trials with binary or time to event outcomes where stratification can lead to type 1 errors thus, false-positive outcomes of the efficacy. Thus, the number of covariates should be calibrated with the sample size. When stratified randomization is used, factors used in the stratification should also be considered.

Why adjusting for covariates is important?

Being able to account for baseline covariates is an extremely helpful tool for researchers. It can help create a better understanding of any observed results, which can be crucial in developing effective treatment regimens. FDA recommends optimal approaches for adjusting for imbalance in covariate distributions in randomized clinical trials. The main advantage of covariate adjustment is that it increases precision and improves statistical power.

How will this Guidance affect external and internal stakeholders?

This guidance document from the Food and Drug Administration (FDA) was a useful reminder that this is an important area. We expect that the use of covariates in trial design will lead to a more precise estimation of treatment effects and improve trial efficiency. As with any analytical strategy, sponsors must understand when covariate adjustment would be helpful and how to design the best study possible with the available data.

Of course, while this Guidance centered on drug clinical trials, covariate adjustment is also a standard approach to adjusting for baseline differences in observational studies and Medical Device trials. The use of covariates in clinical trials will become a standard, established procedure and is thus expected to continue.

The guide can be found on the FDA page.