UPDATED: FDA revises and expands draft guidance on adjusting for covariates in RCTs
The FDA on Thursday revised and expanded a 2019 draft guidance that spells out how to adjust for covariates in the statistical analysis of randomized controlled trials (RCTs).
Building on the ICH’s E9 guideline on the statistical principles for clinical trials, the 3-page draft has been transformed into an 8-page draft, with more detailed recommendations on linear and nonlinear models to analyze the efficacy endpoints in RCTs.
Baseline covariates are the demographic factors, disease characteristics, or other information collected from participants before they are randomized in a trial, FDA explains. “Covariate adjustment refers to the use of baseline covariate measurements for estimating and testing treatment effects between randomized groups,” the guidance notes.
Such adjustments will generally reduce the variability of estimating treatment effects and lead to narrower confidence intervals and more powerful hypothesis testing.
While linear models are regarded by FDA as generally providing “reliable estimation and inference for the average treatment effect,” additional issues should be considered before sponsors use nonlinear models, the guidance says.
“Sponsors should discuss with the relevant review divisions specific proposals in a protocol or statistical analysis plan containing nonlinear regression to estimate conditional treatment effects for the primary analysis,” FDA says. “When estimating a conditional treatment effect through nonlinear regression, the model will generally not be exactly correct, and results can be difficult to interpret if the model is misspecified and treatment effects substantially differ across subgroups.”
The revised draft also features an example of six steps for one statistically reliable method of covariate adjustment for an unconditional treatment effect with binary outcomes that produce a resulting estimator.
Frank Harrell, biostatistics professor at Vanderbilt’s medical school, told Endpoints News via email, “Overall it’s really excellent and there is nothing there to prevent modern flexible statistical models from being used in this context. Several modern methods are fostered by this guidance.
“The only reservation I have is its emphasis on causal inference methods that are not really needed when the design is already causal (especially when intent-to-treat is respected). To be fair, those methods may be more robust if you are really bad at pre-specifying the covariate adjustment. Causal methods estimate a quantity that depends on the inclusion/exclusion criteria and the achieved mix of patients, so does not transport to patient populations that don’t have the same covariate distribution as what’s in the trial,” he said.