Abstract:
|
Count and recurrent event endpoints are common key efficacy endpoints in clinical research. Count data are often analyzed by generalized linear models with a log link. Recent FDA draft guidance (2021) for randomized control trials specifically discusses how covariate adjustment in nonlinear models yields a conditional, or subgroup-specific, treatment effect which may differ from the marginal treatment effect. Alignment of the target estimand and estimation approach is a critical component of the estimand framework outlined in the ICH E9 Addendum. In this presentation, we review the impact of covariate adjustment on treatment effect estimation of count and recurrent event endpoints, motivated by pulmonary exacerbation events in cystic fibrosis. We demonstrate that the within-group conditional event rates estimated at the mean of the covariate are always less than or equal to the within-group marginal event rates. We also evaluate the collapsibility of the rate ratio and when the conditional and marginal rate ratios are equivalent. Finally, we discuss clinical interpretations of conditional and marginal rate ratios under the estimand framework.
|