An adaptive imputation method of missing data for sparsely retrieved dropouts in treatment policy strategy

Abstract

The ICH E9 R1 Addendum suggests using a treatment-policy strategy as an approach to handle intercurrent events for estimating de facto estimand. Under this strategy, regardless of the occurrence of intercurrent events, the value for the variable of interest is analyzed. After discontinuing treatment, participants who remain in the trial to complete the assessment of the primary endpoints are referred to as retrieved dropouts, while early withdrawal by participants results in missing data. To mitigate the effects of missing data, strategies like mixed model for repeated measures or the retrieved dropout multiple imputation method are used. The bias of retrieved dropout methods is relatively small. However, if retrieved dropouts are scarce, it could significantly inflate variance.

This article introduces an innovative adaptive model that refines the On/Off Intercepts with Common Slopes using Residuals (RD_OICSR) model, which is a model within the retrieved dropout methods, and evaluates it using simulated data from a depression trial. The findings indicate that when the proportion of retrieved dropouts falls below the predetermined threshold set by researchers, our method minimizes unrealistic variance inflation by incorporating data from placebo completers. Conversely, the model adaptively matches the RD_OICSR model. This ensures that irrespective of the retrieved dropouts' proportions, the analysis remains accurate.