Abstract:
|
Missing data are potential source of bias in clinical trials. The drug regulatory agencies often recommend to adopt missing-not-at-random (MNAR) mechanism for handling missing data. One approach for handling missing data according to MNAR mechanism is selection model which postulates a missingness mechanism. The usual approach is then to generate missing observations from predictive distribution of missing data. Alternatively, especially when interest is to compare two means, an inverse probability weighting (IPW) estimator can be directly obtained adjusting for the bias due to missingness without imputing for missing data. In this current work, the scope of IPW estimator approach has been extended to compare two binary proportions. The comparative performance of IPW estimator was evaluated through simulation and the method was applied to clinical trial data comparing two proportions. The variability of IPW estimate was studied via bootstrapping (for a given missingness model) and sensitivity analysis (for the variability in the missingness model).
|