|
Abstract:
|
Missing data complicates the interpretation of results on an endpoint. We demonstrate the versatility of methods to combine endpoints hierarchically to handle missing data problems. The main advantage of the method is that the reason for missingness goes hand in hand with the imputation method. Thus, for example, the imputation method recognizes that missingness due to an adverse event is worse than missingness due to loss to follow-up. By developing rules that hierarchically order the reasons for missing data, a consistent solution is provided to the problem. The procedure, which can apply for any kind of endpoint, will be described in this session for a continuous outcome and for a recurrent event outcome. Under the conventional approach, data are analyzed by several methods - one of which is considered the primary method (e.g. MMRM) and the other are labeled sensitivity analyses (e.g. pattern mixture). Under the proposed approach a single analysis can suffice.
|
