PURPOSE: This workshop will discuss the importance of missing patient-reported outcome (PRO) data in oncology clinical trials; describe why implementation of methods that make correct assumptions is important; present novel methods to identify heterogeneous groups of treatment responders informed by missing PRO data patterns and their link with survival outcomes; and present examples using these methods. The anticipated audience includes those familiar with PRO analyses who want to learn and incorporate methods that show more accurate and compelling results.
DESCRIPTION: Although much attention has been paid to the development and administration of PRO questionnaires, largely unrecognized is the explicit analysis of missing data and their impact on clinical trials. Pattern mixture models (PMMs) are recommended as a sensitivity analysis when one suspects that data are not missing at random. These models build in covariates to account for differential missing data patterns and allow the derivation of an adjusted mean effect. This workshop presents an extension to PMMs (ePMM) within a latent variable framework: heterogeneity is explicitly examined at the patient level in terms of intercepts and changes over time, while simultaneously building in the effect of missing data patterns. Data-driven subgroups can be identified that allow analysts to determine whether missing data leads to different PRO results. Once identified, survival can be compared between the subgroups. Identifying subgroups who are more, or less, likely to complete oncology clinical trials and assessing the relationship with survival can play a key role in study design and the development of personalized medicine. We will present models, show the effects on survival outcomes, and review randomized clinical trial examples. Following the presentation, the audience will be encouraged to discuss the pros and cons of these methods compared to currently used approaches in the design and analysis of oncology clinical trials.
