This study will develop improved statistical methods for two aspects of clinical trials analysis: quality-of-life (QOL) and treatment compliance. Researchers increasingly consider QOL data in evaluating new therapies, especially in trials for cancer and other chronic diseases. QOL data contribute most when combined with clinical information on survival, since decisions about therapy should address both factors. Unfortunately, missing data problems plague this area of research, complicating statistical analysis. QOL data are often nonignorably missing, since subjects who miss assessments tend to have poor QOL or health status; this makes standard statistical analyses at best difficult and at worst biased and uninterpretable. The proposed work, will (i) analyze QOL in a time-to-event framework, extending survival analysis methods in this unique setting to jointly analyze QOL and clinical outcomes; and (ii) develop diagnostic techniques to assess and evaluate the extent of the nonignorable missing data problem Treatment compliance in clinical trials is generally imperfect. Compliance data can contribute to analyses of treatment effect. An ongoing debate among both statisticians and clinicians centers on """"""""as- randomized"""""""" (AR, or """"""""intent-to-treat"""""""") vs. """"""""as-treated"""""""" (AT) analyses. The former approach groups subjects according to their randomization, value, regardless of compliance with the treatment regimen, the latter according to actual treatment received. The proposed work will (i) develop an AT approach that allows intermediate levels of compliance, and (ii) extend this approach to longitudinal data.
| Troxel, Andrea B; Esserman, Denise A (2004) Frailty models for quality of life in oncology. J Biopharm Stat 14:145-54 |
