PRO data have now become an integral part of clinical trials to evaluate the efficacy of treatment for severe and terminal illness. However, it is often fraught with missing data due to illness, death and early termination of a trial (time to event study or superior efficacy). OBJECTIVES: To evaluate different methods of data imputation when trials are terminated early and there is missing data due to deteriorating health. METHODS: Using data from a large cancer (multiple myeloma) trial which included both missing data due to illness and early termination due to study termination or non-illness-related reasons, several statistical methods were evaluated. A total of 598 subjects completed the EORTC QLQ-C30 at least one post-baseline timepoint and were available for analysis. After setting PRO scores of all subjects who died to the worst possible scores, remaining missing data were imputed and analyzed by either multiple imputation (M=4) using a generalized estimating equation technique, Sun and Song method for censored data, and Pattern-Mixture models. Global Health was the primary PRO endpoint with all other scales adjusted for multiplicity using the Hochberg-Benjamini method. RESULTS: All methods found similar results, although the multiple imputation method found the most number of scales/symptoms significant (N=10). Sun and Song and the Pattern-Mixture model each found the same four scales/symptoms significantly different between groups – Global Health, Cognitive Functioning, Emotional Functioning and Dyspnea. CONCLUSIONS: All methods are useful approaches to handle missing data imputation and analysis. The multiple imputation method appears to be less conservative, finding ten significant differences versus four with the other methods. The Sun and Song approach provides an insight into what the treatment differences would have been had all the subjects stayed in the study. The Pattern-Mixture model was the most complex method and did not provide any additional information over the other methods.