OBJECTIVES: Missing PRO data can introduce bias and interfere with the ability to evaluate treatment effects.  Approaches to handing missing PRO data during data analysis should be pre-specified in the statistical analysis plan.  This study sought to guide sponsors by critically demonstrating different approaches to handling missing PRO data where entire measurements are missing. METHODS: Four (4) approaches to handling missing PRO data were evaluated: full analysis dataset, complete case analysis, last observation carried forward (LOCF) and pattern mixture model.  Analysis was conducted on data that comprised a dummy data set designed to represent a 12 week clinical trial data set comparing fictional treatments A and B, with PRO data based on the EORTC QLQ-C30.  The resulting four imputed datasets were analysed using mixed model for repeated measures (MMRM).  Results were presented in tabular and graphic format, and were compared with the full analysis dataset to evaluate their performance. RESULTS: Analysis performed under the assumption of missing at random (MAR) produced similar results to the complete case analysis.  Analysis performed under the assumption of missing not at random (MNAR) produced notably different results.  The pattern mixture model provided a degree of confidence around the complete case analysis that appeared related to the extent of missing data i.e. the more missing data, the greater the uncertainty.  The LOCF approach was the least robust with the most unpredictable results. CONCLUSIONS: Results based on analysis of the dummy data demonstrated that the extent of missing data and the pattern of missing data affected the similarity of the comparisons.  Some form of sensitivity analysis is highly advisable, ideally performed to link the approach to analysis to the pattern of missing data identified in the data set.  LOCF is not recommended as an appropriate approach to handling missing PRO data.