OBJECTIVES: Assessment of Quality of Life (QoL), a Patient Reported Outcome (PRO), has gained acceptance as a study endpoint. An open-label, multicenter phase II, single arm oncology study was conducted with a QoL endpoint aiming to assess change of scores from baseline to 12-week or End of Study (whichever occurred first). This required the availability of the baseline and at least one post-baseline assessment. Unfortunately, missing data affects the validity of QoL assessment. A set of different techniques to deal with missing data was compared.METHODS: The analysis addressed the global health status scale (QL range [1;100]) of QLQ-C30, the EQ-5D utility index (Utility range [0;1]) and Visual Analysis Scale (VAS range [0;100]). Five multiple imputation (MI) techniques were carried out with two softwares (SAS, IVEware) and compared with Rubin’s efficiency: Monte-Carlo Markov Chain (MCMC), Expectation-Maximization (EM), Regression (REG), Propensity score (PROP) and Sequential regression (SEQ), using 5 simulations per technique. RESULTS: Changes significance varied depending on the imputation technique. At baseline, mean scores were: QL 0.73, EQ-5D index 63.4, and VAS 68. For QL score, the change estimations were (mean [95%CI]): -1.699 [-3.322;-0.076] (MCMC), -1.558 [-3.132;0.015] (EM), -1.795 [-3.449;-0.141] (REG), -1.197 [-3.067,0.673] (PROP), -0.895 [-5.622;3.832] (SEQ). For EQ-5D index, estimations were: -0.020 [-0.042;0.003] (MCMC), -0.018 [-0.043;0.007] (EM), -0.018 [-0.034;-0.002] (REG), -0.015 [-0.045;0.014] (PROP), -0.010 [-0.049;0.030] (SEQ). VAS changes varied from 0.019 (SEQ) to 0.791 (PROP), no change estimation was significant. Rubin’s efficiency was comprised between 88.32% and 94.43% depending on score and technique.CONCLUSIONS: Results have to be carefully interpreted since they vary according to the MI method. SEQ is the only method not assuming a normal distribution of the data and consequently displays large confidence intervals. Nevertheless, multiple imputation is told to be robust to normality. A sensitivity analysis is advised in order to compare the different results.