Tan, Agnes W.H.

6116 Background: Analyses of patient-reported outcomes rely on the dependability of patients to complete and submit assessments in a timely manner – not all data is obtained. In recent work focusing on quality of life (QOL) data and imputation, it has been found that most methods do not alter study results. But how much data can be missing before study results are affected? Methods: Missing data was investigated using a 2-arm study (109 patients) who completed Linear Analogue Self Assessments at 4 intervals. Patients (11%) had missing data at the second interval. Existing data was analysed for differences in scores between arms, then cases were randomly deleted to create increasing percentages (12%-20%) of missing data. Ten simulations were conducted per percent. Imputation methods applied were carrying forward the last value (LVCF), average value (AVCF), and maximum value (MVCF). Student’s t-tests were performed between arms for each simulation. Results: Imputation did not alter results of our study data which was statistically significant (SS) between arms for overall QOL (p=0.036) and spiritual well-being (SWB) (p=0.006), and not statistically significant (NS) for mental well-being (MWB) (p=0.174). After data deletion and t-test calculations, AVCF did not impact results. For overall QOL, data deletion changed the p-value to NS in 1 of 10 simulations starting at 12% missing data and 5 of 10 simulations starting at 16% missing data. No matter what percentage of missing data, imputation produced a SS p-value over 80% of the time. Data deletion and subsequent imputation did not affect the study decision for SWB. For MWB, all differences between arms were NS prior to imputation. After imputation, there was at most a 7% disagreement in conclusions. LVCF and MVCF performed equally in all simulations. Conclusions: For this particular study, when p-values are close to the study-defined alpha, the increase in missing data can change the study results and imputation methods are more likely to determine SS differences. The further the p-values are from the study alpha, there is little effect from increasing missing data or applying imputation. These results are for one particular study and further research is needed. No significant financial relationships to disclose.