OBJECTIVES: Collection and analysis of quality of life (QoL) data is complex in patients with metastatic breast cancer (mBC) as missing values are often associated with worsened QoL, progressive disease (PD) or death. Nonetheless, it is a relevant endpoint when assessing therapeutic options in mBC. Proper interpretation of data requires sensitivity analyses with different imputation rules for missing data to avoid bias. We compare protocol-specified analysis (no imputation) on QoL in the AVADO trial with two imputation strategies. METHODS: Patients (n=736) were randomised to docetaxel (doc) plus either placebo (control) or bevacizumab (Bv) 7.5 or 15 mg/kg. The self-assessed Functional Assessment of Cancer Therapy-Breast (FACT-B [FACT-G+BCS]) questionnaire was administered at pre-defined timepoints. QoL data were analysed according-to-protocol (no imputation), by imputation rule 1 (missing data for PD set to zero [favours treatment arm]) and imputation rule 2 (missing data for PD replaced with last value carried forward [favours control arm]). Missing values due to death were set to zero for both imputation rules. RESULTS: QoL assessments were completed at baseline and weeks 9, 15 and 33 (after chemotherapy completed), in 656, 570, 510 and 289 patients, respectively. Compared with baseline, mean changes with no imputation were near zero or better for both doc+Bv doses, and were consistently worse for control. Applying imputation rules 1 and 2 gave consistent deterioration of mean values for all arms; however, compared with control, both doc+Bv arms reported statistically significant better scores (p<0.05) at most evaluation timepoints. At week 9, doc+Bv7.5 reported a significant difference with both rules (two extremes). CONCLUSIONS: Sensitivity analysis using different imputation rules to account for missing values is an important exercise to understand accurately the potential QoL benefit of a cancer therapy. This sensitivity analysis suggests that patients receiving doc+Bv may have a significant QoL benefit compared with control group.