The aim of the project was to identify clinical and quality of life (QL) factors that together predict survival and response to chemotherapy in advanced breast cancer. Potential prognostic factors were studied in 187 women with baseline QL data from a trial of paclitaxel versus doxorubicin as ®rst-line chemotherapy. Demographic and clinical factors studied were age, performance status, dominant site of disease and preceding disease-free interval (DFI). Factors from the EORTC QLQ-C30 were all function scales, fatigue, nausea/vomiting, pain, dyspnoea, insomnia, loss of appetite and global QL. The proportional hazards regression model with strati®cation for treatment, and the logistic regression model adjusting for treatment arm were used for univariate and multivariate analyses of survival and response to treatment, respectively. For survival, multiple sites of visceral disease, pain, global QL and fatigue were signi®cant prognostic factors in the univariate analysis. The ®nal multivariate model predicted poor survival with multiple sites of visceral disease (P=0.003), DFI 42 years (P=0.026) and pain (P=0.003). For response, age, dyspnoea, fatigue and global QL were signi®cant predictive factors in the univariate analysis. The ®nal multivariate model for response selected DFI (P=0.009), multiple sites of visceral disease (P=0.037) and dyspnoea (P=<0.001) using forward selection, but model instability was indicated by the inclusion of fatigue and emotional function in the ®nal model when backward selection was used. In addition to known clinical factors, patient-assessed QL variables appear to be prognostic for survival and response to chemotherapy in women with advanced breast cancer. However, identi®cation of prognostic factors from responses to questionnaires may be unstable, and their reliability and clinical utility should be tested prospectively. #