Survival Prediction of Patients with Advanced Cancer: The Predictive Accuracy of the Model Based on Biological Markers

To determine whether the addition of biological markers to performance status (PS) and physical symptoms would improve survival prediction among patients with advanced cancer, we developed two prediction models with a scoring system based on 294 consecutive patients with advanced cancer (training set), and then tested its validity on another 93 patients (testing set). We assessed the predictive accuracy of the models using receiver-operating characteristic analysis. Albumin (ALB), lactate dehydrogenase (LDH), and lymphocyte percentage (Lymp%) were significantly and independently associated with survival length. For prediction of 60-day survival, the predictive accuracy of Model 2, based on the above biological markers in addition to PS and symptoms, was significantly better than that of Model 1, based on PS and symptoms alone (area under the curve [AUC] for Model 2, 0.80 ± 0.03; AUC for Model 1, 0.69 ± 0.04; P < 0.001). Addition of ALB, LDH, and Lymp% to PS and physical symptoms improved prediction accuracy, especially for longer survival.