Comparison of Survival Prognostic Tools in Terminal-Stage Tumor Patients: A Multicenter Prospective Cohort Study

Previous studies have discussed the effectiveness of the Karnofsky Performance Scale (KPS), Palliative Performance Scale (PPS), Palliative Prognostic Index (PPI), and Delirium-Palliative Prognostic Score (D-PaP) in predicting the survival of patients with terminal-stage tumors, but their effectiveness in the Chinese terminal-stage tumor population still needs to be explored. Therefore, we conducted this prospective cohort study to investigate the applicability of KPS, PPS, PPI, and D-PaP in predicting survival in Chinese patients with terminal-stage tumors.

This is a prospective cohort study that collected data from 449 terminal-stage tumor patients across four medical institutions in China from January to December 2024. The feasibility of using KPS, PPS, PPI, and D-PaP to predict the 7-day and 30-day survival of terminal-stage tumor patients was discussed through ROC curves. Kaplan-Meier survival curves were used to examine the survival differences of terminal-stage tumor patients using four survival prediction tools with new cut-off values.

The study ultimately included 449 patients with terminal-stage tumors. For the prediction of 30-day survival, the area under the ROC curve of KPS, PPI, PPS, and D-PaP was 0.842, 0.749, 0.778, and 0.898, respectively. In our study cohort, the optimal cut-off values were 25, 3, 35, and 9.25. Their corresponding sensitivity was 86.3%, 92.3%, 77.7%, and 76.0%, while their specificity was 69.7%, 44.4%, 67.7%, and 93.9%, respectively.

The results of this multicenter cohort analysis reveal that in terminal-stage tumor patients in China, KPS, PPS, PPI, and D-PaP are all effective for predicting survival. Specifically, D-PaP is more applicable for predicting 30-day survival. Although D-PaP showed the most significant predictive power, its accuracy in predicting 7-day and 30-day survival outcomes was not significantly superior to that of experienced palliative care professionals. Therefore, future research still needs to develop or update scoring tools with better predictive performance.