Abstract
Background: Prognosis of overall survival is instrumental for patient management and can improve conduct of clinical trials and real-world data analysis. With the shift towards cancer immunotherapy, modeling of overall host fitness becomes increasingly important. Here, we compare the performance of contemporary prognostic scores constructed from routinely measured biomarkers.
Patients and methods: We used patient data from the Flatiron Health electronic-health record de-identified oncology database and from 16 clinical studies sponsored by Roche. A total of 64,233 patients were analyzed, covering the most prevailing solid tumor and hematology cancer types.
We compared the Royal Marsden Hospital score (solid tumors), international prognostic index (IPI) (blood tumors), the Eastern Cooperative Oncology Group (ECOG) performance status, and the ‘Real wOrld PROgnostic score (ROPRO)’. OS was modeled from the start of treatment using Kaplan-Meier analysis and Cox regression.
Results: All investigated scores proved to be prognostic, both in RWD and clinical trial data, and in all indications from the respectively intended range of application. The ROPRO uniformly outperformed other prognostic scores. Concordance indices / hazard ratios in the range of [0.64;0.73]/[2.80;4.50] were found for ROPRO, and in the range of [0.53;0.65]/[1.55; 3.10] for the remaining scores. In hematology trials, the IPI came close to the performance of ROPRO.
Conclusions: Strong and easy-to-apply prognostic scores for overall survival exist. The usage of all investigated scores can be recommended. With moderate extra effort, the implementation of ROPRO can create considerable improvement.