An AI Model to Impute ECOG Scores When Missing in Real-World Patient Data

The ECOG Performance Score is a strong prognostic indicator of cancer outcomes. However, it is infrequently collected in real-world treatment settings. Real-world data therefore often does not have ECOG scores in the structured patient record, limiting the ability of data from these patients to be incorporated into clinical trials and External Control Arm studies. In this research abstract, accepted at ASCO 2020, Senior Principal Scientist Vivek Vaidya analyzes findings from a machine learning model built by Concerto HealthAI researchers to impute ECOG values using information from different points in the patient journey using structured data.