OBJECTIVES : Advances in real-time data collection and analysis represent an emerging theme that has integrated stakeholders in pursuit of value-added health management guidance. This research aims identify outcomes to be evaluated in a value-based program for stroke patients and present preliminary results of a data analysis structure used to develop computational models to automatize outcomes identification from EMR.

METHODS : Outcomes identification step was based on literature review, including ICHOM standard stroke set and specific studies discussing stroke outcomes, followed by discussion with experts. The outcomes identified were classified into three categories of variables: structure (safe and effective service), process (clinical management and treatment process), and clinical (Rankin, self-care, communication, and cognitive capability). A spreadsheet was built for Python algorithms training process. A multidisciplinary team is working on extracting structured information from EMR using supervised machine learning approaches.

RESULTS : A total of 25 outcomes were identified and transformed in variables that can be extracted from EMR, 12 (Structure), 05 (Process), 08 (clinical answer). The definition of each semantic concept and the value assigned to each word is predetermined through an automatically generated language model. Initial tests with patients’ evolution data began to be performed to validate algorithm performance. A total of 4932 sentences were already evaluated and our preliminary results for 19 variables we have trained showed promising results with assertiveness around 0.70 for few variables and in average 50% of assertiveness evaluated by F1-scores and MCC. The research is being continued and further data will contribute to robustness. Algorithm may be scaled to evaluate approximately 950 patients from 4 hospitals.

CONCLUSIONS : Progress in artificial intelligence for data analysis is essential to enable more effective health-care management. This research is unprecedented in Brazil in discussing feasibility of making value-oriented priority care lines and has potential to contribute, on national scale, to value-based health-care management programs.