MACHINE LEARNING IN STROKE OUTCOME PREDICTION: SYSTEMATIC REVIEW OF PROGNOSTIC MODELS AND CLINICAL DECISION SUPPORT APPLICATIONS

Abstract

Stroke is the second leading cause of death and the leading cause of acquired disability globally, with approximately 13.7 million new strokes occurring annually and 5.5 million resulting in death. Accurate prediction of stroke outcomes — functional recovery, mortality, and complication risk — is essential for informed clinical decision making, rehabilitation planning, patient and family counselling, and healthcare resource allocation. Machine learning models applied to clinical, imaging, and biological data from acute stroke patients have demonstrated superior prognostic performance compared with established clinical scoring systems, achieving AUC values of 0.80–0.95 for 90-day functional outcome prediction compared with 0.70–0.80 for traditional scores (NIHSS, ASPECTS). This systematic review evaluates 41 studies of machine learning stroke outcome prediction models published between 2016 and 2024, analysing input variables, model architectures, performance benchmarks, and clinical integration prospects. The implications for stroke care in Uzbekistan, where stroke incidence is among the highest in Central Asia and specialist neurology capacity is constrained, are examined.