Machine learning for early dynamic prediction of functional outcome after stroke

Prediction of outcome after stroke is critical for treatment planning and resource allocation but is complicated by fluctuations during the first days after onset. We propose a machine learning model that can provide hourly predictions based on the integration of continuous variables acquired within 72 h of hospital admission. We analyzed 2492 admissions for ischemic stroke in the Geneva University Hospital from 01.01.2018 to 31.12.2021, amounting to 2’131’752 unique data points. We developed a transformer model that continuously included clinical, physiological, imaging, and biological data recorded within 72 h of admission. This model was trained to generate hourly predictions of mortality and morbidity. Shapley additive explanations were used to identify the most relevant predictors to explain outcomes for each patient. The MIMIC-III database was used for external validation. Our transformer model predicts mortality, with an area under the receiver operating characteristic curve of 0.830 (95% CI 0.763–0.885) on admission, reaching 0.893 (95% CI 0.839–0.933) 72 h later for a 3-month outcome. Validated in an independent cohort, it outperforms all static models. Based on their mean explanatory weights, the top predictors included continuous clinical evaluation, baseline patient characteristics, timing from admission to acute treatment, and markers of inflammation and organ dysfunction. The performance of our transformer model demonstrates the potential of machine learning models integrating clinical, physiological, imaging, and biological variables over time after stroke. The clinical applicability of our model is further strengthened by access to hourly updated predictions along with accompanying explanations. Stroke is the most frequent cause of disability in industrialized countries. To determine the best treatment and allocate resources, an early and accurate prediction of outcome is essential. Although modern stroke units gather a continuous stream of data, existing tools for outcome prediction are rarely used as they are static and fail to adapt to the evolving condition of the patient. We developed a machine learning model, a computer system learning from existing data, to provide real-time predictions of in-hospital mortality and 3-month outcomes. Our model was able to provide accurate hourly prediction of outcome based on regularly updated clinical data obtained from the patient. This study demonstrates the potential of integrating the continuous data stream recorded in the electronic health record after stroke. Similar predictive models could help personalize treatment planning, empower patients and their families through counseling, and facilitate resource allocation. Klug et al. present a machine learning model for continuous monitoring and prediction of functional outcome after acute ischemic stroke. Integrating clinical, physiological, and biological variables over time, the system detects patients at risk as well as potential causes of deterioration.