Stroke Prognostication in Patients Treated with Thrombolysis Using Random Forest

Background: Early identification and accurate prognostication of acute ischemic stroke are crucial due to the narrow time frame for treatment and potential complications associated with thrombolysis intervention. Objectives: This pilot study in the Southeast Asian region using Indonesian data, aims to develop a novel machine learning model for predicting the clinical outcome of acute ischemic stroke patients following thrombolysis. The model seeks to aid clinicians in identifying eligible candidates for thrombolysis therapy. Methods: This retrospective study at Cipto Mangunkusumo Hospital’s medical records from 2014 to 2023 used non-contrast brain CT, clinical, and lab data to develop a Random Forest (RF) algorithm predicting Δ NIHSS (National Institutes of Health Stroke Scale) score, indicating functional outcome. The developed RF model was applied to a validation dataset, with performance evaluated. The study also compared RF with a previous Convolutional Neural Networks (CNN) algorithm. Results: This study included 145 acute ischemic stroke patients treated with thrombolysis. It demonstrated the promising feasibility of using machine learning algorithms to predict clinical outcomes in this population. Integration of CT, clinical, and laboratory data as inputs to the RF models shows the best prediction performance (Accuracy = 0.75, AUC = 0.72, F1=0.50, Precision=0.60, Sensitivity=0.43, Specificity=0.88) Conclusions: The application of machine learning shows the potential to enhance the selection process for thrombolysis intervention in treating acute ischemic stroke. Further research with larger multicenter datasets and additional imaging modalities is required to improve predictive ability.