Abstract Number: Esoc2026a291 Patient Perspectives on Machine Learning Tools for Outcome Prediction and Decision-Making in Intracerebral Haemorrhage: A Mixed-Methods Study

Abstract Background and aims Machine learning (ML) tools hold promise in outcome prediction to assist clinicians in their decision making for patients with acute intracerebral hemorrhage (ICH). Little is known of attitudes and barriers to adoption of this new technology in practice. Our study aimed to explore stroke patients’ perspectives on ML-based tools for ICH. Methods Surveys (n=186) and focus group interviews (n=9) of people who had experienced ICH were conducted. Qualitative analyses were conducted on survey data. Survey free-text responses and focus group transcriptions were analysed qualitatively using inductive thematic analysis. Results Most patients (81%) were comfortable with the idea of ML-based software supporting clinicians in outcome prediction for ICH. Concerns included the accuracy and quality of prognostic outputs, use of relevant and contemporaneous data for model training, and patient knowledge of and trust in artificial intelligence. Some patients (19%) had concerns over the collection and use of medical data for the training of the model. Strategies suggested to improve patient comfort and confidence included evidence of the model's performance, involvement of clinicians in the design and implementation of the model, and more information generally on the model. Other concerns raised included ethical use, potential overreliance by clinician users, and the need for clear communication to the patient and family around the use of ML models in their care. Conclusions Patients were comfortable with ML-based tools for ICH. Strategies to improve patient comfort and confidence during implementation were discussed. Conflict of interest Alexandra Hurden: nothing to disclose; Menglu Ouyang: nothing to disclose; Leibo Liu: nothing to disclose; Xiaoying Chen: nothing to disclose; Craig Anderson; nothing to disclose.