Abstract Number: Esoc2026lt35 Protocol: A.I. Based Stroke Risk Factor Classification and Treatment (Abstract) Study

Abstract Background and aims Stroke is a leading cause of death and disability worldwide, costing the UK approximately £26 billion annually. While lifestyle modification and preventative medications reduce risk, 30% of stroke patients have no identifiable risk factors. Hence, there is a need to better identify individuals who are at risk of stroke. ABSTRACT is a three-phase UK Medical Research Countil (MRC) funded study that looks to address this issue by (I) using AI to predict stroke risk from routine hospital data, (II) validating models on external datasets, and (III) evaluating the clinical utility of AI-guided risk classification. This presentation focuses on Phase I. Methods We analysed data from 9,155 stroke patients and 109,581 controls from southwest England (Jan 2003–Nov 2025). Stroke cases were identified via SSNAP. CT/MRI, ECG/echocardiography, laboratory test and medical history data were obtained from hospital and GP records. Separate ML models for each modality will be trained and then ensembled to create a high-performance stroke prediction model. Data from MIMIC-IV and the UKBiobank will be used externally validate models. Results Provisionally, models have achieved stroke prediction accuracies up to 75% for CT/MRI, 76% for ECG/echo, and 92% for laboratory test data. Ensemble model and external validation results are anticipated by May 2026. Conclusions This presentation announces the promising preliminary findings from ABSTRACT Phase I. We also outline our methodological approach in developing a multimodal stroke prediction model, along with plans to perform prospective validation and a subsequent clinical trial comparing AI-guided risk classification against standard care. Conflict of interest