ABSTRACT NUMBER: ESOC2026LB126 OPPORTUNISTIC RISK STRATIFICATION OF INTRACRANIAL ANEURYSMS USING INTEGRATED ROUTINE BLOOD TESTS: A HIGH-SENSITIVITY MACHINE LEARNING EXCLUSION TRIAGE MODEL

Abstract Background and aims Definitive diagnosis of intracranial aneurysms (IA) requires cerebral angiography, which has limited availability. A high-sensitivity exclusion triage biomarker is developed using machine learning and routine haematological and biochemical tests, age, and sex. Methods 247 patients who underwent computed tomography angiography were analysed, retrospectively. 128 (52%) patients were confirmed to have IA, and 119 patients had no IA. Twenty-one routinely available blood components, along with age and sex, were used to train machine learning models to exclude control cases. Optimisation prioritised sensitivity and negative predictive value (NPV) to minimise false negatives. Fifteen exclusion triage models were compared. Results AdaBoost is the top-performing model. Mean age is 57 ± 15 years. Females are more prevalent in the IA group (75.7%) than among controls (44.7%). Subarachnoid haemorrhage is present in 75% (96/128) of IAs. The model has an AUROC of 0.75 (95% CI: 0.62–0.88), a sensitivity 99% (95% CI: 99–99%), and an NPV 99% (95% CI: 99–99%). Specificity was 18% (95% CI: 6–31%) and positive predictive value 50% (95% CI: 29–70%). The model successfully flagged and excluded 32 unnecessary controls without reducing IA detections, demonstrating utility as an opportunistic safety-focused triage. Conclusions This model achieved 99% sensitivity and NPV, supporting its role as an opportunistic exclusion triage tool for IA. Implementation leverages and repurposes routine laboratory data and health records, enabling near real-time risk stratification without requiring additional diagnostic infrastructure. Prospective external validation is in progress. Conflict of interest