Immuno-metabolic biomarkers for 90-day prognostication after acute ischemic stroke: a classically solved QUBO-COPE modeling and biological contextualization study.

BACKGROUND &

objectiveEarly and accessible prognostication after acute ischemic stroke (AIS) is critical for risk stratification, yet translating systemic immuno-metabolic responses into bedside tools remains challenging. This study aimed to develop and externally validate an interpretable prognostic framework based on routine immuno-metabolic biomarkers for predicting 90-day functional outcomes after AIS.

methodsIn this dual-center retrospective cohort study, consecutive AIS patients from two tertiary hospitals in the Huaibei Economic Zone were enrolled. The primary endpoint was unfavorable functional outcome at 90 days, defined as a modified Rankin Scale score ≥ 3. After feature selection via classically solved Quadratic Unconstrained Binary Optimization (QUBO) with simulated annealing, a CRITIC-Optimized Poly-Ensemble (COPE) model integrating six base learners was constructed. A Full Model (including cellular population data) and a Core Model (excluding cellular population data) were internally locked and then evaluated in a held-out external validation cohort. Model interpretation was performed using CRITIC-weighted SHapley Additive exPlanations. Supporting biological contextualization utilized Mendelian randomization, single-cell RNA sequencing, and unsupervised clustering.

resultsA total of 3,812 patients were included (3,093 in the development cohort, 719 in external validation, including 107 with an unfavorable outcome). The QUBO algorithm identified a 10-feature panel comprising neurological severity, inflammatory, and metabolic indices. In external validation, the Full COPE Model achieved an area under the receiver operating characteristic curve of 0.860 (95% CI = 0.815-0.898), a Brier score of 0.118 (95% CI = 0.106-0.130), a specificity of 0.931, and a positive predictive value of 0.592. The Core Model retained comparable discrimination (AUC = 0.870). SHAP analysis revealed that admission neurological severity and inflammatory burden dominated predictions, and unsupervised clustering identified two reproducible sub-phenotypes with divergent outcomes. Mendelian randomization and transcriptomic data provided supportive biological context linking neuroinflammatory signals to prognosis.

conclusionThe framework offers an interpretable, laboratory-based prognostic tool for 90-day AIS outcomes by integrating routine immuno-metabolic biomarkers. Its balanced performance and clinical accessibility support potential utility in risk stratification, though prospective multicenter validation across diverse populations is required before clinical implementation.