A clinical machine learning model for 1-year functional outcome prediction in acute ischemic stroke: temporal validation across evolving guidelines

Accurate prediction of 1-year excellent functional outcome (modified Rankin Scale [mRS] 0–1) in acute ischemic stroke (AIS) patients is vital for guiding long-term rehabilitation. However, existing tools primarily focus on short-term (3-month) outcomes and often lack validation in temporally distinct cohorts, particularly when clinical guidelines and treatment landscapes evolve. To address this, we trained six machine learning models on a derivation cohort (n = 965, admitted 2020–2023) managed under the 2018 Chinese Guidelines for Diagnosis and Treatment of Acute Ischemic Stroke. The optimal logistic regression (LR) model included eight key predictors: admission NIHSS, admission mRS, age, neutrophil‑to‑lymphocyte ratio (NLR), glucose, blood urea nitrogen (BUN), D‑dimer, and B-type natriuretic peptide (BNP). The LR model was rigorously assessed on an independent temporal validation cohort (n = 144, admitted 2024) treated under the 2023 Guidelines, which expanded indications for reperfusion therapy. Although the validation cohort showed significantly higher thrombolysis rates and milder symptoms than the derivation cohort, the LR model demonstrated robust performance (AUC = 0.80, 95% CI: 0.72–0.87), significantly outperforming admission National Institutes of Health Stroke Scale (NIHSS) score (AUC = 0.73, 95% CI: 0.64–0.81). The model also showed substantial incremental value with a net reclassification improvement of 0.71 and an integrated discrimination improvement of 0.14 (both P < 0.001). Finally, an open‑access web‑based predictor was deployed to facilitate clinical implementation within the first 24 h of admission. In summary, we developed and temporally validated a robust, interpretable prediction model for 1-year functional outcome in AIS, offering a practical tool for long-term prognosis.