Development and Internal Validation of a Clinical Imaging–Based Stroke Prediction Model in a Community Cohort

作者：Yiqing Wang, Dingding Zhang, Fei‐Fei Zhai, Ziyue Liu, Ning Su, Lixin Zhou, Jun Ni, Ming Yao, Liying Cui, Bin Peng, Mingli Li, Zhengyu Jin, Shuyang Zhang, Fei Han, Yi‐Cheng Zhu · 发表于：Journal of the American Heart Association · 年份：2025 · DOI：10.1161/jaha.125.041413 · 被引用次数：3 · 研究领域：Acute Ischemic Stroke Management、Cerebrovascular and Carotid Artery Diseases、Dementia and Cognitive Impairment Research

BACKGROUND: Existing stroke prediction models tend to overestimate contemporary stroke risk and inadequately incorporate neuroimaging parameters. This study aimed to develop a novel and accurate stroke prediction model for a community-based population by integrating comprehensive neuroimaging data. METHODS: A prospective cohort study was conducted involving 1586 eligible participants from northern rural China. Baseline clinical and neuroimaging data were collected, with annual follow-ups to assess incident stroke. Least absolute shrinkage and selection operator regression was used to identify key predictors, which were subsequently incorporated into a Cox proportional hazards model to develop the final predictive model. Internal validation was performed via 500 times bootstrap resampling. Model performance was evaluated using time-dependent receiver operating characteristic curves, Brier scores, calibration plots, and decision curve analysis. RESULTS: During a mean follow-up of 8.0 years, 54 incident strokes occurred among 1173 participants (4.6%). The final model incorporating least absolute shrinkage and selection operator-selected predictors (current smoking, diabetes, high cerebral small-vessel disease burden, and severe intracranial artery stenosis) showed strong discrimination, with area under the curve values of 0.88 (95% CI, 0.83-0.92) for 5-year and 0.84 (95% CI, 0.79-0.90) for 7-year prediction. Bootstrap validation confirmed model robustness (area under the curve v...