Abstract: Stroke is the second leading cause of death and the primary cause of disability worldwide, and its treatment outcomes are highly dependent on early recognition and timely therapeutic intervention. Currently, prehospital stroke prediction tools mainly include three categories: traditional scales, machine learning models, and biomarkers, each with distinct characteristics but significant limitations. Traditional scales (such as FAST and Cincinnati prehospital stroke scale), due to their operational simplicity, serve as primary screening tools at the grassroots level but are insufficient in identifying posterior circulation strokes. Specialized scales for large vessel occlusion (such as the Los Angeles motor scale and rapid arterial occlusion evaluation scale) show relatively high specificity but still face challenges with high false positive rates. Machine learning models (such as extreme gradient boosting and random forest) exhibit superior performance in stroke subtyping and large vessel occlusion prediction, but are constrained by insufficient data dimensions and clinical translation barriers. Biomarkers (such as glial fibrillary acidic protein and S100 calcium-binding protein B) exhibit significant potential in differentiating stroke subtypes, but their prehospital application remains limited due to complex detection techniques and lack of standardization. Future advancements require optimized scale design, integration of multimodal data, development of portable detection technologies, and enhanced prehospital to in-hospital coordination to improve the accuracy of prehospital stroke prediction, so as to guide early prehospital intervention and improve patient prognosis.

Key words: Stroke prediction; Prehospital emergency care; Large vessel occlusion; Machine learning model; Biomarker

摘要： 卒中作为全球第二大死亡原因和首要致残因素，其救治效果高度依赖早期识别与及时治疗干预。目前，院前卒中预测工具主要包括传统量表、机器学习模型及生物标志物三大类，这些诊断工具各具特点但均存在明显局限性。传统量表（如FAST、辛辛那提院前卒中量表）因其操作简便成为基层筛查的主要手段，但对后循环卒中识别不足，而针对大血管闭塞的专项量表（如洛杉矶运动量表、动脉闭塞快速评价量表）虽特异性较高，但仍面临假阳性率偏高的问题。机器学习模型（如极端梯度提升、随机森林）在卒中分型及大血管闭塞预测中展现出优越性能，但受限于数据维度不足和临床转化障碍。生物标志物（如胶质纤维酸性蛋白、S100钙结合蛋白B）在区分卒中亚型方面具有显著潜力，但因检测技术复杂和标准化不足难以在院前应用。未来需通过优化量表设计、整合多模态数据、开发便携检测技术及加强院前院内衔接来进一步提高院前卒中预测的准确性，为指导早期院前干预、改善患者预后提供支持。

关键词: 卒中预测; 院前急救; 大血管闭塞; 机器学习模型; 生物标志物