Multimodal models play a critical role in mortality prediction for ICU patients.

Traditional emergency patient outcome models rely solely on structured data, overlooking critical insights from unstructured text (e.g., medical notes). We develop a multimodal framework integrating structured clinical metrics with natural language processing (NLP) of text data to predict ED outcomes. Compared to structured-only models, the multimodal approach improves the performance for mortality prediction, with NLP features enhancing interpretability. Combining structured and unstructured data captures holistic patient profiles, demonstrating that multimodal modeling boosts predictive accuracy in emergency care. These findings advocate for routine NLP integration in ED prediction systems.传统急诊患者预后模型仅依赖结构化临床数据，却忽略了非结构化文本（如医疗笔记）中蕴含的关键临床信息。本研究构建了多模态框架，将结构化临床指标与文本数据的自然语言处理（NLP）技术相结合，用于预测急诊（Emergency Department，ED）患者的预后结局。相较于仅使用结构化数据的模型，该多模态方法在死亡率预测任务中表现更优异，且NLP特征可有效提升模型的可解释性。整合结构化与非结构化数据能够构建全面的患者画像，证实多模态建模可提升急诊医疗场景下的预测准确率。本研究结果支持在急诊预后预测系统中常规集成NLP技术。