Abstract:

Objective To conduct an external validation study of the postoperative cardiac delirium patients risk prediction model to provide a basis for clinical screening of appropriate prediction models. Methods The data of patients who underwent cardiac surgery from October 2023 to January 2025 in a tertiary general hospital in southwest Sichuan Province were retrospectively collected. Totally 3 postoperative cardiac delirium patients risk prediction models screened by systematic evaluation were externally validated. The predictive performance of the models was evaluated by plotting and calculating the Area Under Curve（AUC）；the predictive accuracy of the models was evaluated by the Hosmer-Lemeshow test and plotting calibration curves；the clinical utility of the model was evaluated by applying Decision Curve Analysis（DCA）. Results A total of 562 cardiac surgery patients were included，of which 76 had postoperative delirium，and the incidence rate of postoperative cardiac delirium was 13.52%. The results of external validation showed that the AUC of the 3 postoperative cardiac delirium patients risk prediction models ranged from 0.710 to 0.859；the results of the Hosmer-Lemeshow test showed that the predictive accuracy of 2 of the models was better（P=0.865，P=0.270），and the DCA indicated that the net clinical benefit of the perioperative delirium screening in cardiac surgery model had the greatest net clinical benefit. Conclusion All 3 predictive models demonstrated moderate or above discriminatory efficacy（all AUC values >0.700），but only 2 models were well calibrated，and DCA analysis showed some utility for all models. Combining model performance and clinical applicability，external validation concluded that the perioperative delirium screening in cardiac surgery model was superior to the remaining 2 models. Further multicenter studies are needed to optimize the generalizability of the model and to develop intelligent assistive systems to promote clinical application.

Key words: Cardiac Surgical Procedures, Delirium, Prediction Model, External Validation, Nursing Care

摘要：

目的 对心脏术后患者谵妄风险预测模型进行外部验证研究，为临床筛选合适的预测模型提供依据。 方法 回顾性分析四川省西南地区某三级甲等综合医院2023年10月至2025年1月于心脏大血管外科接受手术患者的相关资料。对前期经系统评价筛选出的3个心脏术后患者谵妄风险预测模型进行外部验证。通过绘制并计算受试者操作特征曲线下面积（area under curve，AUC）评价模型预测性能，采用Hosmer-Lemeshow检验、绘制校准曲线评价模型预测准确度，应用决策曲线分析评价模型临床效用。 结果 共纳入562例心脏手术患者，其中76例发生术后谵妄，心脏术后谵妄发生率为13.52%。验证结果显示，3个心脏术后患者谵妄风险预测模型的AUC为0.710~0.859；Hosmer-Lemeshow检验结果显示，其中2个模型校准度良好（P=0.865，P=0.270），决策曲线分析表明，围手术期心脏手术谵妄筛查评分模型的临床净获益最大。 结论 3个预测模型均展现中等及以上的区分效能（AUC值均>0.700），但仅2个模型校准度良好，决策曲线分析显示所有模型均具备一定效用。综合模型性能与临床可操作性，经外部验证，最终得出围手术期心脏手术谵妄筛查评分模型相比其余2个模型更具优势。未来还需进一步通过多中心研究优化模型普适性，并开发智能化辅助系统推动预测模型在临床应用。

关键词: 心脏手术, 谵妄, 预测模型, 外部验证, 护理