Li Li, He Hongye, Xiang Linjun, Wang Yongxiang
Nursing Department, The Third Xiangya Hospital, Central South University, Changsha, Hunan, China.
Emergency Department, Xiangya Third Hospital of Central South University, Changsha, Hunan, China.
Perioper Med (Lond). 2025 Jun 2;14(1):60. doi: 10.1186/s13741-025-00544-6.
Postoperative admission to the ICU for surgical patients is a significant burden in nursing care, and there is currently a lack of corresponding assessment tools.
Clinical information of patients was extracted from the VitalDB database. LASSO regression and random forest algorithms were used to screen clinical variables related to postoperative ICU admission. Subsequently, the effectiveness of logistic regression, random forest, support vector machine, and multi-layer perceptron algorithms was compared using ROC curves. After selecting the best algorithm, postoperative ICU admission probability prediction nomogram was constructed.
This study identified 18 clinical factors that influence postoperative ICU admission. The factors influencing patient outcomes include three physiological characteristics: age, weight, and gender; five preoperative laboratory tests:platelet count, prothrombin time(%),activated partial thromboplastin time, albumin, and blood urea nitrogen; and seven intraoperative anesthesia details: anesthesia duration, propofol dosing during surgery, midazolam dosing during surgery, phenylephrine dosing during surgery, calcium chloride dosing during surgery, American Society of Anesthesiologists (ASA) classification, and anesthesia method. Additionally, three other factors are considered: whether the surgery is classified as an emergency, the department category, and the type of surgery. The logistic regression model developed using these 18 variables was identified as the most effective predictive model for postoperative ICU admission, achieving an ROC AUC of 0.925.
The postoperative admission warning model constructed in this study can effectively predict the probability of patients being admitted to the ICU after surgery, providing a corresponding management tool for postoperative care in surgical patients.
外科患者术后入住重症监护病房（ICU）给护理工作带来了沉重负担，目前缺乏相应的评估工具。
从VitalDB数据库中提取患者的临床信息。使用LASSO回归和随机森林算法筛选与术后入住ICU相关的临床变量。随后，使用ROC曲线比较逻辑回归、随机森林、支持向量机和多层感知器算法的有效性。在选择最佳算法后，构建术后入住ICU概率预测列线图。
本研究确定了18个影响术后入住ICU的临床因素。影响患者预后的因素包括三个生理特征：年龄、体重和性别；五项术前实验室检查：血小板计数、凝血酶原时间（%）、活化部分凝血活酶时间、白蛋白和血尿素氮；以及七项术中麻醉细节：麻醉持续时间、手术期间丙泊酚给药量、手术期间咪达唑仑给药量、手术期间去氧肾上腺素给药量、手术期间氯化钙给药量、美国麻醉医师协会（ASA）分级和麻醉方法。此外，还考虑了其他三个因素：手术是否列为急诊、科室类别和手术类型。使用这18个变量开发的逻辑回归模型被确定为术后入住ICU最有效的预测模型，ROC曲线下面积（AUC）为0.925。
本研究构建的术后入住预警模型能够有效预测患者术后入住ICU的概率，为外科患者术后护理提供了相应的管理工具。
