急性药物中毒性脑病患者ICU住院时间预测模型的构建

Development a predictive model for prolonged ICU stay in patients with toxic encephalopathy induced by acute drug poisoning

目的 构建急性药物中毒性脑病患者重症监护室(intensive care unit, ICU)住院时间延长的预测模型并评价其效能。方法 选择重症监护医疗信息集市(MIMIC)-Ⅳ2.2数据库中148例急性药物中毒性脑病患者作为研究对象,收集患者临床资料,根据ICU住院时间分非延长组(≤48 h)与延长组(>48 h)。采用最小绝对收缩和选择算子(LASSO)回归联合Logistic回归筛选变量,构建和绘制列线图。分别采用受试者工作特征曲线下面积(AUC)、Hosmer-Lemeshow校准曲线和决策曲线分析(DCA)评价模型的区分度、校准度及临床适用度。结果 患者ICU住院时间1~15 d,其中ICU住院时间延长69例,采取LASSO回归与Logistic回归相结合方法筛选预测变量。结果显示SOFA评分、心率、合并心血管疾病、使用机械通气4个变量为独立危险因素,依据以上预测变量构建和绘制列线图,列线图的AUC为0.837,95%CI 0.774~0.900;Bootstrap内部验证AUC 0.873,95%CI 0.817~0.930,说明该列线图预测模型具有较好的预测能力。校准曲线和Hosmer-Lemeshow检验(χ^(2)=6.392,P=0.603)均显示该模型具有较高的一致性和拟合度;DCA结果表明,患者可从模型中净获益(阈值范围0.05~1.00),具有较好的临床适用性。结论 本研究开发的模型性能良好,有助于评估急性药物中毒性脑病患者ICU住院时间的延长风险。

Objective Develop a prediction model to assess the risk of prolonged ICU stay in patients with toxic encephalopathy induced by acute drug poisoning and evaluate the performance of the model.Methods The clinical data of 148 cases of toxic encephalopathy induced by acute drug poisoning from the Medical Information Mart for Intensive Care(MIMIC)-Ⅳ2.2 database were collected.Patients were categorized into two groups based on ICU length of stay:the non-prolonged ICU stay group(≤48 hours)and the prolonged ICU stay group(>48 hours).Variable selection was carried out utilizing both LASSO and Logistic regression,followed by constructing and plotting a nomogram.The discrimination,calibration,and clinical utility of the nomogram were evaluated using the area under the receiver operating characteristic curve(AUC),Hosmer-Lemeshow calibration curve,and decision curve analysis(DCA),respectively.Results In this study,ICU length of stay for patients ranged from 1 to 15 days,with 69 cases experiencing prolonged ICU stay.The combined method of LASSO and Logistic regression was employed for variable selection.The results indicated that SOFA score,heart rate,presence of cardiovascular disease,and utilization of mechanical ventilation were identified as independent risk factors.A nomogram was constructed using multiple Logistic regression based on these variables.The AUC of the nomogram was 0.837,95%CI 0.774~0.900,and through internal validation using Bootstrap samples,the AUC was determined to be 0.873,95%CI 0.817~0.930,highlighting its excellent predictive capability.The calibration map showed that the calibration curve of the nomogram model was very close to the standard curve,with the goodness of fit test(χ^(2)=6.392,P=0.603).In addition,DCA results suggested the model had a good net benefit(threshold probability was 0.05~1.00).Conclusion The model developed in this study demonstrated better performance.It may help predict the risk of prolonged ICU stay for patients with toxic encephalopathy induced by acute drug poisoning.