内科

目的探讨肝硬化患者出院30 d内非计划再入院的相关预测因子，并基于随机森林算法构建风险预测模型，初步评估其预测性能。方法采用回顾性分析方法，收集2023年5月至2024年1月于南宁市某专科医院住院的292例肝硬化患者的临床资料。根据出院30 d内是否发生非计划再入院将患者分为两组。采用单因素分析及随机森林模型筛选预测变量，并构建预测模型。采用受试者工作特征（ROC）曲线下面积（AUC）、准确度、召回率及特异性评估模型在独立验证集上的性能。结果 292例患者中，86例（29.45%）发生出院30 d内非计划再入院。随机森林预测模型筛选出的前5位重要预测变量依次为终末期肝病模型（MELD）积分、单核细胞/淋巴细胞比值（MLR）、活化部分凝血活酶时间（APTT）、丙氨酸氨基转移酶（ALT）及年龄。在验证集上，模型的ROC AUC为0.746 9，召回率（敏感性）为1.00，特异性为0.91，整体准确度为0.84，精确度约为0.92。结论本研究初步构建的随机森林预测模型显示出对肝硬化患者短期再入院风险一定的预测潜力，其筛选的预测变量具有临床参考价值。但模型性能仍需优化，且结论有待更大样本、多中心的外部数据进一步验证。

Objective To explore the relevant predictors of unplanned readmission within 30 days after discharge in patients with liver cirrhosis, construct a risk prediction model based on the random forest algorithm, and preliminarily evaluate its predictive performance. Methods A retrospective analysis was conducted, and the clinical data of 292 patients with liver cirrhosis who were hospitalized in a specialized hospital in Nanning from May 2023 to January 2024 were collected. The patients were divided into two groups according to whether unplanned readmission occurred within 30 days after discharge. Univariate analysis and random forest model were used to screen predictive variables, and a predictive model was constructed. The area under the receiver operating characteristic (ROC) curve (AUC), accuracy, recall, and specificity were employed to evaluate the performance of the model on the independent validation set. Results Among the 292 patients, 86 (29.45%) had unplanned readmission within 30 days after discharge. The top 5 important predictive variables screened by the random forest prediction model were in the following order: Model for End-Stage Liver Disease (MELD) score, monocyte/lymphocyte ratio (MLR), activated partial thromboplastin time (APTT), alanine aminotransferase (ALT), and age. On the validation set, the model achieved an ROC AUC of 0.746 9, a recall (sensitivity) of 1.00, a specificity of 0.91, an overall accuracy of 0.84, and a precision of approximately 0.92. Conclusion The preliminarily constructed random forest prediction model in this study shows a certain predictive potential for the short-term readmission risk of patients with liver cirrhosis, and the screened predictive variables have clinical reference value. However, the performance of the model still needs to be optimized, and the conclusions need to be further verified by larger-sample and multi-center external data.

ref
ref
ref
ref
ref
ref
ref
ref
ref
ref
ref