最小绝对收缩和选择算子回归-极端梯度提升模型预测老年急性脑梗死后吞咽障碍患者衰弱发生的回顾性研究

Retrospective study on prediction of frailty in older patients with dysphagia after acute cerebral infarction using a LASSO-XGBoost model

中文关键词: 老年    急性脑梗死    吞咽障碍    衰弱    预测模型    极端梯度提升

英文关键词: older adultsacute cerebral infarctiondeglutition disordersfrailtyprediction modelextreme gradient boosting

中文摘要:

      目的 通过最小绝对收缩和选择算子(LASSO)回归-极端梯度提升(XGBoost)模型，对老年急性脑梗死后吞咽障碍患者发生衰弱的风险因素进行识别。方法 回顾性纳入2022年4月至2025年4月天津市第一中心医院收治的652例老年急性脑梗死后吞咽障碍患者，入院后48 h内完成衰弱评估，分为衰弱组(n=298)和非衰弱组(n=354)。采用LASSO回归、XGBoost进行特征筛选，取交集变量纳入Logistic回归，确定老年急性脑梗死后吞咽障碍患者衰弱发生的影响因素。建立预测模型，绘制受试者操作特征(ROC)曲线，结合曲线下面积(AUC)、校准曲线、决策曲线评估模型效能。结果 652例老年急性脑梗死后吞咽障碍患者衰弱298例(45.71%)。与非衰弱组相比，衰弱组糖尿病、独居、低社会支持水平构成比较高，蒙特利尔认知功能评估(MoCA)量表评分、巴塞尔指数(BI)较低，匹兹堡睡眠质量指数(PSQI)评分较高，差异均有统计学意义(均P <0.05)。经LASSO回归、XGBoost筛选得到居住情况、MoCA评分、PSQI评分、BI、社会支持水平5个相关风险因素。Logistic回归分析显示，高MoCA评分、高BI为老年急性脑梗死后吞咽障碍患者发生衰弱的保护因素(OR=0.801、0.903,P <0.01)，高PSQI评分、低社会支持水平为危险因素(OR=3.251、3.592,P <0.05)。据此构建预测模型，绘制ROC曲线可知AUC为0.895(95%CI:0.871～0.919)。校准曲线与决策曲线显示模型具有良好一致性与临床适用性。结论 MoCA评分、PSQI评分、BI及社会支持水平是老年急性脑梗死后吞咽障碍患者衰弱发生的影响因素。

英文摘要:

      Objective To identify risk factors for frailty in older adults with dysphagia after acute cerebral infarction using a least absolute shrinkage and selection operator(LASSO)regression-extreme gradient boosting(XGBoost)model.Methods A total of 652 older adults with dysphagia after acute cerebral infarction admitted to First Central Hospital of Tianjin City from April 2022 to April 2025 were retrospectively enrolled.Frailty assessment was completed within 48 h after admission,and patients were divided into frailty group(n=298)and non-frailty group(n=354).LASSO regression and XGBoost were used for feature selection.The intersecting variables were included in Logistic regression analysis to determine the influencing factors of frailty.A predictive model was established,and the receiver operating characteristic(ROC)curve was plotted.The performance of the model was evaluated using the area under the curve(AUC),calibration curve,and decision curve analysis.Results Among the 652 patients,298(45.71%)developed frailty.Compared with the non-frailty group,the frailty group had significantly higher proportions of patients with diabetes,living alone,and low social support level,as well as lower Montreal cognitive assessment(MoCA)score and Barthel index(BI),and higher Pittsburgh sleep quality index(PSQI)score(all P＜0.05).LASSO regression and XGBoost screening identified five relevant risk factors:living status,MoCA score,PSQI score,BI,and social support level.Logistic regression analysis showed that higher MoCA score and higher BI were protective factors for frailty(OR=0.801,0.903;both P＜0.01),while higher PSQI score and low social support level were risk factors(OR=3.251,3.592;both P＜0.05).The AUC of the predictive model was 0.895(95%CI:0.871-0.919).Calibration curve and decision curve analyses indicated that the model had good consistency and clinical applicability.Conclusion MoCA score,PSQI score,BI,and social support level are influencing factors for frailty in older patients with dysphagia after acute cerebral infarction.