不同算法的老年吞咽功能障碍患者吸入性肺炎风险预测模型比较

Comparison of risk prediction models for aspiration pneumonia in elderly patients with swallowing dysfunction based on different algorithms

目的:对比logistic回归模型与随机森林模型对老年吞咽功能障碍患者吸入性肺炎(AP)风险预测价值。方法:于2021年1月至2022年10月,采用便利抽样法,选取南京市第一医院收治的老年吞咽功能障碍患者450例为研究对象。采用单因素和多因素logistic回归分析法分析AP发生的影响因素,并建立logistic回归模型、随机森林模型。采用受试者工作特征(ROC)曲线评估两种模型对AP发生的预测效能。结果:450例老年吞咽功能障碍患者AP发生率为42.00%;多因素logistic回归分析显示,意识障碍、胃食管反流、鼻饲或胃肠道营养、体位不当、CRP、NLR、LER、洼田饮水试验分级为AP发生的危险因素,而电磁刺激疗法、口腔清洁规范、吞咽训练为AP发生的保护因素(P<0.05);随机森林算法显示,AP发生影响因素重要性前8位依次为洼田饮水试验分级、吞咽训练、电磁刺激疗法、胃食管反流、意识障碍、咳嗽无力、体位不当、口腔清洁规范。Logistic回归模型预测AP发生的ROC曲线下面积(AUC)为0.836,预测一致率为0.432;随机森林模型预测AP发生的AUC为0.938,预测一致率为0.712。结论:随机森林算法对AP发生的预测效能优于logistic回归,临床可依据预测模型预测AP发生风险,并制定针对性干预措施,以防止AP发生。

Objective:To compare the value of logistic regression model and random forest model in predicting the risk of aspiration pneumonia(AP)in elderly patients with swallowing dysfunction.Methods:A total of 450 elderly patients with swallowing dysfunction in Nanjing First Hospital from January 2021 to October 2022 were selected by convenience sampling method.The factors influencing the occurrence of AP were analyzed by univariate and multivariate logistic regression,and logistic regression model and random forest model were established.The receiver operating characteristic(ROC)curve was used to evaluate the prediction efficiency of the two models for AP occurrence.Results:The incidence of AP was 42.00%in 450 elderly patients with swallowing dysfunction.Multivariate logistic regression analysis showed that consciousness disorder,gastroesophageal reflux,nasal feeding or gastrointestinal nutrition,improper posture,CRP,NLR,LER and lowland drinking water test were the risk factors for AP occurrence,while electromagnetic stimulation therapy,oral cleaning standard and swallowing training were the protective factors for AP occurrence(P<0.05).According to the random-forest algorithm,the top 8 influential factors for the occurrence of AP were the grade of lowland drinking water test,swallowing training,electromagnetic stimulation therapy,gastroesophageal reflux,consciousness disorder,cough weakness,improper posture,and oral cleaning standard.The logistic regression model predicted the occurrence of AP with an area under ROC curve(AUC)of 0.836 and a prediction agreement rate of 0.432,while the random forest model predicted the occurrence of AP with an AUC of 0.938 and a prediction agreement rate of 0.712.Conclusion:The prediction efficiency of random forest algorithm is better than logistic regression.Clinical prediction model should be used to predict the risk of AP occurrence,and corresponding targeted intervention measures should be formulated to prevent AP from occurring.