慢性阻塞性肺疾病急性加重住院患者出院状态预测研究

Predictive study on discharge status of hospitalized patients with acute exacerbation of chronic obstructive pulmonary disease

目的为解决肺功能检测不易获取、测量误差大等问题,结合出院状态和住院的时间,构建机器学习预后预测模型,实现慢性阻塞性肺疾病急性加重期(acute exacerbation of chronic obstructive pulmonary disease,AECOPD)患者预后的精准预测。方法选择2011年10月―2020年5月因AECOPD于山西医科大学第二医院呼吸科住院的患者3035例。结局变量为中位住院时长内是否好转出院。通过构建5种机器学习模型[逻辑回归(logistic regression,LR)、支持向量机(support vector machine,SVM)、随机森林(random forest,RF)、Catboost(categorical boosting)、多层感知机(multilayer perceptron,MLP)]建立预测模型,比较受试者工作特征(receiver operating characteristic,ROC)的曲线下面积(area under curve,AUC)等评价指标,选出最优模型。最后使用最优模型进行决策曲线分析,验证其临床实用性。结果RF相较于其他机器学习模型的综合预测性能最佳,AUC为0.780、准确率为69.69%、精确率为64.50%、召回率为75.18%、F1分数为69.44%、布里尔分数为18.77%,校准曲线基本与对角线一致,决策曲线分析有较好的临床收益。结论基于RF的预测模型可以在无法获得肺功能检测相关指标的情况下实现对AECOPD患者预后的准确预测,为临床医生在评估与治疗决策中提供一定的支持。

Objective A machine learning prognosis prediction model was created by innovatively combining discharge status and length of hospital stay,to accurately predict the prognosis of patients with acute exacerbation of chronic obstructive pulmonary disease(AECOPD).This was done in order to address the issues of difficult to obtain pulmonary function tests and large measurement error.Methods A total of 3035 inpatients with AECOPD were recruited from the second hospital of Shanxi Medical University between October 2011 and May 2020.The outcome variable is whether or not the patient recovered and was discharged within the median length of hospitalization.The prediction model is created using five distinct machine learning models:logistic regression,support vector machine,random forest,Catboost,and multi-layer perceptron.By contrasting evaluation metrics like area under the working characteristic curve(AUROC),the optimal model is determined.In order to verify the decision curve's clinical applicability,the best model was used to assess it.Results In comparison to other machine learning models,random forest has the greatest overall prediction performance,with AUC of 0.780,accuracy of 69.69%,precision of 64.50%,recall of 75.18%,F1 score of 69.44%,and Brier score of 18.77%.The decision curve analysis has high clinical value,and the calibration curve is largely compatible with the diagonal.Conclusions The prediction model based on random forest may reliably forecast the prognosis of patients with AECOPD and provide some aid to physicians in evaluation and treatment decision-making when the important indices of the lung function test cannot be acquired.