Latent classes and construction of a risk prediction model of symptom distress in elderly patients with heart failure

ObjectiveTo identify the latent classes of symptom distress in elderly patients with heart failure and to construct a risk prediction model using machine learning algorithms.MethodsA convenience sampling method was used to select 460 elderly patients with heart failure who were hospitalized in the department of cardiology of a tertiary grade A hospital in Hengyang city from September 2023 to April 2024 as the study subjects.Latent class analysis was employed to explore the latent classes of symptom distress in these elderly patients with heart failure. Six risk prediction models of symptom distress in elderly patients with heart failure were constructed using machine learning algorithms.The optimal model was selected and subjected to interpretability analysis.ResultsSymptom distress in elderly heart failure patients was classified into three latent classes.The low⁃distress symptom⁃adaptation group accounted for 44.8%(206 cases).The moderate⁃distress psychological⁃dysregulation group accounted for 26.5%(122 cases).The high⁃distress heart⁃failure⁃predominant group accounted for 28.7%(132 cases).LASSO regression analysis showed that age,educational level,average monthly household income,multimorbidity,smoking,and perceived social support were significant influencing factors of symptom distress in elderly heart failure patients.Machine learning algorithms were used to construct six models,namely logistic regression,decision tree,random forest,K⁃nearest neighbors,extreme gradient boosting and support vector machine.The area under the receiver operating characteristic curve of extreme gradient boosting in the training set and test set was 0.909 and 0.828,respectively.Which was higher than that of the other models.The calibration curve showed that the predictions of extreme gradient boosting were in good agreement with the actual results.Within a certain threshold probability range,the decision curve of extreme gradient boosting demonstrated that the model possessed favorable clinical application performance.The learning curve showed no overfitting of the extreme gradient boosting.The interpretability analysis based on extreme gradient boosting revealed that perceived social support,multimorbidity,and age were of high importance.ConclusionsSymptom distress in elderly patients with heart failure exhibits population heterogeneity.The extreme gradient boosting model demonstrates good performance in predicting the risk of symptom distress in elderly patients with heart failure.