Analysis of related factors and comparison of prediction models for readmission of patients with mood disorder
10.3969/j.issn.1000-6729.2025.04.01
- VernacularTitle:心境障碍患者再入院影响因素分析及预测模型比较
- Author:
Feng XU
1
;
Peixia CHENG
;
Qian WANG
;
Hua FAN
;
Qi GAO
Author Information
1. 首都医科大学公共卫生学院,北京 100069
- Publication Type:Journal Article
- Keywords:
mood disorders;
readmission;
machine learning algorithms;
predictive models
- From:
Chinese Mental Health Journal
2025;39(4):293-300
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To analysis the influenced factors of readmission in hospital among patients with mood disorder,construct predictive models and compare the predictive performance of the models.Methods:The electron-ic medical record data of patients with mood disorder admitted to Beijing Anding Hospital from January 2010 to De-cember 2018 were retrospectively collected.Utilizing stepwise logistic regression to analyze the related factors of re-admission in patients with mood disorder.Logistic regression,support vector machine,random forest,extreme gradi-ent boosting(XGBoost)algorithms and convolutional neural networks(CNN)were used to construct a readmission prediction model and compare the predictive performances of the different models by using the accuracy,precision,recall,et.al.Results:A total of 6 234 patients with mood disorder were enrolled,24.9％(n=1 549)patients were readmitted after discharge.The stepwise logistic regression results revealed that readmissions were more likely to occur in patients with mood disorder who were female,had comorbidities,had a treatment outcome of cure,were treated with MECT and were on second-generation antipsychotics(OR=1.26,1.68,1.26,1.35,1.18).The CNN model demonstrated the highest accuracy,precision,and F1 scores,all at 0.87,with a recall of 0.86.The random forest achieved a recall of 0.86 and an AUC of 0.95.Conclusion:Readmission of patients with mood disorder is influenced by a multitude of factors.The convolutional neural networks and random forest models outperform other models in prediction.
