Related Experiment Video
Updated: Jul 30, 2026
A Machine Learning Approach to Design an Efficient Selective Screening of Mild Cognitive Impairment
Published on: January 11, 2020
Comparison and validation of machine learning-based screening models for elevated depressive symptoms in peritoneal
Yugang Cao1,2,3, Dongzhi Yin1,2,3, Xiaoming Yu1,2,3
1Department of Hepatobiliary Surgery, Huangshi Central Hospital, Affiliated Hospital of Hubei Polytechnic University, Huangshi, Hubei, China.
Purpose:
To establish an accurate and generalizable concurrent screening system for elevated depressive symptoms (defined as Self-Rating Depression Scale (SDS) score ≥53) in peritoneal dialysis (PD) patients, identify core associated factors and their associative pathways, and develop a clinically practical tool to support early screening and individualized intervention.
Methods:
A multicenter retrospective cross-sectional study was conducted, enrolling 601 PD patients from two centers, including 482 patients from Huangshi Central Hospital (356 in training group, 126 in internal validation group) and 119 patients from Honghu People's Hospital as the external validation group. LASSO regression was used to screen key predictors. Nine machine learning models were constructed and validated, with SHAP analysis to improve model interpretability. Structural Equation Modeling (SEM) quantified direct associations of key factors, and a R Shiny-based online visualization tool was developed for clinical application. Stratified analysis confirmed significant subgroup differences in risk of elevated depressive symptoms.
Results:
Six key predictors were identified: Age, Peritonitis, Catheter-Related Complications, anxiety status (SAS score), SSRS score, and Peritoneal Dialysis Vintage. The XGBoost model showed optimal performance (external validation AUC = 0.869, F-measure = 0.63). SEM confirmed significant direct associations of all 6 factors with elevated depressive symptoms. The developed online visualization tool enabled rapid risk assessment, and stratified analysis showed significant risk differences across subgroups (all p < 0.05).
Conclusion:
This study develops an interpretable screening system with promising performance for elevated depressive symptoms in PD patients. The XGBoost-based online visualization tool provides a user-friendly clinical tool, while identified key factors clarify intervention targets, facilitating early screening and personalized care to improve patients' mental health and long-term prognosis.
