Authors
Hui Chen, Fengli Zuo, Jie Zhang, Feili Li, Fangrong Liu, Xu Tian, Rongying Tang
Published in
Chronic obstructive pulmonary diseases (Miami, Fla.). Nov 12, 2026. Epub Nov 12, 2026.
Abstract
Patients with stable chronic obstructive pulmonary disease (COPD) experience psychological distress that evolves in dynamic and heterogeneous patterns. Conventional cross-sectional screening and baseline-only prediction models fail to capture longitudinal heterogeneity, and few tools integrate trajectory analysis with machine learning (ML) for individualized risk stratification. This study aimed to identify longitudinal trajectories of psychological distress and develop, validate, and deploy a web-based ML risk calculator for clinical use.
We enrolled 323 stable COPD outpatients from three tertiary hospitals in Chongqing, China, between March and December 2024. Baseline data included sociodemographic, clinical, and key psychosocial variables. Psychological distress was reassessed at 1 week, 1 month, 3 months, and 6 months. Growth mixture modeling (GMM) identified distinct longitudinal trajectory classes. The trajectory outcome was dichotomized for prediction: patients in the persistent low-distress group were defined as low-risk, and all others as high-risk. Candidate predictors were screened using univariate analysis and least absolute shrinkage and selection operator (LASSO) regression. Eight ML algorithms were trained, internally validated, and compared, and the best-performing model was selected for final development.
GMM revealed four trajectories: persistent high distress (8.0%), transient threshold-exceeding distress with rapid decline (15.2%), persistent near-threshold distress with gradual decline (65.0%), and persistent low distress (11.8%). Using the persistent low-distress group as reference, nine predictors were retained: symptom burden, number of children, negative coping, payment method, perceived stress, acute exacerbation frequency, dyspnea severity, educational level, and gender. The XGBoost model achieved the best-balanced performance, with AUC values of 0.859 (training), 0.686 (validation), and 0.739 (test), indicating moderate discriminative ability. The model showed high positive predictive value but limited negative predictive value, constraining its utility for ruling out distress. A web-based calculator was subsequently developed for clinical implementation.
Stable COPD patients follow four distinct longitudinal trajectories of psychological distress. The XGBoost model provides moderate predictive performance as an exploratory risk-stratification tool. The web-based calculator may support clinicians in early identification of individuals at elevated risk of persistent psychological distress, but further validation in independent, diverse cohorts is required before widespread clinical adoption.
PMID:
43023932
Bibliographic data and abstract were imported from PubMed on 13 Nov 2026.
Read full publication at:
Please sign in
to see all details.
Advertisement
Stats
- Recommendations n/a n/a positive of 0 vote(s)
- Views 2
- Comments 0