JMIR Publications

ABSTRACT

Background:

Pulmonary dysfunction-depression comorbidity (PDDC) constitutes a clinically significant syndemic affecting over 500 million individuals worldwide, with depression prevalence reaching up to 39.5% among pulmonary tuberculosis patients compared to 10.6% in the general Chinese population. Despite growing understanding of their bidirectional relationship, critical gaps remain in methodological approaches for identifying associated factors.

Objective:

This study aimed to develop a hybrid analytical framework integrating traditional statistical models and machine learning algorithms to identify key associated factors of PDDC while balancing analytical performance with clinical interpretability.

Methods:

Data from 1,146 participants with pulmonary dysfunctions in the China Health and Retirement Longitudinal Study (the 2011 and 2015 waves) were analyzed. Three analytical models—Logistic Regression (LR), Bayesian Networks (BN), and eXtreme Gradient Boosting (XGBoost)—were constructed and validated using comprehensive metrics including area under the receiver operating characteristic curve (AUC).

Results:

LR and BN models demonstrated comparable discriminative capacity (AUC: 0.734 vs. 0.735) but different operational characteristics. Six core associated factors were consistently identified across models: mental illness (strongest predictor, LR: OR=7.63; BN: OR=5.99), nephropathy, arthritis, gastropathy, household registration status, and BMI. Urban household registration (LR: OR=0.62; BN: OR=0.66) and higher BMI (LR: OR=0.93; BN: OR=0.93) showed protective effects.

Conclusions:

This multi-model comparative approach provides important guidance for analytical tool selection in clinical practice. The identified core associated factors, particularly mental illness and chronic multimorbidity, offer a basis for risk stratification and targeted interventions in PDDC management.