Development and validation of a predictive model for depression in patients with advanced stage of cardiovascular-kidney-metabolic syndrome

Comprehensive Summary

This study by Zha et al. investigates the development of a clinically interpretable model to predict depression risk in patients with advanced-stage cardiovascular-kidney-metabolic (CKM) syndrome, a population known to have high psychiatric vulnerability but limited targeted screening tools. Using NHANES data from 2011-2020, the authors included 1,072 adults with advanced CKM and split them into training and test sets, with three additional NHANES cycles used for external validation to assess generalizability. Depression was defined using the PHQ-9, and 35 demographic, clinical, laboratory, and medical history variables were considered. LASSO regression was first applied to reduce dimensionality and address collinearity, followed by multivariate logistic regression, while random forest, support vector machine, and decision tree models were constructed for comparison. The final LASSO-logistic model demonstrated the strongest and most balanced predictive performance (AUC ≈ 0.77 in the test set), outperforming more complex machine-learning approaches, and identified sleep disorders, younger age, sex, lower poverty-income ratio, greater waist circumference, and elevated gamma-glutamyl transferase (GGT) as independent predictors of depression. In the discussion, the authors interpret these findings as reflecting a convergence of metabolic dysfunction, inflammation and oxidative stress, socioeconomic strain, and psychosocial burden, and they argue that the hybrid modeling strategy preserves predictive accuracy while maintaining transparency and clinical interpretability.

Outcomes and Implications

This research is important because depression in advanced CKM patients is frequently underrecognized despite its strong association with poorer quality of life. By relying on routinely collected clinical measures rather than specialized psychiatric testing or advanced biomarkers, the model provides an objective approach for early risk testing in both primary care and specialty settings. The inclusion of a nomogram makes this test more usable clinically, allowing clinicians to translate abstract risk estimates into individualized assessments that could prompt mental health screening, referral, or preventive interventions on time. Although the authors do not suggest immediate changes to treatment guidelines, they position the model as a short-term decision support tool that could be integrated into electronic health record systems, with future clinical implementation depending on prospective validation and potential augmentation with proteomic or imaging biomarkers to further refine predictive accuracy.