Deep learning for depression prediction in older adults: A retrospective cohort study from CHARLS (2011-2020) with independent cohort validation in CLHLS (2008-2018).

Journal: Journal of affective disorders
Published Date: Jan 17, 2026

Abstract

BACKGROUND: Geriatric depression is highly prevalent yet under-recognized, severely affecting older adults' quality of life and social functioning. There is an urgent need for individualized early prediction tools. While deep learning (DL) holds potential, its systematic application in this domain remains underdeveloped. METHODS: Utilizing five waves (2011-2020) of the China Health and Retirement Longitudinal Study (CHARLS) comprising 2781 older adults aged >60 years, we developed and validated a DL model-the Dual-Attention Residual Network (DARNet)-to predict outcomes for the final wave (2020) based on preceding waves. Key predictors were initially screened using LASSO regression, followed by comparison of DARNet against fourteen benchmark models and validation on an independent cohort from the Chinese Longitudinal Healthy Longevity Survey (CLHLS, 2008-2018). Finally, multidimensional interpretability analysis was conducted via the SHAP method. RESULTS: DARNet demonstrated superior predictive performance on the CHARLS (testing set: Accuracy 0.7384, F1-score 0.7306, AUROC 0.7876 (95%CI: 0.7323-0.8377), AUPRC 0.8021 (95%CI: 0.7343-0.8658)), significantly outperforming benchmark models. Additionally, the model achieved robust performance on the CLHLS independent validation set (AUROC: 0.7440 95%CI: 0.6784-0.7993). SHAP analysis identified sleep duration, age, self-rated health status, and physical pain as core factors, and revealed gender- and age-related heterogeneity in depression risk profiles. CONCLUSION: DARNet effectively models the dynamic trajectory of geriatric depression in the Chinese population, combining excellent predictive performance with interpretability. It enables identification of high-risk individuals and guides stratified interventions, offering a practical tool for intelligent depression risk monitoring. Future integration of multi-center clinical data from diverse populations would enhance its real-world application value.

Authors

  • Qiuyu Long
    Department of Big Data Management and Application, Nanjing University of Chinese Medicine, Nanjing, Jiangsu 210023, China.
  • Junyan Li
    Department of Breast and Thyroid Surgery, Sichuan Provincial Hospital for Women and Children (Affiliated Women and Children's Hospital of Chengdu Medical College), Chengdu, China.
  • Nan Zhao
    CAS Key Laboratory of Behavioral Science, Institute of Psychology, Chinese Academy of Sciences, Beijing, China.
  • Renwei Guo
    Department of Cardiovascular Medicine, Fenyang Hospital Affiliated to Shanxi Medical University, Fenyang, 032200, Shanxi, China.
  • Shaoliang Tang
    School of Medical Imaging, Fujian Medical University, Fuzhou, China.

Keywords

No keywords available for this article.

External Resources

Popular Topics
Recent Journals