Advertisement

AI Model Predicts Depression in Patients with COPD, Study Suggests

Published on: 

XGBoost displayed a high rate of accuracy, sensitivity, and precision, indicating its effectiveness at highlighting depressive symptoms before diagnosis.

Investigators have successfully developed an artificial intelligence (AI) system to help clinicians predict signs of depression in patients with chronic obstructive pulmonary disease (COPD).1

Depression, anxiety, and panic disorders frequently occur in COPD and can result in increased hospital admissions and longer stays, as well as overall worsened quality of life and, in some cases, premature death. According to previous studies, mental state plays a substantial role in the progression and diagnosis of COPD.2

While prediction models for depression already exist, these are often limited in both scope of symptoms and sample size during development. Additionally, these models are typically constructed using statistical methods such as logistic regression analysis; this prevents these models from operating in the complex linear relationships of mental health.1

“Our study fills a gap in the literature on machine learning’s role in linking COPD and depression risk,” wrote Xuanna Zhao, department of respiratory and critical care medicine, Affiliated Hospital of Guangdong Medical University, and colleagues. “While the comorbidity is well-known, traditional predictive methods struggle to capture it. The incorporation of machine learning algorithms…not only enhances predictive accuracy but also broadens the multidimensional understanding of underlying risk factors.”1

Zhao and colleagues collected data from the 2018 China Health and Retirement Longitudinal Study (CHARLS) database. Depressive symptoms were evaluated using the CESD-10 scale, with 10 items rated from 0 to 3, reflecting a range from “none” to “almost every day”. Higher scores on the total scale of 0 to 30 indicate more severe symptoms, with a score ≥10 classified as clinically significant depression.1

A total of 18,230 patients were initially identified in the database. After restricting age range to 45-85 years and excluding patients with a sleep duration of ≤1 hour or ≥15 hours, a final total of 2921 individuals diagnosed with COPD were included. Of these, 1451 exhibited depressive symptoms while 1470 did not. Significant differences were noted across demographic characteristics, comorbidities, and health status between participants with and without depression.1

The team divided the participant data into a training set (70%) and a test set (30%). Data processing and model construction were performed on the training set. A total of 11 variables were selected, followed by 10-fold cross-validation. Final selected features included gender, self-perceived health status, arthritis, kidney disease, digestive disease, life satisfaction, disability, pain, sleep time, history of falls, and Activities of Daily Living (ADL) score.1

Six machine learning models were utilized to predict depression occurrence among all patients with COPD. Of these, the algorithm XGBoost exhibited the most proficient and robust performance, with an area under the receiver operating curve (AUROC) of .811 (95% CI, .79-.829), an accuracy of 78.91%, a sensitivity of 77.31%, and a precision of 79.74%. It also achieved a specificity of 80.51% and an F1 score of 78.5%.1

In the time series validation set, XGBoost exhibited the highest accuracy (70.63%), sensitivity (59.05%), and F1 score (63.17%), as well as satisfactory precision (67.92%), specificity (79.25%), and an AUROC of .748 (95% CI, 0.716-0.779). Investigators noted that these data highlight its superior generalizability and stability across various populations.1

These findings indicate that AI models like XGBoost can prevent the development of a vicious cycle between COPD and depression, as well as enhance overall outcomes. Zhao and colleagues suggest the implementation of such models on a cloud server will facilitate quick intervention, such as medication, psychotherapy, and lifestyle modifications.1

“By deploying online prediction tools and integrating them into clinical practice, healthcare providers can efficiently and accurately assess whether COPD patients are at risk of depression and implement appropriate interventions and treatments,” they wrote.1

References
  1. Zhao X, Wang Y, Li J, et al. A machine-learning-derived online prediction model for depression risk in COPD patients: A retrospective cohort study from CHARLS. J Affect Disord. 2025;377:284-293. doi:10.1016/j.jad.2025.02.063
  2. Yayan J, Rasche K. Risk factors for depression in patients with chronic obstructive pulmonary disease. Respiratory Physiology & Neurobiology. 2023;315:104110. doi:10.1016/j.resp.2023.104110

Advertisement
Advertisement