Related Experiment Video
Updated: Jul 16, 2026
A Machine Learning Approach to Design an Efficient Selective Screening of Mild Cognitive Impairment
Published on: January 11, 2020
Construction and verification of a prediction model for sleep disorders in older patients with coronary heart disease
Dali Dong1, Xiang Peng2, Siling Tan2
1Department of Medical Affairs, the Second Affiliated Hospital of Hunan University of Chinese Medicine, Changsha, Hunan, China.
Objective:
To construct and validate a predictive model for sleep disorders in older adults with coronary heart disease (CHD) using machine learning algorithms.
Methods:
A cross-sectional survey design was employed to collect clinical data from 338 older adult patients with CHD who were admitted to our hospital between January 2023 and December 2024. Participants were divided into two groups based on the presence of sleep disorders (PSQI > 7). Data were collected from questionnaires and electronic medical records, and subjects were randomly divided into training (241 cases) and validation (97 cases) sets in a 7:3 ratio. Least absolute shrinkage and selection operator (LASSO) regression was used to screen risk factors from the training set, and Logistic regression, random forest (RF), and gradient boosting machine (GBM) models were built. Receiver operating characteristic (ROC) and calibration curves evaluated model performance, with a temporal validation cohort of 152 older CHD patients from our hospital between January and October 2025.
Results:
Among 338 patients, 82 (24.26%) had sleep disorders, and 58 (24.07%) in the training group.. In the training set, LASSO identified five risk factors: sex, duration of CHD, chronic gastritis, anxiety, and depression. The three models showed good predictive performance with different strengths. In the training set, the RF model achieved the highest AUC (0.839), accuracy (0.766), and F1 score (0.818); both RF and Logistic regression achieved the highest sensitivity (0.812). In the validation set, the GBM model achieved the highest AUC (0.838) and accuracy (0.733); the Logistic regression model demonstrated the best specificity (0.875) and F1 score (0.825). External validation of the Logistic regression model showed an AUC of 0.785 (95% CI: 0.694-0.876), sensitivity of 0.798, and specificity of 0.658, with calibration curves showing good consistency.
Conclusion:
The Logistic regression, GBM, and RF models based on sex, duration of CHD, chronic gastritis, anxiety, and depression all demonstrated good predictive performance.
