A bayesian network model for neurocognitive disorders digital screening in Chinese population: development and...

Area of Science:

• Medical Informatics
• Public Health
• Machine Learning

Background:

• Neurocognitive disorders (NCDs) pose a significant global health challenge, necessitating efficient screening methods.
• Early detection of NCDs improves patient quality of life and reduces healthcare expenditures.
• Current screening methods require optimization for large-scale application.

Purpose of the Study:

• To develop and validate a cost-effective and convenient classification model for screening NCDs.
• To leverage electronic health records for building a robust NCD screening tool.
• To enhance the efficiency of NCD screening in primary healthcare settings.

Main Methods:

• A Bayesian network classification model was constructed using electronic health record data (2015-2017).
• Variables were selected through univariate logistic regression and Bayesian Information Criterion optimization.
• Model performance was assessed using ROC curves, calibration curves, and decision curve analysis, benchmarked against logistic regression.

Main Results:

• The Bayesian network model demonstrated strong predictive discrimination with AUCs ranging from 0.800 to 0.849 across datasets.
• The model showed good calibration and clinical applicability, with robust performance under sensitivity analysis for missing data.
• Thirty-one demographic and clinical variables were incorporated, with eight directly linked to NCD prediction.

Conclusions:

• The developed Bayesian network model accurately predicts NCD risk in primary care using electronic health records.
• This model is suitable for large-scale NCD screening in adult populations within electronic health record systems.
• The model's robustness to missing data supports its potential for clinical decision-making.