Development and validation of a machine learning-augmented algorithm for diabetes screening in community and primary...

Area of Science:

• Medical Informatics
• Public Health
• Machine Learning in Healthcare

Background:

• Diabetes screening is vital for reducing morbidity, mortality, and socioeconomic burdens.
• Machine learning (ML) offers enhanced predictive accuracy for disease detection.
• Developing ML-augmented models for diabetes screening in community and primary care is the focus.

Purpose of the Study:

• To develop and evaluate ML-augmented models for diabetes screening.
• To compare the performance of ML models against the New China Diabetes Risk Score (NCDRS).
• To assess predictive accuracy and cost-effectiveness in community and primary care settings.

Main Methods:

• Utilized data from 8425 participants in a population-based study in Hubei, China.
• Developed predictive models using seven ML algorithms, with non-laboratory features for community settings and laboratory features for primary care.
• Compared ML models (ML and ML+lab) against the NCDRS using AUC, auPR, and average detection costs.

Main Results:

• ML models demonstrated superior performance over NCDRS, with higher AUC (0.697) and auPR (0.303).
• The ML model achieved a 12.81% lower average detection cost than NCDRS at the same sensitivity.
• The ML+FPG model offered the lowest detection cost among ML+lab models.

Conclusions:

• ML-augmented models provide higher predictive accuracy and lower detection costs than NCDRS-based approaches.
• ML-augmented algorithms are a potential tool for effective diabetes screening in community and primary care.
• These findings support the integration of ML for improved diabetes detection and management.