Prognostic Modeling and Prevention of Diabetes Using Machine Learning Technique

Area of Science:

• Biostatistics
• Machine Learning
• Epidemiology

Background:

• Diabetes risk stratification is crucial for targeted interventions.
• The Framingham Diabetes Risk Scoring Model (FDRSM) is a widely used prognostic tool.
• Validating prognostic models with advanced machine learning techniques is an ongoing area of research.

Purpose of the Study:

• To explore the efficacy of a Hidden Markov Model (HMM) in predicting 8-year diabetes risk.
• To validate the performance of the FDRSM using HMM.
• To assess if HMM can effectively identify individuals at high risk for developing diabetes.

Main Methods:

• Utilized Electronic Medical Record (EMR) data from 172,168 primary care patients.
• Applied a Hidden Markov Model (HMM) to derive 8-year diabetes risk.
• Analyzed a subset of 911 individuals with complete risk factor and follow-up data.

Main Results:

• The HMM achieved an Area Under the Receiver Operating Characteristic Curve (AROC) of 86.9%.
• This AROC surpasses previously reported FDRSM validation AROCs (78.6% in Canadian population, 85% in Framingham study).
• The HMM demonstrated superior discrimination capability compared to FDRSM validations.

Conclusions:

• Hidden Markov Models (HMM) are effective in predicting 8-year diabetes risk.
• HMM shows superior performance in identifying individuals at increased risk for diabetes.
• This machine learning approach offers a valuable tool for diabetes risk stratification.