Predicting Diabetes in Canadian Adults Using Machine Learning.

Area of Science:

• Public Health
• Health Informatics
• Machine Learning in Medicine

Background:

• Rising diabetes rates increase healthcare costs and complications.
• Undiagnosed diabetes cases are prevalent, necessitating early detection.
• Accurate diagnosis is vital for mitigating disease progression.

Purpose of the Study:

• To predict diabetes prevalence in Canadian adults using machine learning.
• To leverage electronic medical record data for diabetes risk assessment.
• To identify key clinical factors for early diabetes diagnosis.

Main Methods:

• Utilized the Canadian Primary Care Sentinel Surveillance Network (CPCSSN) database.
• Developed and validated seven machine learning classification models.
• Trained models on clinical patient characteristics predictive of diabetes.

Main Results:

• The XGBoost model demonstrated the highest performance with an Area Under the Curve (AUC) of 92%.
• Key predictors identified include HbA1c level, LDL cholesterol, and hypertension medication.
• The study successfully predicted diabetes likelihood using primary care data.

Conclusions:

• Machine learning models can enhance early diabetes diagnosis in primary care settings.
• Identifying critical patient characteristics aids targeted interventions and public health planning.
• This approach can help reduce healthcare system burdens associated with diabetes.