Predicting Diabetes in Canadian Adults Using Machine Learning

Rising diabetes rates have led to increased health-care costs and health complications. An estimated half of diabetes cases remain undiagnosed. Early and accurate diagnosis is crucial to mitigate disease progression and associated risks. This study addresses the challenge of predicting diabetes prevalence in Canadian adults by employing machine learning (ML) techniques to primary care data. We leveraged the Canadian Primary Care Sentinel Surveillance Network (CPCSSN), Canadas premier multi-disease electronic medical record surveillance system, and developed and tuned seven ML classification models to predict the likelihood of diabetes. The models were tested and validated, focusing on clinical patient characteristics influential in predicting diabetes. We found XGBoost performed best out of all the models, with an AUC of 92%. The most important features contributing to model prediction were HbA1c, LDL, and hypertension medication. Our research aims to aid healthcare professionals in early diagnosis and to identify key characteristics for targeted interventions. This study contributes to an understanding of how ML can enhance public health planning and reduce healthcare system burdens.