Context: The Chronic Care Management model specifies that care should be panel-based and proactive. We used primary care Electronic Medical Record (EMR) data from the National Diabetes Repository; results should be applicable across Canada.
Objective: Preliminary results using Machine Learning (ML) approaches indicate the potential to identify patients with Type 2 diabetes that miss multiple elements of care at recommended frequencies.
Dataset: Data from Primary Care Ontario Practice-Based Learning and Research Network (POPLAR) will be used for prediction and intensify outreach to patients predicted to be less likely to receive recommended services to manage diabetes.
Outcome Measures: Outcome scores were derived using presence or absence of recommended number of visits, blood pressures, weights, HbA1c, lipid panel, ACR, eGFR, statin prescription in each year of interest, from January 1 2016 to December 31st 2019. This score was further categorized as patients (i) not receiving adequate diabetes care (<= 3); (ii) receiving adequate diabetes care (>=4). Scores for service use were generated; a higher score indicated that a patient received more of the basket of recommended services for diabetes. The maximum score was 8, with the mean score of 4.96; median value of 5 and most common score of 7. We used a collection of ML algorithms to predict the future diabetes care.
Results: Preliminary results indicated the potential for accurately identifying patients with Type 2 diabetes that miss multiple elements of care at recommended frequencies.