Justin Yan: Derivation and Internal Validation of a Clinical Prognostic Tool for Recurrent Emergency Visits for Hyperglycemia


Diabetes mellitus (DM) is an increasingly prevalent chronic disease, with estimates that up to 40% of adults may develop DM over their lifetime. Patients with poorly controlled DM often visit the ED for management of hyperglycemic episodes, including diabetic ketoacidosis (DKA) and hyperosmolar hyperglycemic syndrome (HHS).

A previous study by our group attempted to identify predictors of an unplanned recurrent ED visits for hyperglycemia within 30 days of initial presentation. However, this study had its limitations due to its retrospective nature and inability to capture certain patient characteristics that may have influenced the risk of a subsequent ED visit. The objectives of this study are to: 1) prospectively confirm previously reported risk factors identified and determine additional variables that may be predictive of adverse outcomes within 30 days of initial presentation, and 2) derive and internally validate a prognostic clinical risk tool to identify those at higher risk of adverse outcomes including repeat ED visits within 30 days of initial ED presentation.


This will be a multicentre prospective cohort study of consecutive adult patients with an ED diagnosis of hyperglycemia, DKA or HHS. Potential cases of patients will be identified from the local site ED patient tracking system. Trained research assistants will screen the charts of these cases, and all eligible patients will be followed for recurrent ED visits. Emergency physicians will confirm eligibility for enrolment and obtain informed consent for the research team to contact the patient for telephone follow-up.

Research assistants will then contact the participants enrolled in the study at 14 and 30 days via telephone for follow-up with respect to obtaining data regarding clinical outcomes. Research assistants will also follow these patients electronically to determine if they have any further ED visits, admissions or ICU admissions after their ED visit for hyperglycemia.

Distributions of relevant patient characteristics will be summarized using descriptive statistics and differences between groups will be assessed using chi-squared and t-test as appropriate. Patient characteristics that could contribute to the estimation of the probability of the outcomes of interest will be selected based on expert opinion, prior research, as well as regression analysis in the current study. We will perform an initial univariate, followed by a multivariate logistic regression analysis. Because some patients might have more than one ED visit during the study, our multivariable analysis will be conducted in a generalized estimating equations (GEE) framework with an auto-regression correlation structure for the residuals (to correct for individual-level clustering due to repeated observations on the same patient.

Impact on EM

We hypothesize that we will be able to identify potentially modifiable risk factors that may predict recurrent visits for hyperglycemia, such as specific patient demographics, comorbidities and lack of access to follow-up. The results of this project will assist clinicians to better identify these patients and enable them to intervene either medically or educationally to prevent subsequent visits to the ED. As a result, patients may have improved care, better glucose control, and be identified for closer follow-up with a family physician or diabetes specialist.