A Probabilistic Case-finding Algorithm for Chronic Disease Surveillance

We developed and validated a multivariable probabilistic case-detection model to detect known cases of diabetes mellitus (DM) using clinical and demographic data. We applied our method to a cohort of older adult residents of the region of Sherbrooke, Quebec. Predictors were added to a logistic regression model and internally validated using a 2:1 split sample approach. Models were compared using measures goodness of fit, discrimination and accuracy. The best model incorporated all predictors into the model: male sex, age, at least one hospitalization, physician visit and drug dispensed for diabetes.