The accuracy of using integrated electronic health care data to identify patients with undiagnosed diabetes mellitus

Rationale, aims and objectives Diabetes mellitus is a growing health and economic burden. Identification of patients with unrecognized diabetes, or those at high risk for diabetes, provides an opportunity for timely intervention. This study assessed the accuracy of using electronic health care data to identify patients with undiagnosed diabetes. Methods The study was conducted at a tertiary-care teaching facility in Ottawa, Canada. The study cohort was a stratified random sample of hospitalizations between 1 January 2003 and 31 December 2008. We used diagnostic codes, pharmacy orders and serum glucose tests to classify patients into six groups: recognized diabetes (a diabetes diagnostic code or any diabetes medication), probable diabetes (maximum glucose = 11.1 mmol L-1), possible diabetes (maximum glucose between 7.8 and 11.1 mmol L-1), unlikely diabetes (maximum glucose between 6.0 and 7.8 mmol L-1), no diabetes (maximum glucose < 6.0 mmol L-1) and unknown diabetes status (no glucose test). We compared this electronic classification to a reference standard chart review performed by a blinded abstractor. Results A total of 500 hospitalizations were included. The prevalence of each diabetes group was: recognized 17%; probable 4%; possible 15%; unlikely 20%; none 15%; and unknown 29%. Our electronic algorithm correctly classified 88.8% (95% confidence interval 85.791.3) of hospitalizations (weighted-Kappa = 0.885; 95% confidence interval 0.8510.919). The sensitivity, specificity and positive predictive values of our algorithm for known diabetes was 0.842, 0.988 and 0.941, respectively. For patients at a high risk for diabetes (maximum glucose > 7.8 mmol L-1), the corresponding values were 0.921, 0.971 and 0.872. Conclusions Patients with diagnosed and undiagnosed diabetes can be accurately identified using electronic health care data.