Development and validation of algorithms to identify newly diagnosed type 1 and type 2 diabetes in pediatric population using electronic medical records and claims data.

PURPOSE: To develop and validate algorithms to classify diabetes type in newly diagnosed pediatric patients with DM.

METHOD: Data from the United States Department of Defense health system were used to identify patients aged 10 to 18 years with incident DM. Two independent sets of 200 children were randomly sampled for algorithm development and validation. Algorithms were developed based on clinical insight, published literature, and quantitative approaches. The actual DM type was ascertained via chart review. Finally, the sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were evaluated.

RESULTS

Among the 400 patients, mean age was 14.2 (±2.5 years), and 50% were female. The best performing algorithms were based on data available in claims. They consisted of several logical expressions based on one predictor or more, which classified patients by use of glucose-lowering drugs or testing, DM ICD-9 diagnosis codes, and comorbidities. The best performing T2DM and T1DM algorithms achieved 90% and 98% sensitivity, 95% and 95% specificity, 87% and 98% PPV, and 96% and 96% NPV, respectively.

CONCLUSIONS

Our results suggest that claims algorithms can accurately identify newly diagnosed T1DM and T2DM pediatric patients, which can facilitate large database studies in children with T1DM and T2DM. However, external validation in other data sources is needed.