Background: Surveillance of incident diabetes among youth is essential to identify health disparities and inform health care planning.
Objective: To evaluate the validity of a large-scale algorithmic approach to ascertaining diabetes incidence in Colorado youth < 20 years of age in 2018, by comparing it against manual chart review.
Methods: The algorithm for identifying incident diabetes status was based on diagnostic codes, medications, and laboratory data. Manual chart review was the gold standard. We calculated the sensitivity, sensitivity, positive predicted value (PPV) and negative predictive value (NPV).
Results:A total of 3779 youth were identified as youth onset diabetes cases by both methodologies between 2002 and 2018. To identify new onset (incident diabetes) the algorithmic approach was 96.4% sensitive, and 99.4% specific compared to chart review while the PPV and NPV were 95.1% and 99.6%, respectively. For incident type 1 diabetes, the algorithmic approach was 94.2% sensitive and 99.3% specific and the PPV and NPV were 93.7% and 99.4%, respectively. For incident type 2, the algorithmic approach was 52.9% sensitive and 99.9% specific and the PPV and NPV were 85.7% and 99.6%, respectively.
Conclusion: The algorithmic approach had high validity in identifying incident cases of type 1 diabetes among youth but performed poorly for type 2.
T.L. Crume: None. A. Bellatorre: None. S. Burgett: None. R.B. Conway: None. T. Anderson: None. B.A. Shiferaw: None. D. Dabelea: None.
Centers for Disease Control and Prevention (U18DP006517; U18DP006518)
