OBJECTIVES: Diabetic ketoacidosis (DKA) at onset of type 1 diabetes (T1D) is life-threatening, expensive, and affects 60% of children diagnosed with T1D in Colorado. Current evidence on DKA is limited to single centers. All-payer claims databases (APCDs) could be used to estimate DKA events at T1D diagnosis to inform T1D screening and monitoring efforts. Our objective was to develop an approach to estimate DKA rates using the Colorado APCD against validated electronic medical record (EMR) data.

METHODS: The Barbara Davis Center for Diabetes validated EMR data includes over 90% of all Colorado children diagnosed with T1D with distinction between DKA events at diagnosis. We matched medical records to the all-payer claims data in Colorado. Gestational diabetes and individuals ≥18 years of age were excluded. Advancements to existing T1D algorithms included criteria associated with emergency room (ER), inpatient (IP) and outpatient (OP) visits associated with T1D and DKA codes in addition to inpatient insulin use and lab orders. We estimated combinations of criteria against patients’ registry records to maximize sensitivity, specificity, and positive predictive value (PPV).

RESULTS: After applying inclusion and exclusion criteria among N=1,407 individuals with both claims and EMR records, a final sample of N=728 included continuously insured children residing in Colorado with a diagnosis of T1D between 2014 – 2019. Mean age was 16 years and 56% had DKA at diagnosis using EMR alone. An algorithm with ≥ 2 criteria (IP or ER visit with an associated T1D code, DKA code, or J-code for insulin use) optimized sensitivity (87.5%), specificity (60.3%), and PPV (73.8%). This algorithm resulted in a proportion with DKA at diagnosis of 66%.

CONCLUSIONS: APCDs can identify T1D cases and distinguish between DKA events with varying levels of bias compared to EMR using encounters with T1D and DKA codes in combination with inpatient insulin.