O-52: External validation of the risk index for diabetic ketoacidosis (RI-DKA) in the T1D exchange quality improvement collaborative
D D Schwartz, S Rompicherla, D DeSalvo, N Rioles, S Dei-Tutu, S LyonsAbstract
Introduction
The Risk Index for Diabetic Ketoacidosis (RI-DKA) is a clinical prediction model that has been shown to reliably identify children and adolescents with type 1 diabetes (T1D) at increased risk for DKA in a single-center internal validation analysis. However, it is unknown whether the model generalizes to the broader population of individuals with T1D in the U.S.
Objectives
The goal of the present study was to validate the RI-DKA using data from T1D Exchange Quality Improvement Collaborative (T1DX-QI) and population health network, a consortium of over 60 pediatric and adult diabetes centers across the U.S.
Methods
This was a retrospective multicenter external validation study examining predictive validity of the RI-DKA in 57,408 individuals with T1D in the T1DX-QI cohort. Predictor weightings for the three RI-DKA risk factors (most recent HbA1C, health insurance [public/self-pay vs private], and prior DKA) were determined using multivariable logistic regression, and positive likelihood ratios were used to determine category cutoffs for low, moderate, high, and very high-risk groups. Data from the preceding two years was used to predict DKA events in the subsequent year (2024-2025) within each risk group. Model performance was evaluated using area under the ROC curve (AUC).
Results
All three risk factors remained highly predictive of DKA (p < 0.001) in the T1DX-QI cohort. Odds ratios were 80.15 (95% CI, 72.08 - 89.23) for prior DKA, 2.32 (95% CI, 2.16-2.49) for health insurance, and 1.028 (95% CI, 1.027-1.029) for each level of HbA1c. Model validation showed good discrimination between individuals at different risk levels (AUC = 0.824, 95% CI 0.815−0.832; Figure 1).
Conclusions
The RI-DKA showed strong predictive performance in a large multicenter sample, indicating that the risk model generalizes across much of the U.S. T1D population. The simple three-factor RI-DKA can therefore be used nationally to support efforts at preventing DKA.