The Development and Validation of Multivariable Electronic Health Record-Based Models to Predict Diabetic Ketoacidosis-Related Hospitalizations for Adults with Type 1 Diabetes
Authors' conclusion
We developed EHR-based logistic regression models that effectively predict DKA-related hospitalizations in adults with T1D. Future work will enhance model performance by incorporating additional features and applying advanced machine learning methods.
Study context
- Population
- Adults with T1D treated at an academic health system in the United States, January 1, 2017 to April 30, 2023667 (8.6%) with ≥1 diabetic ketoacidosis (DKA) event
- Participants
- 7,798 adults
- Comparison
- 90-day model vs 180-day model
- Outcome
- DKA-related hospitalizations
- Follow-up
- 90 and 180 days
Methodological detailslogistic regression, discovery and prospective validation cohorts, single academic health system, 7 predictors: prior DKA, number of prior DKA events, mean time between DKA events, time since last DKA, most recent HbA1c, absent HbA1c in past 2 years, insurance type; Electronic health record (EHR) data from 2 years before index date
- Logistic regression
- Discovery and prospective validation cohorts
- Single academic health system
- 7 predictors: prior DKA, number of prior DKA events, mean time between DKA events, time since last DKA, most recent HbA1c, absent HbA1c in past 2 years, insurance type; EHR data from 2 years before index date
Performance
As reported in the abstract
Discrimination for DKA-related hospitalization
| Test or model | Value | Recall (top 5% risk) | Precision (top 5% risk) | Specificity (top 5% risk) |
|---|---|---|---|---|
| 90-day model | mean AUC 0.87 (SD ± 0.02) | 0.45 | 0.11 | 0.95 |
| 180-day model | mean AUC 0.84 (SD ± 0.02) | 0.42 | 0.17 | 0.96 |
Value
90-day modelmean AUC 0.87 (SD ± 0.02)
180-day modelmean AUC 0.84 (SD ± 0.02)
Recall (top 5% risk)
90-day model0.45
180-day model0.42
Precision (top 5% risk)
90-day model0.11
180-day model0.17
Specificity (top 5% risk)
90-day model0.95
180-day model0.96
Original publication
The Development and Validation of Multivariable Electronic Health Record-Based Models to Predict Diabetic Ketoacidosis-Related Hospitalizations for Adults with Type 1 Diabetes. Diabetes Technol Ther. Published 7 November 2025. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →