Predicting and Ranking Diabetic Ketoacidosis Risk Among Youth with Type 1 Diabetes with a Clinic-to-Clinic Transferrable Machine Learning Model
Authors' conclusion
Our DKA risk model effectively predicts youths' relative risk of experiencing hospitalization for DKA and is readily deployable to other diabetes centers that map diabetes data to the T1DX-QI schema.
Study context
- Population
- Youth with T1D, tertiary care pediatric diabetes clinic network, Midwest USA; data September 1, 2017 to November 1, 2022
- Participants
- 3,097 participants24,638 clinical encounters
- Outcome
- DKA-related hospitalization or emergency care
- Follow-up
- 6-month horizon
Methodological detailsmachine learning, ensemble of gradient-boosted, tree-based models, out-of-sample testing, single clinic network, 15 features chosen from electronic health record (EHR) data mapped to the T1DX-QI data schema
- Machine learning
- Ensemble of gradient-boosted, tree-based models
- Out-of-sample testing
- Single clinic network
- 15 features chosen from EHR data mapped to the T1DX-QI data schema
Performance
As reported in the abstract
Average precision for top-ranked highest-risk youth (out-of-sample testing set)
- Average precision, top 10
- 0.96
- Average precision, top 25
- 0.81
- Average precision, top 50
- 0.75
- Average precision, top 100
- 0.70
- Lift of the model (vs random selection), top 100
- 19
Further findings
Top predictive features
- Top three features: average time between diabetic ketoacidosis (DKA) episodes, time since last DKA episode, T1D duration
Original publication
Predicting and Ranking Diabetic Ketoacidosis Risk Among Youth with Type 1 Diabetes with a Clinic-to-Clinic Transferrable Machine Learning Model. Diabetes Technol Ther. Published 6 January 2025. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →