Development and Validation of Inpatient Hypoglycemia Models Centered Around the Insulin Ordering Process
Authors' conclusion
Hypoglycemia within the next 24 hours can be predicted at the time an insulin order is placed, providing an opportunity to integrate decision support into the medication ordering process to make insulin therapy safer.
Study context
- Population
- Adult patients admitted in 2019 to non-ICU floors of a large teaching hospital with orders for subcutaneous insulin
- Participants
- 21,052 insulin orders
- Comparison
- Logistic regression vs random forest vs XGBoost for predicting hypoglycaemia at insulin ordering
- Outcome
- Hypoglycaemia: BG <70 mg/dL after ordering insulin
- Follow-up
- Within 24 hours
Methodological detailssingle-centre, retrospective
- Single-centre
- Retrospective
Performance
As reported in the abstract
Model performance for predicting hypoglycaemia
| Test or model | AUC | Recall at precision 0.30 |
|---|---|---|
| Logistic regression | 0.81 | 0.44 |
| Random forest | 0.80 | 0.49 |
| XGBoost | 0.79 | 0.32 |
AUC
Logistic regression0.81
Random forest0.80
XGBoost0.79
Recall at precision 0.30
Logistic regression0.44
Random forest0.49
XGBoost0.32
Further findings
Hypoglycaemic event after insulin order · within 24 hours
- Insulin orders followed by hypoglycaemic event
- 1839 (9%)
Predictor importance
- Most significant predictor: lowest blood glucose (BG) in the 24 hours preceding the order
- Insulin-order predictors useful but less important than the patient's BG history over time
Original publication
Development and Validation of Inpatient Hypoglycemia Models Centered Around the Insulin Ordering Process. J Diabetes Sci Technol. Published 1 September 2022. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →