Early, unplanned readmission in diabetes inpatients: Comparing the utility and performance of machine learning and traditional prediction models to guide timely diabetes management team review
Authors' conclusion
EUR prediction models facilitate targeted, timely DMT review. Regularised regression models offer a feasible, practical approach for identifying high-risk patients in resource-limited hospital settings. Combining model-identified risk with abnormal glucose refines patient selection, optimising resource allocation.
Study context
- Population
- Inpatients with diabetes; models built on 14 routinely available, primarily early admission variables
- Comparison
- Classification tree vs logistic group lasso vs elastic net vs neural network for predicting early unplanned readmission (EUR)
- Outcome
- Prediction of early unplanned readmission (EUR); performance by AUC
Methodological detailsmachine learning vs traditional models
- Machine learning vs traditional models
Performance
As reported in the abstract
Group lasso, elastic net and neural network performed similarly, AUC test cohort
0.722-727
Group lasso, elastic net and neural network, AUC validation
0.653-0.667
Tree (outperformed by the other models), AUC test cohort
0.663
Tree, AUC validation
0.595
Further findings
Admissions identified as high risk at sensitivities of 25%, 50% and 75%
- High risk at 25% sensitivity
- 9%
- High risk at 50% sensitivity
- 21%-23%
- High risk at 75% sensitivity
- 41%-42%
Group lasso: Diabetes management team (DMT) review cohort when high risk combined with abnormal glucose (≥1 glucose <4 or >15 mmol/L)
- DMT review cohort at 25% sensitivity
- 4.9%
- DMT review cohort at 50% sensitivity
- 10.8%
- DMT review cohort at 75% sensitivity
- 19.2%
- About half of high-risk patients also had abnormal glucose
Original publication
Composed by AI from the published abstract and checked against it. How Briefs are made →