Literature‐informed ensemble machine learning for three‐year diabetic kidney disease risk prediction in type 2 diabetes: Development, validation, and deployment of the Prince Sultan Military Medical City (PSMMC) NephraRisk model
Authors' conclusion
Our developed and demonstrated model provided accurate and well-fair DKD/DN risk prediction with excellent calibration, allowing for better decision making with deployment as a web-based research tool and framework for future prospective clinical validation. Further validation and testing are warranted from different centres and healthcare systems to increase confidence and dissemination of our model findings for better utilisation purposes in the future.
Study context
- Population
- Adults with T2D, Prince Sultan Military Medical City (PSMMC) registry, Riyadh, Saudi Arabia, 2019–2024
- Participants
- 18,742 adults
- Outcome
- Incident/progressive diabetic kidney disease (DKD)/diabetic nephropathy
- Follow-up
- 3 years
Methodological detailssingle-centre registry, temporal patient-level splitting, stacked ensemble machine learning (LightGBM + CoxBoost), Bayesian multiple imputation by chained equations (MICE) imputation with external study priors, Deployed as an interactive web-based application
- Single-centre registry
- Temporal patient-level splitting
- Stacked ensemble machine learning (LightGBM + CoxBoost)
- Bayesian MICE imputation with external study priors
- Deployed as an interactive web-based application
Performance
As reported in the abstract
Discrimination: AUROC
0.852
95% CI 0.847–0.857
Calibration slope
0.98
Calibration intercept
−0.012
Further findings
Decision curve analysis
- Net benefit vs treat-all strategies at 10% threshold (events prevented per 1000 patients)
- +22
Bootstrap validation
- C-statistic optimism
- 0.005
Algorithmic fairness across demographic subgroups
- No algorithmic bias detected; maximum |Δ-AUROC|
- 0.010
Prior sensitivity analysis
- AUROC variation
- ≤0.008
Original publication
Composed by AI from the published abstract and checked against it. How Briefs are made →