Predicting mycophenolic acid exposure in paediatric lupus nephritis: a real-world study across three clinical scenarios
Authors' conclusion
This study establishes the first ML-driven framework for personalized MPA dosing in paediatric LN, addressing TDM limitations and enhancing precision medicine.
Study context
- Population
- Children with childhood-onset SLE and lupus nephritis (LN) on mycophenolic acid (MPA)1,376 follow-ups
- Participants
- 154 children
- Comparison
- Machine learning (ML) algorithms for predicting MPA-AUC across three clinical scenarios
- Outcome
- Prediction of MPA-AUC: therapeutic window attainment, precise AUC estimation and post-dose-adjustment exposure
Methodological detailsreal-world, machine learning, SHAP interpretation, A web-based tool was developed to support clinical decisions
- Real-world
- Machine learning
- SHAP interpretation
- A web-based tool was developed to support clinical decisions
Main finding
As reported in the abstract
Performance of ML models predicting MPA-AUC across three scenarios (1: therapeutic window attainment; 2: precise AUC estimation; 3: post-dose-adjustment exposure)
| Group | Scenario 1 AUC | Scenario 1 Precision | Scenario 1 Recall | Scenario 1 F1 | Scenario 2 R2 (clinical indicators + C0.5, C1.5, C4) | Scenario 2 mean absolute error (MAE) | Scenario 2 root mean square error (RMSE) | Scenario 3 R2 | Scenario 3 MAE |
|---|---|---|---|---|---|---|---|---|---|
| LightGBM | 0.67 | 0.64 | 0.66 | 0.65 | |||||
| Random Forest (RF) | 0.84 | 3.97 | 5.97 | ||||||
| RF | 0.51 | 20.57 |
Scenario 1 AUC
LightGBM0.67
Random Forest (RF)— not stated in the abstract
RF— not stated in the abstract
Scenario 1 Precision
LightGBM0.64
Random Forest (RF)— not stated in the abstract
RF— not stated in the abstract
Scenario 1 Recall
LightGBM0.66
Random Forest (RF)— not stated in the abstract
RF— not stated in the abstract
Scenario 1 F1
LightGBM0.65
Random Forest (RF)— not stated in the abstract
RF— not stated in the abstract
Scenario 2 R2 (clinical indicators + C0.5, C1.5, C4)
LightGBM— not stated in the abstract
Random Forest (RF)0.84
RF— not stated in the abstract
Scenario 2 mean absolute error (MAE)
LightGBM— not stated in the abstract
Random Forest (RF)3.97
RF— not stated in the abstract
Scenario 2 root mean square error (RMSE)
LightGBM— not stated in the abstract
Random Forest (RF)5.97
RF— not stated in the abstract
Scenario 3 R2
LightGBM— not stated in the abstract
Random Forest (RF)— not stated in the abstract
RF0.51
Scenario 3 MAE
LightGBM— not stated in the abstract
Random Forest (RF)— not stated in the abstract
RF20.57
Further findings
Key predictors (SHAP analysis)
- Dose, weight, renal function and concentration timepoints were key predictors
Original publication
Predicting mycophenolic acid exposure in paediatric lupus nephritis: a real-world study across three clinical scenarios. Rheumatology (Oxford). Published 17 February 2026. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →