Prediction of ineffectiveness of biological drugs using machine learning and explainable AI methods: data from the Austrian Biological Registry Biologics Registry (BioReg)
Authors' conclusion
Ineffectiveness of biological drugs could be predicted with promising accuracy.
Study context
- Population
- Patients treated with bDMARDs in the Austrian Biologics Registry (BioReg), at least 100 treat-to-target courses per drug19 variables
- Participants
- 1,397 patients
- Comparison
- Abatacept vs Adalimumab vs Certolizumab vs Etanercept vs Tocilizumab
- Outcome
- Prediction of risk of ineffectiveness of each bDMARD (model quality by AUROC)
- Follow-up
- First 26 weeks
Methodological detailsmachine learning, cross-validation, hyperparameter optimization, explainable AI, registry-based
- Machine learning
- Cross-validation
- Hyperparameter optimization
- Explainable AI
- Registry-based
Performance
As reported in the abstract
AUROC of best model per drug for predicting ineffectiveness
Abatacept
0.66
Area under the receiver operating characteristic (AUROC) · 95% CI 0.54–0.78
Adalimumab
0.70
AUROC · 95% CI 0.68–0.74
Certolizumab
0.84
AUROC · 95% CI 0.79–0.89
Etanercept
0.68
AUROC · 95% CI 0.55–0.87
Tocilizumab
0.72
AUROC · 95% CI 0.69–0.77
Further findings
Explainable AI: risk-increasing and risk-reducing factors
- Most risk-increasing variable: Visual analytic scores (VAS) for abatacept and etanercept; GC co-therapy for adalimumab
- Dosage most important variable for certolizumab, associated with lower risk of non-response
- Some variables (gender, rheumatoid factor (RF)) showed opposite impacts depending on the bDMARD
Original publication
Prediction of ineffectiveness of biological drugs using machine learning and explainable AI methods: data from the Austrian Biological Registry BioReg. Arthritis Res Ther. Published 8 February 2024. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →