Predicting EQ-5D full health state in systemic lupus erythematosus using machine learning algorithms.

Area of Science:

• Rheumatology
• Health Outcomes Research
• Machine Learning in Medicine

Background:

• Systemic lupus erythematosus (SLE) significantly impacts patient quality of life.
• Understanding factors influencing perceived health states is crucial for patient management.
• EuroQol 5-Dimensions (EQ-5D) is a key patient-reported outcome measure.

Purpose of the Study:

• To identify demographic, clinical, and patient-reported factors associated with achieving full health state (FHS) in SLE patients.
• To utilize machine learning algorithms for robust prediction of FHS before and after intervention.

Main Methods:

• Post hoc analysis of two Phase 3 belimumab trials (BLISS-52, BLISS-76).
• Applied Monte Carlo Feature Selection, Support Vector Machine (SVM), LASSO, Neural Network (NNet), and Logistic Regression (LR).
• Evaluated models for linear and non-linear relationships, interpretability, and robustness.

Main Results:

• 12.9% of 1642 SLE patients reported FHS at baseline, increasing to 23.1% at week 52.
• Key predictors included age, sex, ancestry, disease activity (cSLEDAI-2K, SELENA-SLEDAI PGA), urine protein:creatinine ratio (UPCR), and baseline EQ-5D-3L index.
• Models showed good performance (AUC up to 0.77) with high negative predictive values (0.88-0.94).

Conclusions:

• Machine learning identified older age, female sex, non-Asian ancestry, high disease activity, and low UPCR as associated with lack of FHS.
• High baseline EQ-5D-3L index scores were the strongest predictor for achieving FHS at week 52.
• These findings enhance understanding of FHS determinants in SLE patients undergoing treatment.