Predicting Response to Tocilizumab Monotherapy in Rheumatoid Arthritis: A Real-world Data Analysis Using Machine...

Area of Science:

• Rheumatology
• Clinical Pharmacology
• Data Science in Medicine

Background:

• Tocilizumab (TCZ) demonstrates comparable efficacy in rheumatoid arthritis (RA) whether used as monotherapy or in combination therapies, as evidenced by randomized controlled trials (RCTs).
• A predictive score for TCZ monotherapy (TCZm) remission was previously developed using RCT data.

Purpose of the Study:

• To externally validate the TCZm remission prediction score using real-world data (RWD).
• To compare the performance of prediction models in RWD versus RCTs and explore methods to enhance predictive accuracy.

Main Methods:

• Patients from the Corrona RA registry on TCZm (n=452) were identified and matched to patients from 4 RCTs (n=853).
• Remission status was assessed at 24 weeks. Prediction models were compared using logistic regression and random forest, with and without an extended variable set including other biologic disease-modifying antirheumatic drug monotherapies (bDMARDm).

Main Results:

• Remission rates in RWD (12%) were comparable to RCTs (15%).
• The prediction score derived from RCTs showed good discrimination in RWD (AUROC 0.69).
• Model performance improved in RWD when retrained on an extended variable set (AUROC 0.76).

Conclusions:

• Remission prediction scores developed in RCTs are applicable to RWD for TCZ monotherapy in RA.
• Retraining prediction models using RWD and expanded variable sets can significantly enhance the accuracy of predicting RA remission.