Individualised treatment effect prediction: the benefit of initial biological treatment in early rheumatoid arthritis

Abstract

OBJECTIVES: Initiating biologics upfront in early disease-modifying antirheumatic drugs (DMARDs)-naïve rheumatoid arthritis (RA) may benefit selected patients more than starting with methotrexate (MTX) monotherapy. Current guidelines recommend using prognostic markers to define high-risk groups for early biological treatment. Predictive models combining these markers could better identify individuals likely to benefit. This study aimed to develop and validate models predicting the benefit of starting tocilizumab (TCZ) plus MTX over MTX alone in early DMARD-naïve RA. METHODS: We used data from the FUNCTION trial (NCT01007435, n = 856) for model development and U-Act-Early (NCT01034137, n = 299) for validation. Models predicted 6-month time-averaged Clinical Disease Activity Index (CDAI) benefit of TCZ + MTX vs MTX alone using the following 2 approaches: (1) imputation of the unobserved counterfactual outcome, followed by prediction of the individual treatment effect; and (2) direct modelling with linear regression including treatment-covariate interactions. Model performance was evaluated via discrimination, calibration, and simulated clinical impact. RESULTS: Both models aligned with observed outcomes. In U-Act-Early, patients selected by the models for TCZ + MTX had an average CDAI reduction compared with MTX alone of 7 points, compared with 3 points in those not selected. The number needed to treat (NNT) to achieve an average CDAI ≤22 (below high disease activity) over 6 months was 12 for the full trial population and 5 in model-selected patients. CONCLUSIONS: Our prediction models identified early patients with RA more likely to benefit from initiating TCZ + MTX over MTX monotherapy, reducing the NNT to a clinically acceptable threshold of 5.