Post-hoc analysis of RCTs and cohorts identifies predictors of remission in ACPA-positive early RA treated with abatacept plus methotrexate or methotrexate alone.

OBJECTIVE: To predict which early RA-patients achieve clinical remission on initial treatment with methotrexate monotherapy or with abatacept+methotrexate. METHODS: This study is a subanalysis of the AVERT and AVERT-2 randomized controlled trials, which were performed in ACPA-positive early RA-patients who received methotrexate monotherapy or abatacept + methotrexate. External model validation of patients on methotrexate monotherapy was performed in the observational Leiden Early Arthritis Clinic (EAC) cohort. Primary outcome was DAS28-CRP remission at 6 and 12 months follow-up. Prediction models were developed using logistic regression analysis. First, a model including clinical baseline variables only was estimated. Subsequently, it was assessed whether adding serological or imaging data, shared epitope or early DAS28 response improved model performance. RESULTS: In the methotrexate-monotherapy group (n=388), 27% and 39% of patients achieved DAS28-remission after 6 and 12-months. In the abatacept + methotrexate group (n=743) this was 43% and 53%. Baseline DAS28-CRP was predictive for clinical remission in all models. Optimism-adjusted model performance (AUROC) for DAS28-remission at 6 and 12-months was 0.66/0.65 in the methotrexate-group and 0.68/0.59 in the abatacept + methotrexate group. Adding baseline MRI-detected joint-inflammation, baseline serology or HLA-shared epitope alleles did not significantly improve model performance. Early DAS28-response did improve model performance. In the external validation cohort model performance was very similar. CONCLUSION: Determining which patients achieve clinical remission upon methotrexate or abatacept+methotrexate treatment remains challenging. Disease activity at disease presentation and early DAS response were the only consistent predictors for achieving clinical remission. Genetic, imaging and serology parameters did not improve model performance.