Objective Successfully stopping or reducing treatment for patients with rheumatoid arthritis (RA) in low disease activity (LDA) may improve cost-effectiveness of care. We evaluated the multi-biomarker disease activity (MBDA) score as a predictor of disease relapse after discontinuation of TNF inhibitor (TNFi) treatment. Methods 439 RA patients who were randomized to stop TNFi treatment in the POET study were analyzed post-hoc. Three indicators of disease relapse were assessed over 12 months: 1) restarting TNFi treatment, 2) escalation of any DMARD therapy and 3) physician-reported flare. MBDA score was assessed at baseline. Associations between MBDA score and disease relapse were examined using univariate analysis and multivariate logistic regression. Results At baseline, 50.1%, 35.3% and 14.6% of patients had low (<30), moderate (30−44) or high (>44) MBDA scores. Within 12 months, 49.9% of patients had restarted TNFi medication, 59.0% had escalation of any DMARD and 57.2% had 1 physician-reported flare. MBDA score was associated with each indicator of relapse. At least one indicator of relapse was observed in 59.5%, 68.4% and 81.3% of patients with low, moderate or high MBDA scores, respectively (P = 0.004). Adjusted for baseline DAS28-ESR, disease duration, BMI and erosions, high MBDA scores were associated with increased risk for restarting TNFi treatment (OR = 1.85, 95% CI 1.00–3.40), DMARD escalation (OR = 1.99, 95% CI 1.01–3.94) and physician-reported flare (OR = 2.00, 95% 1.06–3.77). Conclusion For RA patients with stable LDA who stopped TNFi, a high baseline MBDA score was independently predictive of disease relapse within 12 months. The MBDA score may be useful for identifying patients at risk of relapse after TNFi discontinuation.