Early divergence of treatment response trajectories to adalimumab or its biosimilar in active ankylosing spondylitis: consensus clustering analysis of a randomized controlled trial.

Background: Patients with active ankylosing spondylitis (AS) exhibit substantial heterogeneity in their clinical responses to tumor necrosis factor alpha inhibitors (TNFi). Consensus clustering, an unsupervised cluster discovery method, may identify AS subgroups with more homogeneous treatment response patterns to adalimumab (ADA), a widely prescribed TNFi. Methods: We performed longitudinal consensus clustering based on 8 repeated measurements of 10 core response variables in 438 patients with active AS enrolled in a 24-week phase III randomized controlled trial of ADA or its biosimilar, IBI303. Baseline characteristics and important endpoints-including the Assessment of SpondyloArthritis International Society (ASAS)-based and Ankylosing Spondylitis Disease Activity Score Inactive Disease (ASDAS)-based response criteria-were compared between the identified clusters. Predictive models of cluster membership reconstructed from data beyond week 2 were developed and internally validated to facilitate early, prospective identification of the clusters based on baseline and week-2 data. Results: Two longitudinal clusters were characterized: a favorable-response cluster (C1, n = 246, 56.2%) and a less favorable-response cluster (C2, n = 192, 43.8%). Compared with C1, C2 was characterized by older age, longer disease duration, and more frequent prior TNFi exposure, despite comparable baseline C-reactive protein (CRP) and erythrocyte sedimentation rate (ESR). For the majority of clinical endpoints except CRP and BASMI, significant divergence emerged as early as week 2 and was sustained through week 24. For instance, C1 achieved significantly higher rates of ASDAS-Inactive Disease (ASDAS-ID) than C2 at week 2 (0.21 [95% CI 0.16, 0.26] vs. 0.01 [95% CI 0.00, 0.02]; Risk Difference [RD]: -0.21 [95% CI -0.26, -0.15], p < 0.001) and week 24 (0.63 [95% CI 0.57, 0.69] vs. 0.14 [95% CI 0.09, 0.18]; RD: -0.49 [95% CI -0.57, -0.42], p < 0.001). A multivariable model incorporating one baseline and eight week-2 variables had an optimism-corrected C-statistic of 0.880 [95% CI 0.859, 0.915] to correctly identify C2. Conclusions: These observed response trajectories exhibited distinct baseline characteristics and early response divergence at week 2. While these findings provide a hypothesis-generating framework for early patient stratification, their utility for prospective clinical patient triage requires further independent validation.