Abstract
Introduction: Axial Spondyloarthritis is a complex and heterogenousisorder. The disease varies significantly leading to a diverse spectrum of management choices. We analysed retrospective clinical data from our centre to identify factors associated with multiple biologic switches. We used clustering analysis, an unsupervised machine learning algorithm, and multivariate logistic regression.
Aim: To identify factors associated with a higher frequency of biologic switches in axial spondyloarthropathy patients in a real-world clinical setting.
Materials and Methods: Data were collected retrospectively from the consultations of 166 patients receiving biologic treatment for axial spondyloarthropathy at our centre from 2003 until 2021. Feature selection included: demographics; body mass index; clinical phenotype (axial involvement; peripheral arthritis; enthesitis; uveitis; psoriasis; inflammatory bowel disease); HLA-B27 positivity; radiographic disease; chronic widespread pain diagnosis; disease activity measures (baseline and aggregate scores over disease course) – Bath Ankylosing Spondylitis Disease Activity Index; Spinal pain Visual Analogue Score; Bath Ankylosing Spondylitis Functional Index; C-reactive protein; time to start biologic from diagnosis; number of biologics and mode of action. Clustering analysis included two additional variables: – response to Tumour Necrosis Factor inhibitors and Interleukin-17 inhibitors. Patients were defined as high biologic switchers if they received three or more biologics (not including non-medical switches to biosimilar agents). Multi-variate logistic regression was performed using MNLogit algorithm and clustering analysis using the k-means algorithm (Anaconda Distribution 2.7).
Results: Clustering partitioned our dataset into three clusters: Low Disease Burden (LDB), High Disease Burden 1(HDB1) and High Disease Burden 2(HDB2). The LDB cluster showed good response to treatment, lower disease activity scores and fewer treatment switches. HDB clusters had higher disease activity scores; however, the HDB1 patients had significantly fewer biologic switches. Common features of the HDB1 cluster were female sex, HLA-B27 negativity, less radiographic disease, and more chronic widespread pain diagnosis. Multivariate logistic regression showed that HLA-B27 positivity and higher disease activity scores were positively associated with more biologic switches, whereas time to start biologic and a diagnosis of chronic widespread pain were negatively associated.
Conclusion: HLA-B27 positivity, male sex, higher radiographic burden, higher disease activity scores and early biologic requirement were associated with more biologic switches. Females with axial spondyloarthropathy, HLA-B27 negativity and lower radiographic disease burden had significantly fewer biologic switches despite higher disease activity scores and were more likely to have accompanying chronic widespread pain. Despite advances in treatment, patients with high symptom burden pose a challenge in clinical practice. Consideration should be given to objective and holistic assessment of symptoms and treating other associated conditions, as necessary.