Abstract
BACKGROUND: Axial spondyloarthritis (AS) is a chronic inflammatory disease that significantly affects quality of life and imposes a high economic burden on patients due to the cost of biologic disease-modifying antirheumatic drugs (bDMARDs). Dose reduction strategies for bDMARDs may offer a feasible approach to maintaining clinical efficacy while reducing costs. This study aimed to evaluate the clinical effectiveness and cost-efficiency of bDMARD dose reduction in patients with AS and apply machine learning to identify key factors influencing disease control. METHODS: This 12-month prospective study, 368 AS patients receiving ≥ 3 months of full-dose bDMARDs were included. Among 215 initial responders (ASDAS < 2.1), 146 underwent dose reduction while 69 continued full-dose therapy. Clinical outcomes such as C-reactive protein (CRP) levels, the Bath ankylosing spondylitis disease activity index (BASDAI) and ankylosing spondylitis disease activity score (ASDAS) were assessed, along with cost effectiveness using incremental cost effectiveness ratios (ICER). Random forest models were developed to predict the achievement of inactive disease (ASDAS < 1.3) and to identify key predictors. RESULTS: The dose reduction group demonstrated significantly greater improvements in CRP (-4.05 vs. +2.83 mg/L, p < 0.001), BASDAI (-3.00 vs. +0.89, p < 0.001), and ASDAS (-1.42 vs. +0.09, p < 0.001) compared to the full dose group. A greater proportion of patients in the reduced dose group achieved ASDAS < 1.3 at 12 months (93.2% vs. 33.3%, p < 0.001), with a shorter median time to response (4.20 vs. 4.70 months, p < 0.001). The ICER for achieving ASDAS < 1.3 was favorable (-$6,209.78; 95% CI:-$9,048.35 to-$4,015.78), supporting the cost-effectiveness of dose reduction. A random forest model identified reduced dose strategy, baseline ASDAS, BASDAI, treatment duration, and CRP as key predictors of ASDAS < 1.3, achieving an AUC of 0.845 and F1-score of 0.774. CONCLUSIONS: In this cohort, bDMARD dose reduction was associated with preserved clinical outcomes and lower costs, suggesting it may be a viable strategy for selected patients under close clinical supervision. Predictive modeling provided actionable insights to optimize personalized treatment strategies, balancing efficacy and economic sustainability. These findings support further evaluation of dose reduction strategies, especially in resource-limited settings, to inform potential integration into routine practice.