A Meta-Analysis of the Efficacy of JAK Inhibitors in the Treatment of Rheumatoid Arthritis Assisted by Random Forest Algorithm Analysis

Background: Rheumatoid arthritis (RA) was a chronic autoimmune disease that could lead to joint deformity and functional impairment. As novel oral targeted agents, JAK inhibitors required systematic evaluation of their efficacy, safety, and impacts on biomechanically relevant indices in the treatment of RA. Objective: This meta-analysis aimed to determine the effects of JAK inhibitors on ACR20 response rates, safety, and dose differences in patients with RA, and to explore core influencing factors by combining biomechanical indices (DAS28-CRP, HAQ-DI) and the random forest algorithm. Methods: Relevant randomized controlled trials (RCTs) were searched in databases including PubMed and Embase from inception to July 31, 2025. After screening and quality assessment, statistical analyses were performed using RevMan 5.3, and influencing factors were explored in combination with the random forest model. Results: Eight articles with 15 RCTs were included (2814 patients in the JAK inhibitor group and 2553 in the placebo group). The pooled odds ratio (OR) for ACR20 response rate in the JAK inhibitor group was 3.74 (95% CI [3.33, 4.20], I 2 = 6%), and the OR for adverse event risk was 1.31 (95% CI [1.17, 1.47], I 2 = 46%). Dose subgroup analyses showed that high-dose regimens achieved the best efficacy (OR = 3.79), while medium-dose regimens showed the best safety profile (OR = 1.32). No statistically significant differences were observed between the two groups in DAS28-CRP (MD = 0.02) or HAQ-DI (MD = -0.01). However, the improvements in these two indices were core features in the random forest model. The model achieved an accuracy of 86.7% and an AUC of 0.91; core influencing factors included administered dose (28.6%) and reduction in DAS28-CRP (21.3%). Conclusion: JAK inhibitors significantly improved ACR20 response rates in patients with RA, and medium-dose regimens achieved a more balanced efficacy-safety profile. Biomechanical indices were important for efficacy prediction, and the random forest model could assist individualized treatment decisions. More long-term RCTs were needed to strengthen the evidence.