Clinical predictors of tocilizumab response in moderate-to-severe thyroid eye disease: a large single-center retrospective cohort study

Background Tocilizumab (TCZ), an interleukin-6 receptor antagonist, is recommended as a second-line therapy for thyroid eye disease (TED), yet the clinical determinants of treatment response remain poorly defined. Identifying reliable predictors of response is essential for patient selection and precision use. Methods We conducted a single-center retrospective cohort study of 234 patients with moderate-to-severe TED treated with intravenous TCZ (8 mg/kg every 4 weeks) at Beijing Tongren Hospital from January 2023 to December 2025. Treatment response was defined as a reduction in Clinical Activity Score (CAS) of ≥2 points from baseline. Candidate predictors were entered into a multivariate binary logistic regression. ROC analysis identified optimal cut-off values. Results Following TCZ treatment, CAS decreased significantly from 3.42 ± 0.61 to 1.70 ± 0.67 (P < 0.001), and TRAb declined from 10.35 ± 11.99 IU/L to 4.46 ± 6.73 IU/L (P < 0.001). Of 234 patients with complete paired data, 155 (66.2%) were Responders. Responders had higher TRAb (12.17 vs. 7.07 IU/L), fibrinogen (FIB; 2.83 vs. 2.58 g/L), and HDL-cholesterol (HDL-C; 1.37 vs. 1.24 mmol/L), and shorter disease duration (7.91 vs. 10.08 months; all P < 0.05). Multivariate analysis identified four independent predictors: shorter disease duration (OR = 0.962 per month, 95% CI 0.928–0.998, P = 0.038), higher TRAb (OR = 1.036 per IU/L, 95% CI 1.004–1.068, P = 0.025), higher FIB (OR = 1.869 per g/L, 95% CI 1.097–3.184, P = 0.021), and higher HDL-C (OR = 2.665 per mmol/L, 95% CI 1.175–6.041, P = 0.019). The composite model achieved AUC = 0.695 (95% CI 0.625–0.765, P < 0.001). Conclusions TCZ produces clinically meaningful and multi-dimensionally validated response with moderate-to-severe TED. Shorter disease duration, higher baseline TRAb, FIB, and HDL-C independently predict response, offering a preliminary framework to inform patient selection. These findings provide the largest real-world evidence to date for guiding precision use of TCZ in TED.