Fig. 1.

Initial mediation models a A (FACIT-F-based) and b B (BASDAI Q1-based). Model A: Treatment was the independent binary variable (tofacitinib 5 mg BID versus placebo). Variables i1–i13 represented items of the FACIT-F scale. Latent factor F3 represented overall fatigue; latent factors F1 and F2 represented ‘experience’ (items i1–i4 and i7 of the FACIT-F) and ‘impact’ (items i5–i6, i8–i13 of the FACIT-F) domains, respectively. Latent factor F4 represented back pain as measured by total back pain due to AS on average during last week (p1) and nocturnal spinal pain due to AS on average during last week (p2). Morning stiffness was considered an observed variable and was represented by the mean of BASDAI Q5 and Q6. CRP was considered an observed variable. Model B: Treatment was the independent binary variable (tofacitinib 5 mg BID versus placebo). Observed variable fatigue was represented by BASDAI Q1. Latent factor F4 represented pain as measured by BASDAI Q2 (p1) and Q3 (p2). Morning stiffness was considered as an observed variable and was represented by the mean of BASDAI Q5 and Q6. CRP was considered an observed variable. AS ankylosing spondylitis, BASDAI Bath Ankylosing Spondylitis Disease Activity Index, BID twice daily, CRP C-reactive protein, FACIT-F Functional Assessment of Chronic Illness Therapy-Fatigue, Q question