A 48-week, randomized, phase 3 study (consisting of a 16-week double-blind phase, followed by a 32-week open-label phase) including 269 adult patients with active AS and an inadequate response to ≥2 NSAIDs. Randomization was stratified by bDMARD treatment history in patients who had inadequate response to ≤2 TNFis or had prior bDMARD (TNFi or non-TNFi) use without IR. At baseline, 207 patients were bDMARD-naïve, and 62 patients were TNFi-IR or prior bDMARD use (non-IR). NSAIDs/COX-2 inhibitors, MTX, SSZ, or oral corticosteroids were allowed if patients were on a stable dosage at baseline. For patients taking XELJANZ®, the mean age was 42.2 years old (40 years old for placebo), and most patients were male (87.2% for XELJANZ 5 mg BID, 79.4% for placebo). Patients taking XELJANZ in the trial had a mean disease duration since symptoms of 14.2 years (12.9 years for placebo), AS disease duration since diagnosis: mean 8.9 years (6.8 for placebo), and mean baseline BASDAI score: 6.4 (6.5 for placebo). 

The primary endpoint was to evaluate the proportion of patients who achieved an ASAS20 response at week 16. To control for type I error at the 5% level for primary and certain secondary endpoints, 4 families of efficacy endpoints were tested in hierarchical sequences with a step-down approach. The first family, the global type I error–controlled endpoints at week 16, included: ASAS20 response; ASAS40 response; ΔASDAS(CRP); ΔhsCRP, ΔASQoL, ΔSF-36v2 PCS score, ΔBASMI, and ΔFACIT-F total score. Upon meeting statistical significance for ASAS20 response at week 16, the second family included ΔASAS components at week 16, and included testing for ΔPtGA, Δtotal back pain, ΔBASFI, and Δmorning stiffness (inflammation). The third family, ASAS20 response over time, and the fourth family, ASAS40 response over time, were each tested in the following sequence: weeks 16, 12, 8, 4, and 2. In each family, statistical significance could be declared only if the prior endpoint (or time point) in the sequence met the requirements for significance. Nonresponder imputation was applied to missing data for ASAS response rates.