Background Several models have been proposed to refer patients with possible axial spondyloarthritis (axSpA) to a rheumatologist. Our aim was to evaluate performance of these models in a single cohort.

Methods 13 referral models found in the literature were evaluated in the Leiden SPondyloArthritis Caught Early (SPACE) cohort, which includes patients with back pain (≥3 months, ≤2 years, onset <45 years; n=261) referred to a rheumatology outpatient clinic. Imaging was not considered as a referral parameter. Performance of the strategies was evaluated (sensitivity, specificity, positive likelihood ratio (LR+)) using diagnosis by a rheumatologist as an external standard. For secondary analyses, fulfilment of the Assessment in SpondyloArthritis international Society (ASAS) axSpA criteria was used as an external standard.

Results In total, 107/261 patients were diagnosed with axSpA. Most models performed well regarding sensitivity and specificity. The MASTER strategy showed a balanced sensitivity/specificity with the highest LR+. The ASAS and Brandt I strategies are the most sensitive strategies. Using classification by ASAS axSpA criteria as the external standard gave comparable results. Most patients missed by the strategies fulfilled the imaging arm of the ASAS axSpA criteria.

Conclusions Most referral models performed well, although patients in SPACE have already been referred, which may have led to overestimation of performance. If no patient is to be missed, the ASAS strategy would be most preferable. If the number of referrals needs to be limited, the MASTER strategy seems to perform best. The ‘ideal’ referral strategy may be different from country to country, due to differences in healthcare structure and prevalence of referral parameters such as human leucocyte antigen-B27.