Diagnostic performance of sacroiliac joint MRI and added value of spine MRI to detect active spondyloarthritis

•The combination of MRI of sacroiliac joints and thoracic spine allows discriminating between active and inactive spondyloarthritis.•Combining sacroiliac joint and thoracic spine MRI outperforms MRI of sacroiliac joint alone to identify clinically active spondyloarthritis.•Sacroiliac joint MRI is an informative classification model of patients with spondyloarthritis according to disease activity, regardless of the cut-off level on ASDAS-CRP. To investigate the diagnostic performance of sacroiliac joint (SIJ) magnetic resonance imaging (MRI) and the incremental value of spine MRI to “predict” clinical disease activity in patients with axial spondyloarthritis (axSpA). This cross-sectional study included adult patients with known axSpA according to the SpondyloArthritis International Society (ASAS) classification criteria, radiological arm. MRI disease activity was scored semi-quantitatively for SIJ and total spine MRI in each patient. Two cut-off levels (≥ 1.3 and ≥ 2.1) for ankylosing spondylitis disease activity score with C-reactive protein (ASDAS-CRP) were considered for clinical disease activity categorization. MRI scores were first evaluated individually. Then, SIJ score was combined with the score from a spine segment (lumbar, cervical, thoracic or total spine) to build a bi-parametric model using a classification tree. Receiver operating characteristic (ROC) curves were constructed to evaluate the classification performance according to disease activity category of these models. Forty-four patients (30 men, 14 women; mean age, 37 years±10 [SD] [range: 17–64 years]) with a mean disease duration of 5 years±8 (SD) (range: 0–35 years) were included. Thirty-six patients (36/44; 82%) had ASDAS-CRP≥1.3 and 27 patients (27/44; 61%) had ASDAS-CRP≥2.1. The most frequently involved spinal segment was mid-thoracic (T7-T8). The SIJ MRI score was an informative model to identify active axSpA (AUC≥0.7, regardless of the cut-off level on ASDAS-CRP). Performance of bi-parametric models based on “SIJ+thoracic spine” (for detecting patients with ASDAS-CRP≥1.3) or “SIJ+total spine” (for detecting patients with ASDAS-CRP≥2.1) outperformed that of the individual SIJ score (P