Background/Purpose: In daily practice, radiologists/rheumatologist judge sacroiliac (SI) joints on X-rays (X-SI). However, reliable identification of radiographic sacroiliitis is difficult. Consequently, large inter- and intra-observer variations have been reported, even after specific training¹. In cohorts and clinical trials the reading is usually done by ≥1 trained readers. However, in the DEvenir des Spondylarthropathies Indifferenciées Récentes (DESIR)-cohort, X-SIs at inclusion were first read by the local radiologist/rheumatologist, then by centralized read. We compared the local (by multiple readers in various centers) to centralized read (by a few trained readers) on sacroiliitis yes/no.

Methods: Patients (pts) aged 18-50 with inflammatory back pain (IBP; ≥3 months, ≤3 years) from 25 participating centers were included in the DESIR-cohort (n=708). Available baseline X-SIs were read by local radiologists/rheumatologists with access to clinical data, according to a method derived from the modified New York (mNY) criteria². Grade 2 and 3 from the mNY were pooled together in 1 combined grade ‘DESIR-2’. Local sacroiliitis was defined by at least unilateral grade ≥DESIR-2. Next, 2 well-calibrated centralized readers independently read all X-SIs according to the original mNY, blinded for clinical data. In case of disagreement, an experienced radiologist was adjudicator. An X-SI was marked positive if 2/3 readers agreed on bilateral ≥2 or unilateral ≥3. Agreement between the 2 centralized readers, and between the local and centralized read was calculated (Kappa; % agreement).

Results: Pts with complete X-SI data (n=689) were included in this analysis. Inter reader agreement between the 2 centralized readers is moderate (Kappa 0.54), while percentage agreement (84.3%) is good (table). The low Kappa can partially be explained by the high numbers of normal X-SIs. However, the adjudicator needed to read 108/689 X-SIs (15.7%) because of disagreement among the 2 centralized readers. Comparison between the centralized and the local read shows similar levels of agreement (table). Overall, more X-SIs are read positive by local readers (n=184) than by centralized readers (n=145). In 77 pts, X-SI was read positive by local readers but negative by centralized read; in 38 pts it was the other way around.

Conclusion: Agreement between the centralized and local read, but also the inter reader agreement between the 2 centralized readers, is moderate, thereby showing that early detection of sacroiliitis on X-SIs is a challenge. In patients with recent onset IBP, trained readers do not perform better than local rheumatologists/radiologists in recognizing sacroiliitis on X-SI, suggesting that the role of X-SI as diagnostic criterion for axSpA should be re-evaluated.

References: ¹van Tubergen ARD 2003;62:519-25 ²van der Linden A&R 1984;27:361–8

		Reader 2
Reader 1		modified New York +	modified New York –
	modified New York +	96	58
	modified New York –	50	485
	Kappa (95% CI) / Agreement (%)	0.54 (0.46-0.62)	84.3
		Centralized score (2/3)
Local score		modified New York +	modified New York –
	DESIR mNY +	107	77
	DESIR mNY –	38	468
	Kappa (95% CI) / Agreement (%)	0.54 (0.47-0.62)	83.3

---

« Back to 2013 ACR/ARHP Annual Meeting

ACR Meeting Abstracts - https://acrabstracts.org/abstract/reading-of-the-sacroiliac-joints-on-plain-radiographs-agreement-between-clinical-practice-and-trained-central-reading-of-the-desir-cohort/