Radiologist Performance and Reliability of MR Scoring Systems in Perianal Fistulizing Crohn’s Disease: Retrospective Analysis in a Real-World Clinical Practice

Background Perianal fistulizing Crohn’s disease (CD-PAF) is difficult to treat, and several MR scoring systems have been developed to assess treatment response. Objective To assess the relationship between CD-PAF MR scoring systems and radiologists’ subjective assessment of CD-PAF severity and treatment response on baseline and follow-up pelvic MR. Methods Retrospective single institution study of consecutive symptomatic patients with CD-PAF patients who underwent pelvic MR before and ≥3 months after initiating biologic therapy during a 10-year period (December 2011 to December 2021). One of four radiologists interpreted baseline and follow-up MRs. Scoring systems included variables in the modified Van Assche index (mVAI), magnetic resonance novel index for fistula imaging in CD (MAGNIFI-CD), and pediatric MR-based perianal Crohn’s disease (PEMPAC) index. For initial and follow-up MR, a 5-point Likert scale assessed severity (1=mild, 3=moderate, 5=marked). On follow-up MR, radiologists evaluated fistula response as 1-improved, 3-stable, or 5-worsened CD-PAF severity. All four study radiologists scored the baseline MR in 20 patients to calculate inter-reader agreement statistics. Interrater reliability was assessed with the Krippendorff α coefficient for categorical variables and intraclass correlation coefficient (ICC) for continuous variables. Results The final cohort included 96 CD-PAF patients (50 men; mean age=33.0 years) with 192 baseline and follow-up MRs. Moderate to substantial agreement was observed among radiologists for MAGNIFI-CD, mVAI, PEMPAC, and Likert scores (ICC: 0.716, 0.756, 0.535, and 0.679 respectively). Individual components of MR scoring systems had Fair to Substantial agreement (Alpha: 0.195 to 0.730). Significant univariate associations were found between MR scoring systems and radiologists’ Likert severity assessments (p<0.001, Pearson correlation coefficients ≥0.820). In patients meeting criteria for change in disease severity (n=17), all scoring systems demonstrated AUC values ≥0.93. Conclusion The MR scoring systems for CD-PAF (MAGNIFI-CD, mVAI, and PEMPAC) demonstrated strong associations with radiologists’ subjective assessments of severity and treatment response on baseline and follow-up pelvic MR. Inter-reader agreement of these scoring systems outperformed individual MR factors.