Long-Term Performance of Readers Trained in Grading Crohn Disease Activity Using MRI

Abstract

We aim to evaluate the long-term performance of readers who had participated in previous magnetic resonance imaging (MRI) reader training in grading Crohn disease activity. Fourteen readers (8 women; 12 radiologists, 2 residents; mean age 40; range 31-59), who had participated in a previous MRI reader training, participated in a follow-up evaluation after a mean interval of 29 months (range 25-34 months). Follow-up evaluation comprised 25 MRI cases of suspected or known Crohn disease patients with direct feedback; cases were identical to the evaluation set used in the initial reader training (of which readers were unaware). Grading accuracy, overstaging, and understaging were compared between training and follow-up using a consensus score by two experienced abdominal radiologists as the reference standard. In the follow-up evaluation, overall grading accuracy was 73% (95% confidence interval [CI]: 62%-81%), which was comparable to reader training grading accuracy (72%, 95% CI: 61%-80%) (P = .66). Overstaging decreased significantly from 19% (95% CI: 12%-27%) to 13% (95% CI: 8%-21%) between training and follow-up (P = .03), whereas understaging increased significantly from 9% (95% CI: 4%-21%) to 14% (95% CI: 7%-26%) (P < .01). Readers have consistent long-term accuracy for grading Crohn disease activity after case-based reader training with direct feedback.