Existing Prediction Models of Disease Course in Paediatric Crohn's Disease Are Poorly Replicated in a Prospective Inception Cohort

Abstract

Background: Several groups have proposed models to predict disease outcomes in paediatric Crohn's disease [CD], notably the RISK, GROWTH, and the Porto group, but none were externally validated. We aimed to explore these predictive models and individual predictors summarised by the PIBD-ahead project in a prospective inception cohort of paediatric CD.

Methods: We included children who were diagnosed with CD at two medical centres and followed them at 3 and 12 months thereafter as well as at the last follow-up. Outcomes included steroid-free remission [SFR], surgery, and stricturing/fistulising disease.

Results: In all 155 children were included (median follow-up of 31 [16-48] months, 107 [71%] had moderate-to-severe disease). Stricturing and penetrating disease at diagnosis were noted in 34 [22%] and two [1.3%] children, respectively, and these were excluded from the relevant analyses. At 1 year, 10 [8.3%] developed new stricturing disease, two [1.7%] developed penetrating disease, seven [5%] required intestinal surgery, and 15 [10%] required perianal surgery. The sensitivity/specificity/positive predictive value [PPV]/negative predictive value [NPV] of the GROWTH criteria for predicting SFR at 12 months [occurring in 70% of children] were 20%/85%/76%/31% and for surgery at 2 years were 96%/20%/16%/96%, respectively. Strictures were predicted by the RISK model with sensitivity/specificity/PPV/NPV of 33%/73%/18%/86%, respectively. The sensitivity/specificity/PPV/NPV of the Porto criteria to predict surgery were 86%/10%/4%/94%, respectively. None of the Pediatric Inflammatory Bowel Disease-ahead [PIBD-ahead] predictors were associated with surgery or stricturing disease.

Conclusions: None of the three main predictive models in paediatric CD achieved sufficient accuracy, far from that reported in the original cohorts. This highlights the necessity of external validation in any prediction model prior to its implementation in clinical practice.

Keywords: Crohn’s disease; predictive models; surgery.