Validation of predictive models for disease outcomes in paediatric ulcerative colitis: A multicentre prospective inception cohort

Background: Several studies have proposed models to predict disease outcomes in paediatric ulcerative colitis (UC), notably PROTECT, Schechter and PIBD-ahead, but none has been validated by external cohorts. Aim: To explore these models in a prospective multicentre inception cohort. Methods: Children newly diagnosed with UC in 17 centres were followed at disease onset and 3 and 12 months thereafter, as well as at last visit. Outcomes included steroid-free remission (SFR) and acute severe colitis (ASC). Results: Of the 223 included children, 74 (34%), 97 (43%) and 52 (23%) presented with mild, moderate and severe disease, respectively. SFR rate was 35% at 3 months and 47% at 12 months (62% of those with mild disease at diagnosis vs. 41% in moderate–severe disease; p = 0.01). Thirty-six (16%) children developed ASC during the first month after diagnosis, and 53 (24%) during the first year. The AUC of the PROTECT model for predicting SFR at 3 and 12 months was 0.78 [95% CI 0.65–0.92] and 0.57 [95% CI 0.47–0.66], respectively. The sensitivity/specificity/PPV/NPV of Schechter's criteria to predict sustained SFR at 12 months was 50%/60%/35%/74%. ASC was predicted only by the PUCAI score at diagnosis and at 3 months. Conclusions: The PROTECT model had a good predictive utility for SFR at 3 months, but not at 12 months. The other predictive models did not achieve sufficient accuracy, which was far from that reported in the original studies. This highlights the necessity for external validation of any prediction model prior to its implementation into clinical practice.