A Machine-Learning-Based Predictor for Histological Remission in Ulcerative Colitis

Researchers sought to create a histological index that would help evaluate inflammation and predict clinical outcomes in patients with ulcerative colitis.

The Paddington International virtual ChromoendoScopy ScOre (PICaSSO) Histologic Remission Index (PHRI) and a PHRI-artificial intelligence (AI) system were found to be a simple scoring system and predictor of histological remission in ulcerative colitis (UC), according to study findings published in Gut.

Patients (N=307) with UC were prospectively enrolled for the study at 11 sites in Europe and North America. All patients underwent white light high-definition colonoscopy and gave at least 2 biopsies. These data were used to develop a predictive model of histological remission in 4 stages: deep histological analyses of the correlations between clinical outcomes and histological status using 5 histological indices, development of the PHRI, validation of the PHRI, and development of an AI algorithm. To achieve this, biopsies (N=614) were divided into training (70%) and test (30%) datasets, and 15% of the training set was also used to validate the scoring system.

Patients had a mean age of 48.4±14.8 years; 59.3% were men; 56.0% had subtotal or total disease; 71.7% were in remission, according to the PICaSSO score; and 76.2% had received 5-aminosalicylic acid treatment in the previous 12 months.

The 5 histological indices (Villanacci Simplified Score [VSS]; Robarts Histological Index [RHI]; Nancy Histological Index [NHI]; Geboes Score [GS]; and extent, chronicity, activity, and plus [ECAP] score) were strongly correlated with endoscopic scores (r range, 0.55-0.78) and weakly correlated with adverse clinical outcomes at 12 months (r range, 0.34-0.42). The strongest correlations were observed for neutrophil infiltration of the lamina propria (r range, 0.60-0.76).

PHRI scores were defined by assigning 1 point each for neutrophil infiltration in the lamina propria, surface epithelium, cryptal epithelium, and crypt abscess. The maximum PHRI score is 4.

PHRI was found to correlate with adverse clinical outcomes at 12 months (r, ~0.4) and those with PHRI greater than 0 were associated with more adverse clinical outcomes than those with PHRI equal to 0 (48.65% vs 13.91%; P <.00001).

A receiver operating characteristic (ROC) curve analysis found the best cutoff for predicting adverse 12-month outcomes was a PHRI score of 1. Patients with a PHRI score greater than 0 were associated with poorer survival (hazard ratio [HR], 0.2185; 95% CI, 0.1285-0.3716; P =1.76´10-9).

In the validation stage of the study, the PHRI score was found to be more weakly correlated with endoscopic scores (r range, 0.24-0.36), but was generally superior to other histological indices.

In the final stage of the study, the PHRI-AI system was found to have a sensitivity of 0.8136, specificity of 0.9253, positive predictive value of 0.8683, negative predictive value of 0.9076, and accuracy of 0.8783 for classifying neutrophil infiltration. For classifying UC activity, the system was found to have a sensitivity of 0.6700, specificity of 0.9000, positive predictive value of 0.8000, negative predictive value of 0.8182, and accuracy of 0.8371.

The study was limited by not reevaluating patients by endoscopy at 12 months to determine longer-term predictive accuracy of histological remission.

“PHRI is a simple and reproducible histological index that correlates strongly with endoscopic activity and predicts clinical outcomes in UC,” the study authors noted. “It is therefore ideally suited for adoption in clinical practice as well as for consideration in clinical trials and central readouts if further validated to fulfil requirements of US Food and Drugs Administration or European Medicines Agency requirements.”


Gui X, Bazarova A, del Amor R, et al. PICaSSO Histologic Remission Index (PHRI) in ulcerative colitis: development of a novel simplified histological score for monitoring mucosal healing and predicting clinical outcomes and its applicability in an artificial intelligence system. Gut. Published online February 16, 2022. doi:10.1136/gutjnl-2021-326376