Reproducibility and Generalizability of the Web-Based Tool to Predict Lamina Propria Fibrosis in Eosinophilic Esophagitis

Abstract

We investigated reproducibility and generalizability of the recently developed web-based model to predict lamina propria fibrosis (LPF) in esophageal biopsies with inadequate lamina propria (LP) from patients with eosinophilic esophagitis (EoE) using an independent dataset (N = 183). For grade and stage scores of LPF, the area under the curve for predictive model was 0.77 (0.69–0.84) and 0.75 (0.67–0.82), and its accuracy was 78% and 72%, respectively. These model performance metrics were similar to that of the original model. A significant positive correlation was noted between the models’ predictive probability and the grade and stage of LPF assessed by a pathologist (grade: r 2 = 0.48, P < 0.001; and stage: r 2 = 0.39, P < 0.001). These results support the reproducibility and generalizability of the web-based model to predict the presence of LPF in esophageal biopsies with inadequate LP in EoE. Additional studies are warranted to refine the web-based predictive models to provide predictive probability for sub-scores of LPF severity.