A clinical risk prediction model for Barrett esophagus

The author manuscript of this article is open access and is available at the linked URLBarrett’s esophagus (BE) is the only known precursor to esophageal adenocarcinoma. As definitive diagnosis requires costly endoscopic investigation, we sought to develop a risk prediction model to aid in deciding which patients with gastroesophageal reflux (GER) symptoms to refer for endoscopic screening for BE. The study included data from patients with incident nondysplastic BE (n=285) and endoscopy control patients with esophageal inflammatory changes without BE (“inflammation controls”, n=313). We used two phases of stepwise backwards logistic regression to identify the important predictors for BE in men and women separately: firstly including all significant covariates from univariate analyses; then fitting non-significant covariates from univariate analyses to identify those effects detectable only after adjusting for other factors. The final model pooled these predictors and was externally validated for discrimination and calibration using data from a BE study conducted in western Washington State, USA. The final risk model included terms for age, sex, smoking status, body mass index, highest level of education, and frequency of use of acid suppressant medications (area under the ROC curve, 0.70, 95%CI 0.66–0.74). The model had moderate discrimination in the external dataset (area under the ROC curve, 0.61, 95%CI 0.56–0.66). The model was well calibrated (Hosmer-Lemeshow test, p=0.75), with predicted probability and observed risk highly correlated. The prediction model performed reasonably well and has the potential to be an effective and useful clinical tool in selecting patients with GER symptoms to refer for endoscopic screening for BE