Predicting presence of intestinal metaplasia and dysplasia in columnar-lined esophagus: a multivariate analysis

Abstract

In patients with presumed Barrett esophagus we evaluated clinical risk factors that could predict the presence of intestinal metaplasia and dysplasia in biopsies of columnar-lined esophagus (CLE), independently of histological results. In 908 patients with CLE of length > or = 2 cm, data on age, sex, reflux symptoms, tobacco and alcohol use, medication use, and upper gastrointestinal endoscopy findings were prospectively collected. Multivariate logistic regression analysis was performed, and a model for predicting the histological results was developed. In 127/908 patients, biopsies of CLE did not contain intestinal metaplasia. Of the 781 patients with intestinal metaplasia, 663 patients (85%) had no dysplasia, and 118 (15%) had low grade dysplasia (LGD). The most important predictors for the presence of intestinal metaplasia were length of CLE, size of hiatal hernia, and male sex, while among those with intestinal metaplasia, age and male sex were most important for the presence of LGD. Multivariate combinations of these predictors yielded reliable models, which were able to discriminate intestinal metaplasia well from no intestinal metaplasia (area under receiver operating characteristic [ROC] curve 0.82), but only reasonably discriminated LGD from no dysplasia (area under ROC 0.65). A simple model based on clinical findings can be used to predict the presence of intestinal metaplasia in biopsies from CLE. In contrast, predicting the presence of LGD versus no dysplasia in intestinal metaplasia is more difficult. Predictions from these models may aid decision making on whether a patient with CLE should have surveillance, in view of the known sampling error at endoscopy and interobserver variability at histology.