External validation of a model determining risk of neoplastic progression of Barrett's esophagus in a cohort of U.S. veterans.

ABSTRACT

BACKGROUND AND AIMS: Risk of esophageal adenocarcinoma (EAC) in those with Barrett's esophagus (BE) is 11-fold greater than the general population. It remains unclear which BE patients are at highest risk of progression to EAC. We aimed to validate a predictive model risk-stratifying BE patients. METHODS: We conducted a retrospective cohort study at the Houston Veteran Affairs Medical Center of consecutive patients with a new diagnosis of BE from November 1990 to January 2019. Study follow-up was through February 2020. Patients were excluded if they had no follow-up EGD with esophageal biopsy sampling after the initial BE-diagnosing EGD or evidence of high-grade dysplasia (HGD) or EAC on initial EGD. We performed an external validation study of a risk model containing sex, smoking, BE length, and low-grade dysplasia (LGD) status and assessed discriminatory ability using the area under the receiver operating characteristic curve (AUROC). RESULTS: Among 608 BE patients, 24 progressed to HGD/EAC. The points-based model discriminated well with an AUROC of .72 (95% confidence interval [CI], .63-.82). When categorized into low-, intermediate-, and high-risk groups according to published cutoffs, the AUROC was poor at .57. Restructured into low-risk versus high-risk groups, the AUROC was .72 (95% CI, .64-.80). Excluding baseline LGD did not reduce discriminatory ability (AUROC, .73; 95% CI, .64-.82). CONCLUSIONS: This external validation provides further evidence that the model including sex, LGD status, smoking status, and BE length may help to risk stratify BE patients. A simplified version excluding LGD status and/or reducing the number of risk groups has increased utility in clinical practice without loss of discriminatory ability.