Houston-BEST Predicts Probability of Barrett Esophagus in Patients at High Risk

Researchers sought to develop and validate a risk prediction model for Barrett esophagus (BE), which would incorporate the effect of various BE risk factors.

Researchers developed and validated a prediction model tool, called Houston-BEST, which is compatible with extracting data from electronic health records (EHR) to determine the risk for Barrett esophagus (BE) with a moderate strength of discrimination, according to study findings published in Clinical Gastroenterology and Hepatology.

Researchers identified specific risk factors predictive of the development of BE in a derivation cohort and validated the risk factors in 2 validation cohorts.

The derivation cohort consisted of 1624 adults aged between 40 and 80 years, who underwent esophagogastroduodenoscopies at the Houston Veterans Affairs (VA) Medical Center between 2008 and 2012. Of the 1624 patients in this derivation cohort, 274 had BE and 1350 did not have BE.

The first validation cohort consisted of 513 adults from primary care clinics at the Houston VA between 2008 and 2012 — 44 with BE and 469 without BE. The second validation cohort analyzed data from an external site at the University of Michigan Ann Arbor VA, consisting of 71 patients with BE and 916 without BE.

The researchers analyzed several risk factors in this predictive model (Houston-BEST). The risk factors predictive of BE included:

  • White (odds ratio [OR], 4.94, 95% CI, 3.22-7.57) or Hispanic (OR, 2.63; 95% CI, 1.41-4.90) race and ethnicity
  • male sex (OR, 3.58; 95% CI, 1.71-7.50)
  • family history of esophageal cancer (OR, 1.92; 95% CI, 0.91-4.05)
  • age 50 years and older (OR,1.68; 95% CI, 1.04-2.73)
  • GERD (OR, 1.52; 95% CI, 1.15-2.03)
  • current or former smoking status (OR, 1.24; 95% CI, 0.85-1.81 and OR, 1.06; 95% CI, 0.74-1.50, respectively), and
  • obesity (OR, 1.13; 95% CI, 0.76-1.67).

Compared with current societal clinical practice guidelines, the Houston-BEST predictive model demonstrated moderate, yet superior, discriminative ability (area under the receiver operating characteristics (AUROC) curve, 0.69; 95% CI, 0.66-0.72; sensitivity, 90%; specificity, 39.9%) in the derivation cohort, the first validation cohort at the Houston VA (AUROC curve, 0.68; 95% CI, 0.60-0.76; sensitivity, 84.1%), and the second validation cohort in Ann Arbor (AUROC curve, 0.70; 95% CI, 0.64-0.76; sensitivity, 0%).

“We developed and validated the Houston-BEST model that predicts the probability of BE and contains terms that can be sourced from the EHR,” the study authors said. “Further studies should focus on the development and implementation of an e-trigger tool that contains this EHR-adaptable BE risk model.”

Study limitations include lack of generalizability, as the model could be refined to include nonveteran populations at lower risk for BE.


Wenker TN, Rubenstein JH, Thrift AP, Singh H, El-Serag HB. Development and validation of the Houston-BEST, a Barrett’s esophagus risk prediction model adaptable to electronic health records. Clin Gastroenterol Hepatol. Published online August 17, 2022. doi:10.1016/j.cgh.2022.08.007