Detection of Barrett's esophagus through exhaled breath using an electronic nose device: a multicenter validation study.
Authors: Lotte J. Huibertse, Yonne Peters, Ruud W.M. Schrauwen, Adriaan C.I.T.L. Tan, Niek Warringa, Jordy P. W. Burger, Peter Derk Siersema
Journal: PubMed
Publication year: 2026
DOI: 10.1093/dote/doag124
Abstract
Esophageal adenocarcinoma incidence has risen dramatically in the Western world over the past 30 years, particularly driven by gastroesophageal reflux disease (GERD) and Barrett's esophagus (BE). We aimed to assess whether Volatile Organic Compound (VOC) breath testing can predict BE and to validate its performance in independent cohorts including participants with BE, participants with GERD, and controls. Participants aged ≥18 years referred for gastroscopy between August 2017 and July 2022, were eligible for inclusion. The diagnostic accuracy of VOC breath testing was assessed using four machine learning algorithms (MLAs): (1) random forest, (2) extremely randomized trees, (3) eXtreme Gradient Boosting (XGBoost), and (4) convolutional neural network. Patient tolerability and patient acceptability of breath testing were secondary outcome measures. A total of 1079 participants were included, 602 (55.8%) in the training cohort and 477 (44.2%) in the validation cohort. In the training cohort, XGBoost was the best performing MLA in differentiating BE from GERD and controls (area under the receiver operating characteristic [AUROC] 0.892 [95% CI, 0.862-0.921]). However, in the validation cohort, the performance of XGBoost markedly declined (AUROC 0.530 [95% CI, 0.472-0.589]). Median tolerability score was 2.00 with 0 'no discomfort' and 10 'worst discomfort imaginable', and median acceptability score was 8.00 with 0 'worst experience' and 10 'best experience'. VOC breath testing in its current form does not provide additional diagnostic value for BE screening; however, based on the high degree of patient acceptability, further research on eNose technology seems justified.