Loading ...

TOPLINE: 

Researchers developed and externally validated a machine learning model that accurately predicted the risk for recurrence of Barrett's esophagus (BE) and related neoplasia in patients who had achieved complete eradication of intestinal metaplasia after endoscopic eradication therapy (EET). Longer Barrett's segment, higher BMI, older age, more treatment visits to achieve clearance, and baseline histology were among the key predictors.

METHODOLOGY: 

• EET has transformed care for BE by offering a less invasive alternative to surgery for high‑risk patients, but the rate of recurrence after successful treatment has varied across studies. Current surveillance recommendations rely mainly on baseline histology rather than individualized patient or treatment factors.
• Researchers developed and validated a machine learning-based prediction tool that incorporated patient-level data to predict the risk for and timing of recurrence among patients with BE who had achieved complete eradication of intestinal metaplasia after EET.
• They used data from several US databases to create a development cohort of 1114 patients with BE-related neoplasia (mean age at index EET, 66 years; 77.9% men) who had undergone EET and achieved complete eradication of intestinal metaplasia ( defined as no visible BE and no intestinal metaplasia on biopsies at the first post-EET endoscopy); external validation was performed in 1397 patients (mean age at index EET, 64 years; 82.5% men).
• Recurrence was defined as any subsequent detection of intestinal metaplasia, dysplasia, or esophageal adenocarcinoma in surveillance biopsies or endoscopic mucosal resection specimens.
• The model incorporated 24 predictors, including demographics (age at index EET , race, sex, and BMI), clinical factors (eg, duration and length of BE, presence and size of hiatal hernia, and dysplasia grade), medication use (eg, aspirin and acid‑suppressing drugs), and treatment details (EET modality and the number of sessions required to achieve complete eradication of intestinal metaplasia).

TAKEAWAY:

• Recurrence of BE occurred in 29.2% of patients, and recurrence of BE-related neoplasia was detected in 10.6%; the mean time to first recurrence was 21.3 months.
• The model demonstrated strong and consistent performance for predicting recurrence of BE, achieving an accuracy of 0.87, a sensitivity of 0.73, a specificity of 0.93, and an area under the receiver operating characteristic curve of 0.92 on internal validation; it maintained similar performance on external validation.
• The top predictors, which together accounted for about 69.7% of the model's contribution, were length and duration of BE, BMI, age, number of sessions needed to achieve complete eradication of intestinal metaplasia , baseline histology, and treatment with radiofrequency ablation.
• For recurrence of neoplasia, the model showed good overall accuracy and high specificity but low sensitivity, and its predictions of timing showed moderate discrimination at 1-, 3-, and 5-year timepoints.

IN PRACTICE:

"This tool may be implemented easily, guide decision making and has the potential to personalize surveillance practices in this patient population," the authors of the study wrote.

SOURCE:

The study was led by Venkata Akshintala, MBBS, PhD, Johns Hopkins Medicine in Baltimore, and Samuel Han, MD, Mayo Clinic in Rochester, Minnesota. It was published online in Clinical Gastroenterology and Hepatology.

LIMITATIONS:

The model showed strong discrimination but modest sensitivity at standard thresholds. Follow‑up durations were often short and varied between patients. Surveillance practices varied over time, and limitations of the datasets prevented the inclusion of biomarkers and measures of central obesity in the model.

DISCLOSURES:

The study received support through the Research Investment in the Scientific Enterprise funding policy of the Department of Medicine, University of Colorado Anschutz Medical Campus. One author disclosed being a co-founder of two medical technology companies. Some authors reported serving as consultants or on advisory boards, receiving research funding, and holding stock options in industry.

This article was created using several editorial tools, including AI, as part of the process. Human editors reviewed this content before publication.