Gastroenterology
Partner With Us
Advertisement

Machine learning model helps predict Barrett’s esophagus recurrence 

“This practical tool has the potential to provide a personalized approach to surveillance in this patient population."

The New Gastroenterologist
A new study published in Clinical Gastroenterology and Hepatology found that a machine learning–based tool could help clinicians better predict which patients are most likely to experience recurrence after endoscopic eradication therapy (EET) for Barrett’s esophagus, and when that recurrence might occur.Researchers analyzed data from 2,511 patients across multiple US cohorts who had Barrett’s esophagus-related neoplasia and successfully achieved complete eradication of intestinal metaplasia after treatment. They developed and validated a machine learning model to predict recurrence risk and timing using demographic, endoscopic, pathologic, and treatment-related variables.Over an average follow-up of just over three years, recurrence was relatively common. Barrett’s esophagus recurrence “occurred in 29.2% (n=734) of patients and [Barrett’s esophagus]-related neoplasia recurrence in 10.6% (n=265) with a mean time to recurrence of 21.3 months," reported Venkata Akshintala, MD, of the Division of Gastroenterology, Johns Hopkins Medical Institutions, Baltimore, MD, and colleagues.The model performed well in identifying patients at higher risk of recurrence. Some of the top predictors included the Barrett’s esophagus length, body mass index, age, how many treatment sessions were needed to achieve complete eradication of intestinal metaplasia, and baseline histology.While the tool was strong at estimating recurrence risk, it was less precise in predicting exactly when recurrence would happen, especially further out from treatment. Still, according to the study authors, it may provide useful estimates that could help guide surveillance planning.Recurrence after successful EET occurred in approximately one-third of patients, and risk varied significantly from patient to patient. Current surveillance schedules are largely based on the initial severity of disease, but the study findings suggest a more individualized approach may be possible.“Importantly, the clinical intent of this model is not to exclude recurrence, but to stratify patients by relative risk in order to individualize surveillance intensity,” noted the study authors.They reported several important limitations. Missing data required imputation, although the model remained reasonably robust even when variables with high missingness were excluded. There was also variable and sometimes limited follow-up, nonstandardized surveillance practices across centers, and the inability to account for factors such as biopsy protocols, treatment differences, or anti-reflux surgery.Even with these caveats, the findings highlight a shift toward more personalized care in Barrett’s esophagus. Sachin Wani, MD, one of the authors of the study, said in an interview with GI & Hepatology News that Barrett’s esophagus is the only known precancerous condition for esophageal adenocarcinoma. While EET for patients at the highest risk of progression to invasive esophageal adenocarcinoma have “revolutionized” the management of these conditions, Barrett’s esophagus and related dysplasia can recur after EET.“Currently, we use a one-size-fits-all approach to monitoring these patients for recurrence using endoscopic surveillance procedures,” he noted.  However, the machine learning-based prediction tool may allow clinicians to accurately predict individuals who are at the highest risk for recurrence and the timing of these recurrences. “This practical tool has the potential to provide a personalized approach to surveillance in this patient population — ensure that those patients who need monitoring the most receive this and avoid unnecessary procedures on patients who are least likely to have a recurrence and benefit from these endoscopies,” he said. Several authors reported conflicts of interest, including consulting work for companies like Castle Biosciences, Exact Sciences and CDx Medical; research funding; advisory board membership; and stock options.