A new study, presented at the American Foregut Society (AFS) Meeting 2025, found that an AI-enabled spatial proteomics test assessed progression risk in patients with Barrett’s esophagus (BE) more accurately than standard-of-care tests.
BE is the precursor to esophageal adenocarcinoma (EAC), a cancer with a five-year survival rate of 22 percent. The incidence of EAC continues to rise despite the use of surveillance methods such as endoscopy, clinical history, and pathology review – techniques that can result in misclassification of the patient’s risk of disease progression.
To find out more about the implications of the study on patient outcomes we spoke with lead study author Caitlin C. Houghton, a surgeon at Keck Medicine of USC in Los Angeles, and Emmanuel Gorospe, Medical Director at Castle Biosciences.
How does AI-driven spatial proteomics add precision beyond traditional pathology, and why is that clinically significant for risk assessment of patients with Barrett’s esophagus?
EG: AI-driven spatial proteomics enables the detection of patterns in tissue – such as biomarker expression, cell morphology, and spatial relationships – that are often too complex for manual interpretation using traditional pathology. The spatial proteomics BE test builds on pathology by looking for molecular changes that often precede histologic changes that are visible to a pathologist under a microscope.
While most patients with BE never progress to cancer, a small subset will – around 1 percent per year. Distinguishing who is truly at risk is challenging, since Barrett’s tissue changes can be subtle or obscured by reflux-induced chronic inflammation that can mimic dysplasia.
Spatial proteomics test results complement the pathologist’s interpretation by adding objectivity and precision. For patients with non-dysplastic Barrett’s esophagus (NDBE) who appear low risk under the microscope, spatial proteomics can reveal high-risk biology earlier, when interventions such as endoscopic eradication therapy (EET) have the greatest potential to prevent cancer.
The test uses the same formalin-fixed, paraffin-embedded (FFPE) biopsy samples taken during endoscopy. An AI-enabled computational pathology platform identifies Barrett’s tissue, filters out irrelevant material that can interfere with tissue analysis, and segments defined tissue and cell structures. Within this context, the test measures the expression of nine protein biomarkers plus nuclear morphology, extracting 15 quantitative spatial features that reflect biomarker expression, cell morphology, and spatial relationships. These data are integrated into a locked algorithm that generates a 0–10 risk score, assigns a low, intermediate, or high-risk class, and provides a five-year probability of progression to high-grade dysplasia or esophageal adenocarcinoma.
Your study showed that 15 percent of patients classified as low-risk by pathology were actually intermediate- or high-risk by precision testing. How might these findings change surveillance intervals or treatment decisions in routine practice?
CH: NDBE is generally considered low risk, but as Emmanuel explained, some patients will progress within the recommended 3–5 year surveillance interval. Molecular risk stratification via spatial proteomics helps clinicians to identify the at-risk individuals among the NDBE patients.
In the cohort presented at AFS, 15 percent of patients diagnosed as NDBE by pathology were reclassified as intermediate or high risk by the spatial proteomics test, with median five-year progression risks of 9 percent and 16 percent, respectively. Both exceed the 8.5 percent risk associated with expert-confirmed low-grade dysplasia – where guidelines recommend escalation through EET to remove precancerous tissue or intensified surveillance every 6–12 months.
In practice, this means some patients who would ordinarily follow a 3–5-year surveillance schedule might instead be considered for shorter follow-up, an early repeat endoscopy, referral to an expert BE center, or even EET. By uncovering these patients earlier, care decisions can be more closely aligned to individual risk, ultimately supporting better patient outcomes.
The study found that the spatial proteomics test identified NDBE patients with progression risks similar to, or exceeding, those with low-grade dysplasia. How might that influence treatment decisions?
CH: The findings indicate that a subset of patients with NDBE may warrant the same level of attention as those with low-grade dysplasia. When the spatial proteomics test uncovers higher-risk biology than pathology alone suggests, it provides clinicians with objective evidence to support closer surveillance or earlier intervention, which may help target therapy to those who need it most.
From the patient’s perspective, how could individualized risk stratification improve quality of care?
CH: Traditionally, risk factors can be difficult for patients to interpret. Having a report that clearly states a patient’s five-year progression risk helps them weigh their treatment options and feel confident in the path that is chosen. For low-risk patients, it can mean fewer surveillance procedures, less anxiety, and lower healthcare costs. For higher-risk patients, it can enable earlier action at a stage where interventions such as EET are most effective in preventing progression.
Because the test is performed on the same biopsies already collected for pathology, it fits naturally into existing workflows and supports shared, informed decision-making between patients and their physicians.
What steps are needed to integrate spatial proteomics into current endoscopic surveillance workflows, and are clinicians likely to adopt this test?
EG: Integration into existing workflows is relatively seamless. The assay runs on standard FFPE biopsy tissue that pathologists already process, and Castle Biosciences provides collection kits and logistical support to minimize any added burden. Test reports are straightforward: each patient receives a 0–10 risk score, a low, intermediate, or high-risk class, and a five-year probability of progression to high-grade dysplasia or esophageal adenocarcinoma.
Evidence suggests these results can meaningfully impact management. For example, a Geisinger study showed that the spatial proteomics test changed clinical plans for 55 percent of patients tested – escalating management in about 22 percent and de-escalating in roughly 33 percent – helping ensure surveillance and treatment decisions are better aligned to individual risk.
Clinicians and pathologists who use the test regard it as a valuable adjunct to the shared decision-making process with their patients. It does not replace histology, but rather complements it by providing objective, actionable information that can guide surveillance intervals and treatment decisions with greater confidence.
Do you see this kind of precision diagnostics influencing guideline updates for Barrett’s esophagus?
EG: US gastroenterological societies are beginning to recognize the contribution of molecular and spatial testing as an adjunct to traditional pathology. For example, the American Gastroenterological Association’s (AGA) 2024 guideline on endoscopic eradication therapy acknowledged that not all patients with NDBE are low risk and noted the potential of the spatial proteomics test to help identify those who could benefit from earlier intervention.
The 2022 AGA Clinical Practice Update also included best-practice advice for the spatial proteomics test in NDBE, and the 2022 American College of Gastroenterology guideline noted that molecular testing, including spatial proteomics, may outperform histology, particularly for patients without dysplasia.
As more evidence becomes available, these kinds of precision tests could become increasingly relevant in shaping future updates to clinical guidance.
What impact might this type of testing have on esophageal cancer incidence?
EG: The long-term impact lies in prevention. By identifying high-risk biology earlier, patients with NDBE can be considered for timely interventions such as EET, while those at lower risk can safely extend surveillance intervals. This ability to escalate care for patients most likely to progress, and de-escalate for those unlikely to do so, has the potential to lower the incidence of esophageal cancer, improve patient outcomes, and use surveillance resources more efficiently.