The incidence of gastro-oesophageal junction (GEJ) adenocarcinoma continues to rise worldwide, underscoring the need for evidence-based management of early neoplasia in this anatomically complex region. The GEJ represents the interface between two adjacent organs with distinct treatment recommendations. The presence of Barrett’s epithelium may introduce additional variability in biological behaviour and treatment response. This study aimed to determine predictors of curative resection in Siewert type II GEJ lesions and to compare the outcomes of EMR and ESD performed in this area.

This retrospective two-centre cohort study included consecutive patients who underwent endoscopic resection for early Siewert type II GEJ neoplasia between 2011 and 2025 at Cambridge University Hospital and the Maria Skłodowska-Curie National Research Institute of Oncology in Warsaw. Lesions with macroscopically visible Barrett’s oesophagus were classified as Barrett’s-related. The primary endpoint was curative resection rate, defined as R0 resection (negative deep margin for EMR; deep and lateral margins for ESD), histological grade ≤ G2 and absence of lymphovascular invasion. Secondary endpoints included local recurrence of high-grade dysplasia or adenocarcinoma and surgery within 12 months from endoscopic resection. A multivariable logistic regression model adjusted for age, sex, and Barrett’s status. Comparisons between EMR and ESD, and between Barrett’s- and non-Barrett’s-related lesions, were performed using χ² tests and. A p value < 0.05 was considered statistically significant.

A total of 158 patients were included (mean age 69, ± 10.7 years, 76.6% male). Final histology revealed 137 adenocarcinomas and 21 high-grade dysplasia; 96 EMR and 62 ESD procedures were performed. Curative resection was achieved more frequently in Barrett’s-related neoplastic lesions than in those without Barrett’s (71.7% vs 53.8%; p = 0.026). In the multivariable logistic model, the presence of Barrett’s epithelium was an independent predictor of curative resection (OR 2.54; 95% CI 1.16–5.63; p = 0.020), whereas age, sex, and resection technique were not significant. Lesions without Barrett’s mucosa demonstrated a more aggressive profile, with higher rates of poor differentiation (G3: 21% vs 13%), a greater prevalence of lymphovascular invasion (26% vs 8%) and a higher rate of submucosal invasion (47.6% vs 19.1%), compared with Barrett-related lesions. Among Barrett’s-related Siewert type II lesions (n = 106), curative resection rates were comparable between EMR and ESD (71.2% vs 73.1%; p = 0.86), with similar rates of local recurrence (23.1% vs 23.8%) and additional surgery (5.1% vs 4.2%). There was a trend towards higher curative resection in non-Barrett’s lesions with ESD compared to EMR (61.1% vs 37.5%, p = 0.12) and lower recurrence rate (7.1% vs 25.0%; p = 0.297), although these differences did not reach statistical significance due to limited sample size.

Primary GEJ neoplasia without visible Barrett’s exhibited lower rate of curative endoscopic resection likely in keeping with deeper histological invasion and more frequent lymphovascular invasion compared with Barrett-related lesions. Our data suggest that ESD should be the preferred resection modality for GEJ associated lesion in the absence of Barrett’s epithelium.