Ann Surg Oncol. 2025 Oct 25. doi: 10.1245/s10434-025-18513-0. Online ahead of print.
ABSTRACT
BACKGROUND: Prognostication tools offer a way to combine diverse information and inform personalized survival predictions for patients and their providers. A review of tools aimed at prognostication for patients with esophagus and gastroesophageal junction (GEJ) cancers undergoing surgery did not identify many high-quality tools that may be used.
METHODS: This study developed and externally validated a prognostic model to estimate the probability of dying within 3 years of surgery for patients with resected esophageal or GEJ cancer diagnosed between 2004 and 2016, followed to 2020. We used population-based administrative health and pathology data in Ontario (development) and Manitoba (external validation) from cancer registries, physician billing data, and hospitalization records. Predictor variables included patient (e.g., age, sex), disease (tumor stage, lymph node status), treatment (e.g., extent of surgery, receipt of radiation), and pathology factors (e.g., lymphovascular invasion). Bootstrapped calibration-in-the-large and time-varying area under the curve (AUC) statistics were estimated.
RESULTS: Model development included 2124 patients from Ontario. External model validation included 318 patients from Manitoba. Internal validation demonstrated a calibration plot slope of 1.02, intercept of – 0.01, and AUC of 0.77. In comparison, the external validation reported a calibration plot slope of 1.11, intercept of 0.005, and AUC of 0.73. These results were robust across patient characteristics (e.g., age, sex, income), disease histology, and primary tumor location.
CONCLUSION: Our model demonstrated accurate prognostic capability and may be suitable for application in real-world clinical care. Development of a web-based interface and supporting documentation for communicating risk to personalize prognosis for patients or facilitate shared decision-making is under way.
PMID:41139180 | DOI:10.1245/s10434-025-18513-0
