Pre- and Post-Treatment Predictive Models of Peritoneal Recurrence After Fluorouracil, Leucovorin, Oxaliplatin, and Docetaxel-Chemotherapy and Surgery: An International Study

PURPOSE – Locally advanced gastroesophageal adenocarcinomas commonly relapse early with peritoneal disease, suggesting that a subset of patients harbor occult peritoneal micrometastases at diagnosis. These patients may benefit from peritoneal-directed therapy in the perioperative setting. To facilitate patient selection and evaluation of such therapies, we derived, validated, and compared prediction models for peritoneal recurrence after fluorouracil, leucovorin, oxaliplatin and docetaxel (FLOT)–based multimodality treatment using pre- and post-treatment clinicopathologic predictors. METHODS – A total of 2, 240 patients from the international Survival and Patterns of Care in the Era-FLOT registry were analyzed. Using Fine-Gray competing-risk regression, patients who developed peritoneal recurrence were compared with those with nonperitoneal recurrence, those who died without recurrence, and those who remained disease-free at follow-up. Pre- and post-treatment variables, obtained at staging and post-FLOT/surgery, respectively, were used to construct pre- and post-treatment predictive models of peritoneal recurrence, with internal validation using bootstrap optimism correction. RESULTS – Peritoneal recurrence was the most frequent (41% of all recurrences) and earliest site of disease relapse (median 9.8 v 11.4 months, P =.024) and was associated with poorer postrecurrence (median 4.4 v 9.8 months, P <.001) and overall (median 17.0 v 24.3 months, P <.001) survival compared with nonperitoneal recurrence. Six pretreatment and eight post-treatment variables independently predicted peritoneal recurrence and were incorporated into pre- and post-treatment models, respectively. At 12, 24, and 36 months postsurgery, both models demonstrated good discriminatory performance in predicting peritoneal relapse with comparable accuracy and risk calibration profiles. Decision curve analysis found that both models were superior to treat-none and treat-all approaches, highlighting their potential clinical utility. CONCLUSION – Peritoneal recurrence remains a common and important problem. We derived pre- and post-treatment prediction models for peritoneal recurrence, also available as web-based calculators. These tools can clinically prognosticate and may advance peritoneal-directed therapies for high-risk patients.