P-EGS08 What is the Best Risk Prediction Model for Oesophageal Perforation?

Background Oesophageal perforation is a challenging surgical condition associated with high mortality and morbidity. There is a lack of consensus regarding the optimal treatment strategy, when and whom to operate on. Treatment options include primary repair, t-tube repair, emergency oesophagectomy, endoscopic therapy, and palliation. Whilst many risk prediction models exist, the only specific score to predict mortality in oesophageal perforations is the Pittsburgh Severity Score (PSS). However, there is limited evidence on its validity and even less literature to predict short and long-term morbidity in these patients. Methods We compared and validated commonly used risk prediction models, including the PSS, the National Emergency Laparotomy Audit score (NELA score), the Portsmouth Physiological and Operatic Severity Score for the enumeration of Mortality (P-POSSUM), and the Surgical Outcome Risk Tool (SORT) using a dataset of 83 patients ranging from 2009 to 2021. The power to predict mortality and morbidity was assessed using the comprehensive complication index (CCI). The CCI was calculated using complications for both operative and non-operative cohorts. Results Of the scores assessed, NELA showed the most robust predictive value for in-hospital mortality, 30-day mortality, and 90 mortality (AUROC 0.812, 0.8602, 0.8302, respectively). The PSS also showed significant predictive value for in-hospital mortality, 30-day mortality, and 90 mortality (AUROC 0.792, 0.856, 0.813 respectively). Furthermore, NELA had the strongest correlation between score and CCI (rs 0.644 p Conclusions Despite not being validated for oesophageal pathology, NELA appears to be the optimum scoring model to predict mortality and morbidity for this patient population. This is the first study to compare the efficacy of different risk prediction models in oesophageal perforations and could be used to inform shared decision making and peri-operative outcomes. Further large-scale validation of risk prediction tools is required to corroborate these findings.