SP 9.09 Validation of Machine learning based Predictive OpTimal Trees in Emergency Surgery Risk (POTTER) calculator in laparotomy patients in UK against NELA and NSQIP risk scores

Aims Risk assessment tools routinely used, like NELA and NSQIP, assume that risk factors are linear and cumulative. They are derived by logistical regression models from existing databases. Predictive OpTimal Trees in Emergency Surgery Risk (POTTER) calculator is a novel non-linear risk calculator that was developed with machine learning algorithms. This retrospective study is aimed at validating the use of this score in emergency laparotomies in the UK. Methods All consecutive emergency laparotomies in a single center from October 2023 – November 2025 were retrospectively analysed. The preoperative NELA, NSQIP and POTTER scores were calculated. Calibration plot, Receiver operating characterise curve, c-statistic were calculated. Results Of the 113 emergency laparotomies during the study period, there were 11 deaths (9.7%). Calibration plots were comparable. Brier test scores are NELA: 0.0627, NSQIP: 0.0540 and POTTER: 0.0581. The AUROC (c-statistic) were 0.86 (95% CI 0.76-0.97) for NELA, 0.88 (95% CI: 0.77-0.99) for NSQIP score and 0.9 (95% CU: 0.82-0.98) for POTTER score. The DeLong test curves are NELA vs NSQIP: p=0.317, NELA vs Potter: p=0.332, NSQIP vs Potter: p=0.650. There is no statistically significant difference. All three scores show good accuracy in predicting the actual outcome. Conclusion POTTER is an accurate user-friendly risk assessment tool that can aid decision making for emergency laparotomies and is comparable against other validated tools currently used in the UK. Although, use of machine learning and AI models in deciding futility of emergency surgeries, in very high-risk patients does have ethical implications regarding transparency.