Abstract

Abstract Aims Decision-making in the management of paraoesophageal hernias is complex. Surgery carries the risk of significant complications yet a ‘watch-and-wait’ approach can result in the need for emergency surgery. The aim of this study was to create computerised models that makes personalised predictions of morbidity and mortality for patients undergoing both elective and emergency surgery for their paraoesophageal hernia. Method Using AgenaRisk software, Bayesian Network models were created that made personalised predictions of postoperative morbidity and mortality. Bayesian Networks are based on probability theory and have been widely used in military intelligence and finance to perform risk assessment and predict outcomes across competing strategies. Through Baye's theorem, they can explicitly represent the conditional probability of dependencies between multiple variables. The prior probabilities within these model algorithms were calculated from published studies that identified variables associated with morbidity and mortality following paraoesophageal hernia repair (n=22483). The model's performance was validated against a prospectively maintained dataset of 35 patients. Results The model had an area under the receiver operator curve of 0.875 (P value 0.0027; standard error 0.125, 95% confidence interval 0.717 – 0.962) and a sensitivity of 75% and specificity 100% for predicting significant postoperative morbidity (Clavien-Dindo grade 3+). There were no cases of 30day or 90day mortality in the dataset and the model correctly predicted this. Conclusion This study marks a move towards the delivery of more personalised realistic medicine by potentially supporting better shared decision-making. The predictive performance of these models was strong but further prospective validation is required.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call