Development And Validation Of A Risk Early-warning Machine Learning Model For Acute Abdomen:a Real-World-Date Study

Purpose:We aim to have a knowledge to the performance and efficacy of evidence-based guideline for acute abdomen in the real world and develop a machine-learning algorithm based risk early-warning model to help clinical practitioners to make decisions during the initial assessment.Methods:The study was based on Nanfang Hospital Medical BigData Platform.We included the acute abdomen patients who consulted in the Emergency Department.Clinical features,which be recommended by the evidence-based guideline of acute abdomen,were analyzed and used to train the machine learning model.We validated the model by temporal validation and K-fold validation.We evaluated model performance by calculating the Area Under the Receiver Operating Characteristic Curve(AUROC)with two-sided 95%Confidence Interval(95%CI).Result:From Jan.1 2017 to Dec.31 2017,2594 acute abdomen patients had been included,674(26.0%)patients were at high risk,the total data completeness of the whole 28 clinical features was 52.3%.We Used the LightGBM framework to develop a gradient boosting decision tree model.The mean of external validation AUROC of our model was 0.86±0.01 using extra 306 acute abdomen patients.A sensitivity of 79-81%was achieved when using the preselected threshold of 75%specificity.Conclusion:Our machine learning model may help to improve resource utilization by recognized the high risk acute abdomen patients in the real-time during the initial assessment.