| Background: Acute aortic dissection is an extremely fatal cardiovascular disorder characterized by rapid progression and high mortality.In recent years,diagnostic technology and treatment of acute aortic dissection have improved significantly,but nearly 20% of patients still die during hospitalization,which brings a heavy burden to patients and their families.At present,there have been many studies related to prognosis prediction of acute aortic dissection.However,due to the limitations of traditional prediction methods in terms of prediction performance and development potential,the clinical application of existing prediction is not yet mature,so more advanced prediction methods still need to be explored.With the development of artificial intelligence,machine learning algorithms have begun to attract attention in various fields because of their excellent predictive ability and great development potential.However,there is no relevant research on acute aortic dissection.Objective: Applying machine learning algorithms,a clinical datadriven optimal predictive model was developed to predict the risk of inhospital mortality in patients with acute aortic dissection.Methods: Patients diagnosed with acute aortic dissection between January 2015 to December 2018 were voluntarily enrolled from the Second Xiangya Hospital of Central South University in the study.The diagnosis was defined by magnetic resonance angiography or computed tomography angiography,with an onset time of the symptoms being within 14 days.The analytical variables included demographic characteristics,physical examination,symptoms,clinical condition,laboratory results,and treatment strategies.The machine learning algorithms included logistic regression,decision tree,K nearest neighbor,Gaussian naive Bayes,and extreme gradient boosting.Evaluation of the predictive performance of the models was mainly achieved using the area under the receiver operating characteristic curve.Finally,we used feature importance score,Shapley additive explanation(SHAP)for global interpretation of the final prediction model,and local interpretable modelagnostic explanation(LIME)for local interpretation.Results: A total of 1,344 acute aortic dissection patients were recruited,with an average age of 52.37±11.73 years,including 1080 male patients(80.36%)and 273 patients(20.3%)in the non-survival group.After testing and comparison,we found that the extreme gradient boosting model has the best prediction performance,which was identified as the final model,with the largest area under the receiver operating characteristic curve(0.927,95%CI: 0.860～0.968),and the highest accuracy(91.8%,95%CI: 0.838-0.998)),mean precision(68.3%,95%CI: 0.400-0.966),sensitivity(72.9%,95% CI: 0.457-1.000),specificity(96.6%,95% CI: 0.908-1.000),positive predictive value(0.855,95% CI:0.627-1.000)and negative predictive value(0.934,95% CI: 0.869-0.999).In the feature importance score ranking,treatment modality,Stanford classification of acute aortic dissection,and admission ischemia-modified albumin level were the three most important variables.Through Shapley additive explanation,we further found that characteristics such as conservative medical treatment,Stanford type A acute aortic dissection,and higher admission ischemia-modified albumin levels significantly increased the risk of in-hospital mortality.The results of local interpretable model-agnostic explanation are basically consistent with the results of feature importance scores and Shapley additive explanation.Conclusion: The extreme gradient boosting model is the best predictive model for in-hospital mortality in patients with acute aortic dissection,and the choice of patient treatment,Stanford classification and ischemia-modified albumin levels have the greatest impact on the prediction results of the model.The prediction model not only helps to improve the identification efficiency of patients at high risk of death,but also provides a certain reference for clinicians to make clinical decisions. |
PDF Full Text Request