Risk Assessment And Prediction Of Cardiac Surgery For Critical Patients Based On Machine Learning

Surgery is an invasive way to treat diseases,but sometimes trauma can lead to the deterioration of the patient’s physical function,and even lead to the failure of vital organs,which ultimately defeats the original purpose of surgery.Patient surgery risk assessment plays an important role in clinical surgery.It is of great significance for the determination of surgical indications,the identification of operation-related risk factors,the establishment of scoring criteria and the comparison of surgical efficacy between different centers.This study focuses on the risk of cardiac surgery in severe patients.Clinical studies have found that the duration of ICU treatment fundamentally affects the prognosis of patients.For patients undergoing ICU surgery,it is more necessary to timely and accurately assess their condition and prognosis.The length of hospital stay was divided into 3,7 and 14 days,Logistic regression,random forest and Gradient Boosting Decision Tree（GBDT）models were constructed to predict ICU treatment time.GBDT model has the best prediction effect,with accuracy of 0.712 and Kappa of 0.333.Lasso regression was used to screen the important characteristics affecting the length of hospital stay,and it was found that the number of surgical procedures,the number of diagnosed diseases,lactate and blood oxygen saturation had a significant impact on the length of hospital stay,which was consistent with the results of relevant medical studies.At the same time,white blood cell count,carbon dioxide partial pressure,body temperature and other factors that have no influence on the length of ICU stay were excluded.Furthermore,the GBDT model was further trained with Lasso-screened features,which maintained a good prediction effect compared with the GBDT model trained with all features.The results indicate that the Lasso regression feature screening results have clinical significance and GBDT model can accurately predict the length of stay of ICU patients in cardiac surgery.With the improvement of medical level and the development of medicine,the prediction efficiency of traditional disease scoring system is getting worse and worse.Therefore,it is necessary to explore the prediction model of death risk of patients undergoing severe cardiac surgery based on machine learning method.For the problem of unbalanced classification caused by a small proportion of dead patients in the included data,a cost-sensitive Adaboost algorithm was proposed to make the cost of misclassification of dead patients higher than that of survival patients,making the model more focused on the recognition of dead patients.The results showed that compared with the original algorithm,the improved Adaboost algorithm only sacrificed precision,and the recall increased by 0.201,The F1-score increased by 0.154,and the AUC increased by 0.097,which was more conducive to the identification of ICU death risk in cardiac surgery patients.