Perioperative Complications Of Elderly Patients With Hip Fractures Predicted By Artificial Neural Network

Objective:To retrospectively analyze the risk factors of perioperative complications that affect the elderly patients with hip fracture,and to establish a Logistic regression model and artificial neural network model to predict。To make a new way to manage risk factors of hip fracture.Methods:Collecting843 cases of femoral neck fracture or femoral intertrochanteric fracture treated from August 2013 to March 2018 in the First Affiliated Hospital of Guangzhou University of Traditional Chinese Medicine.A total of 48 variables such as basic information of patients,preoperative basic diseases,operation conditions and postoperative complications were recorded.The data were compared and statistically analyzed by using software such as Python,SPSS and Matlab.BP neural network model and convolution neural network model and Logistic regression model was built by SPSS and Maltab.Finally,the area under the ROC curve and the prediction accuracy are compared by calculation to show the performance of the three models.Results:In this study,843 cases of surgical treatment of hip fracture in the elderly were included.Of the 843 cases in this study,perioperative complications occurred in 193 cases.The incidence of neurological complications was 3.21%,including 14 cases of cerebral infarction,2 cases of cerebral hemorrhage,11 cases of abnormal consciousness,and 27 cases in total.The incidence rate of cardiac complications was 2.49%,of which 3 cases had malignant arrhythmia and 7 had heart failure.Cases,myocardial infarction in 10 cases,cardiac arrest in 1 case,a total of 21 cases;respiratory complication rate was 12.23%,including 100 cases of pulmonary infection,pulmonary embolism in 5 cases,1 case of respiratory arrest,a total of 103 cases;urinary The incidence of systemic complications was9.03%.There were 75 cases of urinary tract infections and 1 case of acute renal failure;the incidence of digestive complications was 0.59%,2 cases of gastrointestinal bleeding,2cases of intestinal obstruction,and 1 case of biliary tract infection.There were 5 cases in total;other complications,shock 1 Cases,diffuse intravascular coagulation in 1 case.A total of 13 deaths were reported.Among them,5 were pulmonary infection,4 heart failure,1 cardiac arrest,1 myocardial infarction,1 pulmonary embolism,and 1 diffuse intravascular coagulation.Multivariate analysis showed that there were respiratory basic diseases,general anesthesia,B-type natriuretic peptide> 100 pg/m L,history of hypertension,age ≥ 80 years,operative time> 7 days,preoperative daily average heart rate>80 times /h,urine white blood cell count> 40,aspartate aminotransferase> 32 U/L,etc.are risk factors for perioperative complications in elderly patients with hip fractures.For different assessment methods,Logistic regression model has an accuracy of 80.0% in predicting training samples,an area under the ROC curve of 0.743(95% CI: 0.701-0.784),and a BP neural network prediction accuracy rate of 82.6% in the artificial neural network.The area under the curve is 0.825(95% CI: 0.789-0.860).The prediction accuracy rate of convolutional neural network was 88.4%,and the area under the ROC curve was 0.901(95%CI:0.875-0.927).The area under the curve was compared between the logistic regression model and the BP neural network(P=0.003),between the logistic regression model and the convolutional neural network(P<0.001),between the BP neural network and the convolutional neural network(P<0.001),all statistically significant.Convolutional neural networks have better prediction performance than BP neural networks and logistic regression models.The BP neural network prediction effect is better than the Logistic regression model.Conclusion:The risk factors for perioperative complications in elderly patients with hip fractures are: basic respiratory diseases,general anesthesia,B-type natriuretic peptide> 100 pg/m L,history of hypertension,age ≥ 80 years,and timing of surgery> 7 days,preoperative daily average heart rate> 80 beats per min,urine white blood cell count> 40,aspartate aminotransferase> 32 U/L and so on.Convolutional neural network model is superior to logistic regression model and BP neural network model in predicting perioperative complications of elderly hip fracture patients.The logistic regression model has good analytical ability of risk factors.In actual clinical work evaluation,several methods should be combined as much as possible for reference.