Establishment And Verification Of An-Early Warning Scoring Model For Acute Heart Failure In Infants

Object:To establish an early warning scoring model for acute heart failure in infants and to compare the accuracy of logistic regression analysis model and neural network model in predicting the outcome of Acute Heart Failure(AHF)in infants.Methods:151 infants with high risk factors of acute heart failure were admitted to CICU and PICU of Women and Children’s Hospital,Qingdao University from January 2012 to October 2020.The clinical data were collected and the binary logistic regression was applied to analyze the influencing factors.The early infant warning score model based on ROC curve was established.The early infant warning scoring model,binary logistic regression analysis model,Multilayer Perceptron(MLP)neural network model and Radial Basis Function(RBF)neural network model were established respectively for prediction.The accuracy and specificity of each model were compared by Area Under the ROC Curve(AUC).Results:1.Establishment of an early warning scoring model for acute heart failure in infants : The factors related to AHF attack were analyzed by Logistic regression analysis of age,gender,blood oxygen saturation,body temperature,heart rate,respiratory rate,urine volume per hour,emotional state.The results showed that heart rate(OR: 1.014,95% CI: 1.002-1.026,P<0.05),respiratory rate(OR: 1.075,95% CI:1.031-1.121,P<0.05),blood oxygen saturation(OR: 0.821,95% CI: 0.809-0.963,P<0.05),urine volume per hour(OR: 0.985,95% CI: 0.981-0.989,P<0.05),emotional state(OR: 1.519,95% CI: 1.065-2.168,P<0.05)were all high risk factors for AHF.Super-score early warning scoring model for infants was established,which was composed of five indexes: Blood oxygen saturation,urine volume,pulse,emotional state and respiratory rate.2.Validation of an early warning scoring model for acute heart failure in infants : The Logistic regression model had a prediction accuracy of 93.16% for AHF attack,85.45% for non-AHF attack and 90.70% for overall prediction,and the Area Under the ROC Curve(AUC)was 0.920.The MLP neural network model had a prediction accuracy of 94.02% for AHF attack,56.36% for non-AHF attack and81.98% for overall prediction,and the Area Under the ROC Curve(AUC)was 0.909.The RBF neural network model had a prediction accuracy of 83.76% for AHF attack,89.09% for non-AHF onset and 85.47% for overall prediction,and the Area Under the ROC Curve(AUC)was 0.917.3.The MLP neural network model had a prediction accuracy of 92.31% for AHF attack,80.00% for non-AHF attack and 88.37% for overall prediction,and the Area Under the ROC Curve(AUC)was 0.923.The RBF neural network model had a prediction accuracy of 86.32% for AHF attack,72.73% for non-AHF onset and81.98% for overall prediction,and the Area Under the ROC Curve(AUC)was 0.873.Conclusions:1.Establish an early warning Super-score model of acute heart failure in infants,which consists of five indexes: blood oxygen saturation,hourly urinary output,heart rate,emotional state and respiratory rate.2.The binary logistic regression analysis model,MLP neural network model and RBF neural network model were applied to validate the early warning score model for infant acute heart failure,and the accuracy and specificity of the early warning score for infant acute heart failure were high.