Artificial Neural Network Prediction Model Predict The Clinical Prognosis Of Drug-induced Liver Injury

Objective:The artificial neural network method was used to establish the back propagation artificial neural network prediction model for the clinical prognosis of drug-induced liver injury,to understand the factors affecting the clinical prognosis of patients with drug-induced liver injury,and to predict the clinical outcome,and evaluating the importance of MIV values for the relevant influencing factors.In order to increase the related measures to the factors affecting the DILI outcome,and ultimately improve the patient's discharge.Methods:(1)In a top three hospital on June 1,2014-June 1,2017 for patients with liver damage caused by drugs such as "drug-induced liver injury","drug-induced liver disease",from the time of admission to discharge to the hospital,clinical data of patients who met the criteria for inclusion and exclusion were recruited and a database was established.(2)Logistic univariate regression analysis was used to screen out the statistically significant factors.Using them as input variables and outcomes as output variables,an artificial neural network back propagation(BP-ANN)model was constructed.A certain amount of samples were randomly assigned to the training set from all the data,and the remaining data was put into the test set,for training and test model.The fitting degree of the model and the accuracy of the model in predicting the clinical outcome of DILI were evaluated by the area under the receiver operator characteristic curve and accuracy.(3)After the model is trained and tested stable,the patient cases that meet the inclusion exclusion criteria were collected from 1 July 2017 to 1 January 2018,and applied the artificial neural network model to predict the outcome of patients discharged from hospital,and compares the predicted results with actual discharge conditions.(4)The mean impact value(MIV)was used to evaluate the importance of the impact of each influencing factor on the DILI.The MIV value was calculated by BP model and the factors were ranked according to the absolute value.Results:(1)Among the 190 samples,there were 129 males and 129 females,and the ratio of males to females was 1:2.1.The youngest person was 15 years old,the oldest was 86,and the mean age(standard deviation)was 47.96(13.61).The age range of high frequency is 41-50 and 51-60 years old,accounting for 27.89% and 31.05% respectively.There were 23 cases have no obvious clinical symptoms,and only show the unusual liver function,accounting for 12.10% of the constituent ratio.The more common clinical symptoms were jaundice(53.20%),anorexia tired of the oil(51.60%),fatigue(45.80%),nausea(44.70%),irritation(30.00%),etc.In clinical classification,there were 146 cases of hepatocyte injury,accounting for 76.80%.22 cases of cholestasis,accounting for 11.60%;mixed type 22 cases,accounting for 11.60%,with hepatocyte injury are most common type.The traditional Chinese medicine in the main types of drugs causing drug-induced liver injury was 84 cases,accounting for 44.20%.There were 21 cases of psychotropic drugs,accounting for 11.10%.There were 14 cardiovascular drugs,accounting for 7.40%.There were 11 cases of unknown drug use and 101 cases of combined drug use,accounting for 53.20%.The liver injury caused by traditional Chinese medicine was the highest in grade 4,with 39 cases,accounting for 46.43%.(2)Logistic univariate analysis showed that the factors affecting clinical outcomes include patient drug allergy history,body mass index(BMI),gastrointestinal symptoms,with and without jaundice,the first abnormal value of alanine aminotransferase(ALT),the first abnormal value of alkaline phosphatase(ALP)and the first abnormal value of aspartate aminotransferase(AST),the first abnormal value of ?-glutamyl transpeptidase(?-GGT),the first abnormal value of direct bilirubin(DBIL),the first abnormal value of total bilirubin(TBIL)and the first abnormal value of indirect bilirubin(IBIL),serum albumin,peak value of ?-glutamyl transpeptidase(?-GGT),peak value of direct bilirubin(DBIL),peak value of total bilirubin(TBIL)and peak value of indirect bilirubin(IBIL),and severity degree.(3)The results of BP neural network training model showed that the median of 1000 times AUC was 0.8310,the mean was 0.8058,and the median of the model test accuracy(%)was 88.5714,with the mean value of 87.6986.The prediction results showed that 43 of the 50 patients were cured or recovered,and 7 were unhealed or died.In the actual situation,37 cases were cured or recovered in the 50 cases,and 13 cases were unhealed or died.The prediction rate of the model was 84%.(4)The absolute value analysis of the influencing factors MIV,gastrointestinal symptoms,the first abnormal value of direct bilirubin(DBIL),and the first abnormal value of aspartate aminotransferase(AST)had the greatest impact on the clinical outcome of patients with DILI.Conclusion:Gastrointestinal symptoms,direct bilirubin,and glutamate transaminase were the most influential factors in the clinical outcomes of patients with drug-induced liver injury.BP artificial neural network model combining the digestive tract symptoms of the patients with DILI,presence of jaundice and biochemical indicators influencing factors such as establishing prediction model can well predict clinical outcome of liver injury.The predictive results of the model are in accord with the actual results.