Construction Of XGBoost Model In Patients With Acute Paraquat Poisoning And Its Forensic Significance For Individual Evaluation And Prediction

Objective: This study aims to establish an e Xtreme Gradient Boosting(XGBoost)model that can prejudge the body condition of patients with acute paraquat poisoning,the predictive effects of the XGBoost model and COX proportional hazard regression model on individual acute paraquat poisoning were compared,so as to provide an objective basis for the forensic identification of the physical state of the survivors of acute paraquat poisoning and the forensic analysis of the cause of death of the victicms,to help forensics judge the damage status of the identified person in paraquat poisoning cases to provide reference for case investigation.Methods: A total of 150 patients with acute paraquat admitted to our hospital and partner hospitals from January 1,2018 to December 31,2020 were selected as the research objects,according to the(Expert Consensus on the Diagnosis and Treatment of Acute Paraquat Poisoning).The subjects were divided into training group and verification group by random number table method,75 cases in the training group were mainly used for training model,75 cases in the validation group were mainly used to validate model.the patient’s sex,age,paraquat dose,perfusion time,first vomiting time,and duration of taking poison to visit,while immediately taking venous and arterial blood were recorded,while immediately taking venous and arterial blood,testing forwhite blood cell(WBC),blood urea nitrogen(BUN),serum creatinine(Scr),and arterial blood lactic acid(Lac),creatine kinase isoenzymes(CK-MB),blood glucose(Glu),blood kaliemia(K+),blood calcium(Ca2+),alanine aminotransferase(ALT),aspartate aminotransferase(AST).Assess the predictive power of the patient’s clinical physiological data on the death of the patient,and use univariate and multivariate Cox regression models to screen possible predictors.The predictive efficacy of the XGBoost model and the Cox regression model on the survival of patients with acute paraquat was analyzed by the Receiver operating characteristic curve(ROC),and the size of the predictive power was determined by the area under the curve(AUC),and Internal verification was performed in the verification group(75 cases).Results: In this study,there were 150 patients with acute paraquat poisoning,including 75 in the training group and 75 in the verification group.The specific data of the training group and the verification group and the comparison between the two groups showed that there was no significant difference in clinical pathological data between the two groups(P>0.05).There was no significant difference in the survival curve of acute paraquat patients between the training group and the validation group(P>0.05).Cox regression analysis results of acute paraquat death in the training group showed that the intake of paraquat dose,the time from taking poison to seeing a doctor,the time for the first perfusion,the time for the first vomiting,AST,ALT,Scr,BUN,WBC,Lac,CK-MB,Glu,Ca2+,K+ were independent predictors of patient death factor.The XGBoost model for predicting the death of acute paraquat patients was successfully constructed,and 14 predictive factors were obtained.The predictive power,from the largest to the smallest,is the time from taking poison to seeing a doctor,the time for the first vomiting,the time for the first perfusion,Lac,WBC,the dose of paraquat ingested,Scr,K+,Ca2+,CK-MB,Glu,AST,BUN and ALT.Compared with the Cox regression model,the XGBoost model has a larger area under the curve(AUC)for predicting death,and it has also been confirmed in the validation cohort.Conclusions: This study successfully established and verified an XGBoost model that can predict death in patients with acute paraquat.The XGBoost model can be used for predictive analysis of deaths in patients with acute paraquat poisoning,and its predictive power is superior to Cox regression models in both training and validation sets.The XGBoost model can be used to evaluate the correlation strength between risk factors and outcomes,its clinical indicators are conventional,inexpensive,and readily available,one of its unique advantages is that it can get the importance score of each,it can assist clinicians to formulate personalized treatment plans for patients and predict the prognosis of patients.This study also contributes to the analysis of causes of death in clinical forensic medicine and the evaluation of the physical status of survivors.