Prediction Of Fatal Adverse Prognosis In Emergency Critically Ill Patients With Fever

Objective1.To analyze and study the relationship between age,maximum body temperature and fatal adverse prognosis of febrile patients in emergency room using big data extraction and analysis technology.2.To explore the key factors affecting the fatal poor prognosis of patients with fever us-ing big data analysis technology,and establish an effective model for predicting the fatal poor prognosis of patients with febrile diseases,in the hope of providing technical support for clin-ical diagnosis and treatment decision-making.3.To explore the key factors affecting the fatal poor prognosis of patients with infectious fever using big data analysis technology,and establish the poor prognosis prediction model of patients with infectious fever using machine learning method.MethodsA retrospective study was conducted based on the clinical data from the Emergency Rescue Database of the Chinese PLA General Hospital,covering febrile patients diagnosed and treated from November 2014 to March 2018.Part 1:The clinical data of febrile patients in the emergency database were retrospectively extracted and analyzed by structured query language.The febrile patients were divided into four groups according to their highest body temperature:low fever,medium fever,high fever,and ultra-high fever.Then the highest body temperature of 38.5℃was used as the dividing temperature to assign the patients into two groups.We explored the relationship between highest body temperature and the clinical prognosis of the febrile patients during emergency treatment.Part 2:Prediction model of fatal adverse prognosis in patients with fever-related diseases was constructed based on machine learning.Patients were divided into the fatal adverse prognosis group and the good prognosis group.The commonly used clinical indicators were compared between the groups.Recursive feature elimination（RFE）method was used to determine the optimal number of the included variables.Logistic regression,random forest,adaboost and bagging were selected as the training models.We also collected the emergency room data from December 2018 to Decem-ber 2019 with the same inclusion and exclusion criterion to validate and evaluate the perfor-mance of the newly established models.Part 3:The prediction model for the poor prognosis of patients with infectious fever was established after a further analysis of the four machine learning methods of logistic regression,decision tree,adaboost and bagging method.The performance of the model was evaluated by accuracy,F1-score,precision,sensitivity and the areas under receiver operator characteristic curves（ROC-AUC）.Results1.There was no significant difference in age between good prognosis group（60.64±20.58 years）and poor prognosis group（62.13±18.38 years）,with t value=-1.524,P=0.128.The highest body temperature of the febrile patients was mainly characterized by low fever,medium degree and high fever,and the proportion of patients with ultra-high fever was small.The difference in clinical prognosis between the four body temperature groups was statisti-cally significant（P<0.05）.The highest body temperature of 38.5℃was used todividethe pa-tients into two groups.The difference in clinical prognosis between the two groups was statis-tically significant（P<0.05）.2.The accuracy of logistic regression,decision tree,adaboost and bagging established for all febrile patients was 0.951,0.928,0.924,and 0.924;F1-scores were 0.711,0.672,0.672,and 0.840;the precision was 0.943,0.938,0.937,and 0.937;and ROC-AUC were 0.808,0.738,0.736,and 0.885,respectively.ROC-AUC of tenfold cross-validation in logistic and bagging models were 0.80 and 0.87,respectively.The top six coefficients and odds ratio（OR）values of the variables in the Logistic re-gression were cardiac troponin T（CTn T）（coefficient=0.346,OR=1.413）,temperature（T）（coefficient=0.235,OR=1.265）,respiratory rate（RR）（coefficient=–0.206,OR=0.814）,serum kalium（K）（coefficient=0.137,OR=1.146）,pulse oxygen saturation（SPO₂）（coefficient=–0.101,OR=0.904）,and albumin（ALB）（coefficient=–0.043,OR=0.958）.The weights of the top six variables in the Bagging model were:CTn T,RR,lactate dehydrogenase,serum amyl-ase,heartrate,and systolic blood pressure.3.In the study of patients with infectious fever,in the process of model training,the ac-curacy of logistic regression,decision tree,adaboost and bagging were 0.971,0.941,0.956 and0.956;the F1-score was 0.447,0.343,0.522and0.344;the Precision was 0.971,0.937,0.951 and0.951;and ROC-AUC were 0.798,0.593,0.796and 0.902,respectively.The logistic,adaboost and bagging models that showed better comprehensive efficiency were cross validated.Their ROC-AUC were 0.81,0.81 and 0.85,respectively.ConclusionsThe main clinical indicators of concern were:CTn T,RR,SPO₂,T,ALB and K.The bag-ging model and logistic regression model had better comprehensive diagnostic performance,which may be conducive to the early identification of critical patients with fever by physi-cians.Big data analysis method was adopted to establish a scientific and objective prediction and evaluation model for adverse prognosis of patients with fever in this study.We identified the main clinical indicators of concern and established the prediction model with high diag-nostic accuracy and reliability,which may be beneficial for physicians to identify early criti-cal patients with fever,thus improving their prognosis.The combination of big data analysis with medical research is helpful to improve the diagnosis and treatment level of febrile critical diseases as well as the prevention and control of infectious diseases.Research on adverse event prediction model for critical patients with fever quantifies the recognition of critical diseases related to fever and provides a reference model for other similar clinical decision support studies.