Application Of Logistic Regression To Construct A TCM Model Of Sepsis Based On Syndrome Factor

Background:Sepsis has become a leading cause of non-coronary artery diseaserelated deaths in the intensive care unit（ICU）.Due to the lack of specific biomarkers,diagnosis is often delayed.The "2021 International Guidelines for Sepsis and Septic Shock" strongly recommend implementing quality improvement programs for sepsis patients in medical settings,including screening for sepsis in emergency and high-risk patients.Methods:By accessing the Intensive Care Management System of Guang’anmen Hospital,China Academy of Chinese Medical Sciences,critically ill patients aged 18 or above,admitted to the ICU for more than 24 hours with complete relevant laboratory information,between January 1,2019,and March 31,2022,were selected.Patient information,including patient ID,hospital number,name,gender,and age,was collected.Disease-related information included the patient’s admission time,admission symptoms and signs,time of sepsis diagnosis,symptoms and signs at the time of sepsis diagnosis,and underlying diseases.Sepsis-related laboratory indicators such as PCT,CRP,WBC,NEUT%,PLT,Lac,D-Dimer,etc.,and relevant scoring systems such as APACHE II score,SOFA score,and GCS score were also collected.Based on the patient’s symptoms,the traditional Chinese medicine（TCM）patterns were categorized,and a logistic regression model was used to fit multiple combined diagnostic indicators or TCM pattern predictive factors to construct a new combined predictive factor.The receiver operating characteristic curve（ROC）was used to compare the combined predictive factor with the area under the ROC curve（AUC）of each original indicator.A simplified model was established using Lasso regression and penalty coefficient λ,followed by model refitting and calculation of sensitivity,specificity,and predictive accuracy.Results:A total of 218 patients were included,of which 114 were sepsis patients and 104 were controls without sepsis despite infection.There were 123 males and 97 females,with an average age of（74.16±14.32）years.Symptom and pattern analysis of the two groups of patients revealed a high incidence of lung pattern,phlegm pattern,heart-brain pattern,qi stagnation pattern,and yin deficiency pattern in sepsis patients.The main associated symptoms included lung（asthma and cough）,phlegm（thick and copious sputum）,heart-brain（delirium and gradual coma）,qi stagnation（delirium,gradual coma,and urinary retention）,and yin deficiency（long-term low-grade fever and frequent constipation）.Statistical analysis identified 15 clinical indicators and TCM patterns as potential risk factors for sepsis.Multiple logistic regression analysis was used to determine independent risk factors and enter them as covariates in the logistic regression model.The regression equation for the full-variable model was as follows:Log（y/1-y）=0.02053X1+0.0375X2+0.3769X3+0.04799X4-0.2186X5+0.006319X6-0.6506X7+0.0003482X8-0.01583X9-1.909X10-0.1897X11+2.726X12-0.8736X13-12.69X14.The regression equation for the pure pattern model was as follows:Log（y/1-y）=-0.234X11-0.9798X14+1.0267X10（X1-X14 represent body temperature,heart rate,WBC,NEUT%,PCT,CRP,INR,APTT,PO₂,excessive heat,large intestine,Qi stagnation,dying Yang,toxin）.The pure syndrome model regression equation is as follows:Log（y/1-y）=-0.234X11-0.979 8X14+1.0267X10（X1X14 respectively represent body temperature,heart rate,WBC,NEUT%,PCT,CRP,INR,APTT,PO₂,fiery heat,large intestine,qi stagnation,death of yang,and poison）,the syndrome element fire heat in the early diagnosis of sepsis is a positively correlated independent risk factor,and the syndrome elements large intestine,poison and sepsis are correlated and have no linear relationship.The critical values are 0.7124,0.3176;the sensitivity is 36.4%,50%;the specificity is 76.5%,64.7%.The positive predictive values were 36.4%,50%,and the negative predictive values were 76.5%,64.7%.Both models have their own advantages.Model 1 is better than model 2 in terms of AIC,training set accuracy,test set accuracy,training set sensitivity,training set specificity,and training set AUC.Model 2 is slightly better than Model 1 in terms of test set sensitivity,test set specificity,and test set AUC.Overall,model 1 performed better.Conclusion:Excessive heat pattern,large intestine pattern,and toxin pattern are independent risk factors for early sepsis diagnosis.The numerical value of the pure pattern regression equation Log（y/1-y）=-0.234X11-0.9798X14+1.0267X10,which is greater than 0.3176,should be considered a risk for sepsis（toxin-heat pattern）.As a TCM sepsis diagnostic model,it assists clinical doctors in early sepsis diagnosis,provides understanding of the disease from a TCM perspective,and guides treatment.However,further research is required to validate its clinical value and update and optimize the model.