Constructing The Diagnosis Models Of The Syndrome From The NAFLD Patients Based On The Traditional Chinese Medicine Theory

Syndrome is the basis and key of TCM to recognize and treat diseases.Objectification of TCM syndromes has always been a hot and difficult point in TCM syndromes research.In recent years,many scholars have applied machine learning to this field and made some progress,providing new methods and ideas for the study of TCM syndrome objectification.This study intends to carry out a cross-sectional clinical study of TCM syndromes in patients with non-alcoholic fatty liver disease(NAFLD)to analyze the correlation between TCM syndromes and clinical detection indicators such as liver function and blood lipid.Decision tree,random forest,support vector machine and extreme gradient lifting algorithm in machine learning were used to construct NAFLD TCM syndrome diagnosis model,and the diagnostic accuracy of the diagnostic models constructed by different methods was evaluated to select the best model.The established diagnostic model is expected to provide a simple,objective and reliable method for the diagnosis of TCM syndromes of NAFLD,and promote the development of objectification of TCM syndromes of related diseases.Result:1.The distribution of TCM syndromes in 335 NAFLD patients was as follows: in the case of combined syndromes,according to the percentage of syndromes,from high to low,the syndromes were phlegm-dampness internal obstruction(38.21%),liver depression and spleen deficiency(35.82%),dampness and heat internal accumulation(24.48%),liver-kidney Yin deficiency(21.19%),and phlegm-stasis interaction(15.22%).2.The correlation between TCM syndromes and clinical indicators of NAFLD: Liver depression and spleen deficiency syndrome was negatively correlated with low density lipoprotein and positively correlated with glutathione reductase.The dampness and heat content was positively correlated with alanine aminotransferase and white blood cell count.The interaction between phlegm and blood stasis was negatively correlated with uric acid and alanine aminotransferase and positively correlated with apolipoprotein A1.The deficiency of liver and kidney Yin was negatively correlated with glutathione reductase and homocysteine.The above correlations were significantly correlated(P<0.05).3.Two models were constructed respectively by using extreme gradient elevation algorithm with the information of the four diagnoses of traditional Chinese medicine and the information of the four diagnoses of traditional Chinese medicine + biochemical indexes as variables.The results showed that the accuracy of XGboost algorithm in Model 1 was72.50%,80.20%,75.25%,86.14%,80.20% for phlegm-dampness internal obstruction syndrome,liver depression and spleen deficiency syndrome,damp-heat internal accumulation syndrome,phlegm-stasis mutual syndrome and liver-kidney Yin deficiency syndrome,respectively.The accuracy of decision tree algorithm for each syndrome was 66.34%,71.30%,76.24%,78.22%,72.28%,respectively.The accuracy of random forest for each syndrome was 68.32%,81.19%,72.28%,84.16%,78.22%,respectively.The accuracy of SVM for each syndrome was 67.33%,78.22%,76.24%,83.17% and 79.21%,respectively.In model 2,the accuracy of XGboost algorithm for the above five syndroms was 67.33%,82.18%,75.25%,85.15% and 78.22%.The accuracy of decision tree algorithm for each syndrome was 63.37%,77.23%,75.25%,80.20% and76.24%,respectively.The accuracy rate of random forest for each syndrome was 67.33%,75.24%,77.23%,87.13% and 75.25%,respectively.The accuracy of each syndrome was 71.28%,84.16%,73.27%,83.17% and76.24%,respectively.4.The diagnostic items of all syndroms obtained in Model 1 were sorted according to the contribution rate: The syndroms of Liver depression and spleen deficiency were hyperpleasure,shortness of breath and lethargy,and depression;The syndrome of internal obstruction of phlegm and dampness was head weight like wrapping.Dampness-heat internal resistance syndrome is moss yellow,halitosis,urine yellow;Mutual syndrome of phlegm and blood stasis is chest pain and dark red tongue.The deficiency of liver and kidney Yin is tinnitus and lumbar acid.5.The diagnostic items obtained in Model 2 were sorted according to the contribution rate: The syndrome of liver depression and spleen deficiency was triglyceride,low density lipoprotein,waist circumference,shortness of breath and lazy speech,hyperpleasure and depression;The symptoms of internal obstruction of phlegm and dampness were BMI,triglyceride,head weight such as wrapping,phlegm,dry mouth and pharynx,and body weight.The dampness and heat accumulation syndrome included urea nitrogen,moss yellow,free fatty acid,halitosis,very low density lipoprotein,white blood cell count.Mutual syndrome of phlegm and blood stasis included ALT,uric acid,apolipoprotein A1,dark red tongue,multiple dreams and bad breath.Liver and kidney Yin deficiency included white blood cell count,fasting blood glucose,glutathione reductase,homocysteine,waist circumference and AST.6.External verification of the diagnostic model: THE AUC values of the constructed model were 0.72,0.71,0.62,0.55 and 0.56 for the syndrome types of phlegm-dampness internal obstruction,liver-depression and spleen-deficiency,damp-heat internal accumulation,phlegm-stasis interaction and liver-kidney Yin deficiency,respectively.Conclusions:In this study,XGboost algorithm was used to preliminarily build a diagnosis and prediction model of TCM syndromes of NAFLD,which has high accuracy and can further improve the diagnostic accuracy of TCM syndromes of NAFLD.Moreover,it is simple,practical and easy to be popularized,and can assist clinicians in syndrome differentiation.It has solved the key scientific problem of objectification of NAFLD syndrome diagnosis in traditional Chinese medicine to a certain extent and has a good application prospect.In this study,the syndrome diagnosis items in model 1 and Model 2are ranked according to their importance,and it is concluded that the symptoms and signs items that contribute most to syndrome diagnosis in model 1 are consistent with clinical practice,providing objective basis for further establishing the diagnostic criteria and weight of syndromes.In model 2,it was found that the biochemical indexes ranked the top six in the order of importance of each syndrome diagnosis item,and showed the correlation between different syndromes and different clinical indexes,indicating that it was feasible to apply extreme gradient elevation algorithm to the study of TCM syndromes and biochemical indexes of NAFLD in methodology.The successful construction of the model provides clinical clues for further standardization of TCM syndromes of NAFLD and further research on the biological basis of syndromes.