| Objective:1.Theoretical researchConduct a systematic conceptual analysis of spleen and kidney deficiency syndrome of mild cognitive impairment(MCI-SKDS),and sort out the etiology and risk factors of MCI-SKDS from the perspective of combination of Traditional Chinese Medicine(TCM)and western medicine,so as to better find the relevant factors of MCI from the perspective of combination of disease and syndrome,as well as combination of TCM and western medicine,to provide theoretical support for the prediction factors in the subsequent prediction model.2.Investigation researchScreening the related factors of MCI and MCI-SKDS from general demographic data,cognitive function,activities of daily living,family history,chronic medical history,hobbies,sleep quality,quality of life,TCM syndrome type or TCM symptom score,and laying a foundation for the follow-up prediction model research.3.Clinical researchFrom physical examination,blood and urine indicators,such as BMI,blood pressure,pulse,blood routine,liver and kidney function,fasting blood glucose,blood lipids,urine routine and other indicators,the related factors of MCI and MCI-SKDS were screened,to provide reference for the determination of prediction model laboratory indicators.4.Experimental research on prediction modelThe prediction model of MCI-SKDS is constructed and evaluated by machine learning methods,facilitating rapid identification of the target population.The model is reported based on TRIPOD statement so that to convenient for the promotion and application of the model,thus laying the foundation for the following targeted intervention research,providing ideas for the research of other elderly diseases.Method:1.Theoretical discussionRetrieve the relevant data of MCI-SKDS through multiple ways like database,dictionary and manual,and conduct an eight-step conceptual analysis of MCI-SKDS by integrating Walker and Avant’s classic concept analysis method and Rodgers’ evolutionary concept analysis method;At the same time,we used content analysis,comparison,mind mapping and other methods to sort out the TCM etiology and modern medical risk factors of MCI-SKDS and their relationship.2.Investigation researchUsing the multi-stage cluster sampling method,according to the diagnostic criteria of MCI,the diagnostic criteria of MCI-SKDS,the inclusion and exclusion criteria of the study object,and the screening process,From January 2022 to July 2022,1007 elderly people were selected as the study object from the communities under the jurisdiction of the Wunancun Community Health Service Station in Wuchang District and Hongshan District Hospital of TCM.Through ten tools including the general situation questionnaire,Montreal Cognitive Assessment Scale,the TCM Syndrome Differentiation Scale for MCI,et al.to conduct a questionnaire survey on the elderly,and Excel was used to establish a special database for this study.SPSS 25.0 and Python 3.9 were used for statistical analysis to preliminarily screen the demographic,interest,sleep quality and other related factors of MCI and MCI-SKDS,and initially explore the independent predictive effect of single TCM symptom on MCI-SKDS.3.Clinical researchThe research object was the same as the investigation research.Collect the laboratory indicators such as blood,urine and physical examination of the elderly,and use SPSS 25.0 and Python 3.9 for statistical analysis.Compare the differences in blood,urine and physical examination of the elderly in the MCI group and the normal group,and further increase the derivative value of blood routine and other indicators in the urine test,and analyze the differences between the elderly in the MCI-SKDS group and the MCI non-spleen and kidney deficiency syndrome(MCI-NSKDS)group,Screening relevant laboratory indicators,and initially exploring the independent predictive effect of single laboratory indicators on MCI and MCI-SKDS.4.Experimental research on prediction model312 elderly people with complete indicators were selected from the elderly of MCI-SKDS.Based on the above literature,survey and Clinical research results,comprehensively considering the professional significance,the variables(P<0.10)in the single factor analysis,and excluding multicollinearity,44 variables were initially screened from all variables to establish a separate database,stored in csv format,and imported into Python 3.9.The data was divided into a training set and a validation set in a 3:1 ratio.Five machine learning methods,including logistic regression(LR),decision tree,naive Bayes,support vector machine(SVC)and gradient boosting(GB)method,were used to build and evaluate the diagnostic prediction model of MCI-SKDS.The model was reported transparently based on the TRIPOD statement.Results:1.Research resultsPart Ⅰ:There was statistically significant differences in the nature of work and monthly income before retirement between the normal group and the MCI group(P<0.05);The total mean scores of MMSE and MoCA in the two groups were significantly different(P<0.05);The total mean scores of BADL,IADL and ADL in the two groups were significantly different(P<0.05);There was no significant difference in family history,chronic disease history and the number of chronic diseases between the two groups(P>0.05);There was no significant difference in the number of hobbies and specific hobbies(such as watching TV,walking,listening to radio,etc.)between the two groups(P>0.05);There was no significant difference between the two groups in the average score of Pittsburgh sleep quality index,quality of life,self-assessment of quality of life and the self-assessment of health status(P>0.05);There was statistically significant difference in the scores of spleen and kidney deficiency syndrome and Qi and blood deficiency syndrome between the two groups(P<0.05),there was no significant difference in the other five syndrome types(P>0.05).Part Ⅱ:The difference of sex constituent ratio between the MCI-SKDS group and the MCI-NSKDS group was statistically significant(P<0.05);There was no significant difference between the two groups in the total average score of MMSE and the average score of each cognitive dimension(P>0.05);The total average score of MoCA,the average score of naming and language of MoCA in MCI-SKDS group were lower than those in MCI-NSKDS group((P<0.05);There was no significant difference in total average score of BADL,IADL and ADL between the two groups(P>0.05);There was no significant difference in the composition ratio of family history between the two groups(P>0.05);In the chronic history of metabolic disease,the proportion of MCI-SKDS group was higher than that of MCI-NSKDS group(P<0.05);There was no significant difference in the number of hobbies and specific hobbies(such as watching TV,walking,listening to radio,etc.)between the two groups(P>0.05).The proportion of elderly people using electronic equipment in the MCI-SKDS group was higher,and the difference was statistically significant(P<0.05);There was no significant difference between the two groups in the average score of Pittsburgh sleep quality index,quality of life,self-assessment of quality of life and health status(P>0.05);There was significant difference between the two groups including waist soreness,waist soreness and leg weakness,ache of lumbar spine,tinnitus like cicada,deafness,loose stool or soft stool after initial hardness,cold hands and feet,fear of cold,tooth movement,tooth loss,frequent nocturnal urination,decreased sexual function,pale tongue,thin white fur,thin white greasy fur,white and slippery fur(P<0.05).Waist soreness,waist soreness and leg weakness,ache of lumbar spine,tooth loss,frequent nocturnal urination,decreased sexual function,pale tongue,thin white fur,thin white greasy fur have independent predictive value for MCI-SKDS(P<0.05),and the predictive efficacy was 62.9%,57%,68.9%,65.4%,59.3%,57.4%,63.2%,58.9%,57.9%respectively.2.Clinical research resultsPart Ⅰ:There was significant difference in MCH,MCV,WBC and P-LCC between the normal group and the MCI group(P<0.05);There was no significant difference in fasting blood glucose between the two groups(P>0.05),but there was significant difference in HDL-C in blood lipids(P<0.05);There was no significant difference in liver and kidney function between the two groups(P>0.05);There was significant difference in UPH and UOB between the two groups(P<0.05);There was no significant difference in BMI,waist circumference,blood pressure and pulse between the two groups(P>0.05);MCH,MCV,PCT,P-LCC and UPH have independent predictive value for MCI(P<0.05),The predictive efficacy was 57.1%,57.3%,56.3%,57.3%,and 57.1%,respectively.Part Ⅱ:The difference of PoL,PLR,LMR,SII,NMLR and MLR between the MCI-NSKDS group and the MCI-SKDS group was statistically significant(P<0.05);There was no significant difference in fasting blood glucose and blood lipids between the two groups(P>0.05);There was significant difference in TBIL and dBIL between the two groups(P<0.05);There was no significant difference in renal function between the two groups(P>0.05);There was significant difference in USG between the two groups(P<0.05);There was no significant difference in BMI,waist circumference,blood pressure and pulse between the two groups((P>0.05);PoL,LMR,HGB,PLR,SII,NMLR and MLR have independent predictive value of MCI-SKDS(P<0.05),the predictive efficacy was 57.2%,56.5%,56.8%,57.7%,57.5%,56.9%and 56.5%respectively.3.Prediction model research results① Predictive indicators and their importance:The predictive indicators of the MCI-SKDS Prediction Model screened by machine learning according to importance were ache of lumbar spine,tooth loss,frequent nocturnal urination,pale tongue,thin white greasy fur,cold hands and feet,deafness,decreased sexual function,urine proportion and metabolic disease,the weights were 0.124,0.098,0.058,0.054,0.041,0.033,0.030,0.013,0.011,0.002 respectively.②Model calibration:The Brier scores of the five models(LR,decision tree,naive Bayes,SVM and GB)were 0.135(0.139),0.156(0.151),0.216(0.187),0.151(0.153)and 0.139(0.135)respectively before(after)calibration.③ Confusion matrix:TP,FN,FP and TN values of the five models in training sets(testing sets)were 103-10-15-81(55-9-9-30),110-3-17-79(56-8-8-31),101-12-27-69(51-13-10-29),102-11-13-83(48-16-7-32),105-8-15-81(54-10-9-30),respectively.④Accuracy:The accuracy of training sets(testing sets)of the five models were 0.880(0.825),0.904(0.845),0.813(0.777),0.885(0.777)and 0.890(0.816),respectively.⑤Sensitivity and specificity:the sensitivity of training sets(testing sets)of the five models were 0.844(0.769),0.823(0.795),0.719(0.744),0.865(0.821),and 0.844(0.769),respectively;The specificity of the five models in training sets(testing sets)were 0.912(0.859),0.973(0.875),0.894(0.797),0.903(0.750)and 0.929(0.844),respectively.⑥Precision:the precision of the positive samples of the five models in training sets(testing sets)were 0.890(0.769),0.963(0.795),0.852(0.690),0.883(0.667),and 0.910(0.750),respectively;The precision of negative samples of five models were 0.873(0.859),0.866(0.875),0.789(0.836),0.887(0.873)and 0.875(0.857),respectively.⑦F1 score:The F1 score of the positive samples of the five models in training sets(testing sets)were 0.866(0.779),0.888(0.795),0.780(0.716),0.874(0.736)and 0.876(0.759),respectively;The F1 score of negative samples were 0.892(0.859),0.917(0.875),0.838(0.816),0.895(0.807)and 0.901(0.850),respectively.⑧Area under curve(AUC)of ROC:The AUC of the five models in training sets(testing sets)were 0.944(0.866),0.970(0.820),0.903(0.818),0.941(0.861)and 0.950(0.864),respectively.⑨Decision curve:The decision curve analysis(DCA)of LR,decision tree and GB models shows that the model has good clinical application value.Conclusion:①Concept analysis method is a recommended method suitable for traditional Chinese medicine literature research.②There are some similarities between TCM etiology and western medical risk factors of MCI-SKDS.MCI-SKDS is related to natural environmental factors,psychosocial factors and lifestyle factors.③The general demographic factors,TCM syndromes,hobbies and sleep quality of the elderly in the community may be related to MCI;General demographic factors,chronic medical history,hobbies,quality of life,naming and language cognitive domain may be related to MCI-SKDS.Some TCM symptoms have independent predictive value for MCI-SKDS.④Indicators related to MCI can be found in blood routine,blood lipid and urine routine indicators,and indicators related to MCI-SKDS can be found in blood routine,liver function and urine routine indicators;Some indicators have independent predictive value,but the predictive value of a single indicator is general,which is worth further exploration.⑤The prediction index of MCI-SKDS prediction model screened by machine learning is simple and effective,the model has good prediction value and clinical applicability,and the decision tree model performs best. |
PDF Full Text Request