A Study On Constructionof Risk Prediction Tools For Postmenopausal Osteooorotic Fractures Based On Two Models

1.BackgroundOsteoporotic fracture is one of the most destructive outcomes of osteoporosis.Postmenopausal osteoporotic fracture is a common disease associated with aging,and is a common clinical end-result in postmenopausal osteoporosis(PMOP)patients.Because of the deficiency of estrogen,the etiology of the disease causes bone mass reduction and bone structure change,which increases bone fragility,and then causes fractures,as well as complications such as bone pain and bone deformation,which seriously affect the quality of life and life expectancy of elderly patients.Related studies suggest that the prevalence of postmenopausal osteoporosis in postmenopausal women is four times that of men.Modern clinical and basic research have shown that the occurrence and development of PMOP fractures can be accelerated by a combination of common risk factors such as age,fracture history,dietary habits,and bone mineral density.Although there is no clear record of PMOP fracture in Chinese medicine,the relevant factors of the disease are scattered in the classic ancient books.It can be seen that the hormone levels and immunoregulatory abilities of premenopausal women are declining.There are often symptoms such as weak blood,severe menstruation,and aging symptoms.These clinical symptoms and characteristics of traditional Chinese medicine may have certain early warnings for fractures in patients with PMOP effect.However,an overview of available literature at home and abroad shows that there are few epidemiological data on osteoporotic fractures in China.Internationally recognized and recommended fracture prediction tools are mostly based on other races and regions,as well as the dietary habits and social cultures of the population in other countries and regions.There are obviously limitations in the early warning of osteoporotic fractures that are used to promote the use of osteoporosis in our population.Therefore,based on the community population,we conducted epidemiological investigations of high-risk populations of postmenopausal osteoporotic fractures,screened modern medical risk factors and clinical symptoms with Chinese medicine characteristics,and concise the syndrome elements.At the same time,based on the selected variables,it was constructed in accordance with our demographics.The characteristic PMOP fracture risk warning tool is necessary.2.Objectives2.1 Identifying the modern medical risk factors and TCM syndromes closely related to PMOP fractures.2.2 Establishing a PMOP fracture risk prediction tool that is both accurate and practical.2.3 To explore the methodological mechanism for constructing PMOP fracture risk prediction models.3.Contents3.1 Early Warning Indicators ScreeningGroup Lasso can simultaneously select categorical covariates(including bivariate variables and ordered categorical variables)and continuous covariates,and simultaneously perform parameter estimation and variable selection under the premise of ensuring prediction accuracy,significantly reducing the error rate of prediction in the early stage of disease.Risk assessment tools have unique advantages in variable screening-Group Lasso was used to screen the modern medical risk factors and TCM syndrome elements closely related to "PMOP→fracture",and combined with clinical experience,the "PMOP→fracture" early warning index was finally determined.3.2 Construction of the modelThe decision tree is one of the representatives of the machine learning classification algorithm.It can extract the classification rules from a set of irregular cases and compare the size of the attribute values in each internal node and judge the node according to the top-down recursive method.The following branches get a classification conclusion at the leaf node.Logistic regression analysis is a generalized linear regression analysis model that can be used to screen for risk factors for disease occurrence and predict the probability of disease occurrence based on risk factors.The"PMOP→fracture" early warning indicator selected by the Group Lasso method was a covariate,and the fracture occurrence(whether or not fracture)was the outcome variable.The early warning tools for PMOP fractures were constructed based on the decision tree classification model and logistic regression model.3.3 The formation and mechanism for the construction of the toolsUsing receiver operating characteristic curve(ROC),according to the area under the ROC curve,the prediction accuracy of the two PMOP fracture early warning models constructed by decision tree model and Logistic regression model were compared respectively,and the accuracy was compared after analysis.The practicable PMOP fracture risk assessment model was further sorted out and summarized to clearly assess the construction mechanism of the model.4.Methods4.1 Study DesignRegistered study.4.2 Questionnaire SurveyTo improve and revise the "Risk Factors and Syndromes of Osteoporosis Risk among 40 to 65-Year-Old Women in the Community"developed in the previous period,it has good reliability and validity.The main contents include:basic characteristics of the crowd,lifestyle,risk factors,clinical symptoms,bone mineral density and other five aspects.From March to June 2009,March to August 2010,and June 2011 to November 2011,field surveys were used to obtain information on general information,risk factors,clinical symptoms,and fractures.No,fracture time,number of fractures,cause of fracture,site of fracture,etc.4.3 ParticipantsIn the Dongcheng District of Beijing City and Xuhui District of Shanghai City,the relevant clinical and social life habits of the 40-65-year-old high-risk patients with PMOP fractures were registered by site investigation.At the same time exclude patients with secondary osteoporosis.4.4 Statistical analysis methodsBased on the Microsoft Excel database,SPSS 23.0 was used to describe the general situation of patients.Using the decision tree package and grplasso package ofR3.3.3 software,a decision tree classification model and Logistic regression analysis model based on SMOTE oversampling data were constructed.The performance of model prediction was evaluated by the area under the ROC curve.5.Results5.1 CompletionIn Shanghai and Beijing,three surveys were conducted for three consecutive years,1,823 questionnaires were sent out,1498 data were complete,and the response rate was 82.17%.Loss of data or incomplete data accounted for 17.83%.5.2 General descriptionWith the 1,498 people surveyed,1,129 were menopause people,49 of whom had PMOP fractures,and the prevalence was about 4.34%.According to the bone mineral density test results of the study population and the causes of fractures,sprains occurred in 12 cases and fell in 37 cases.In the fracture site,there were 14 cases of distal radius fractures,7 cases of upper femoral fractures,6 cases of thoracolumbar vertebral fractures,and ankle fractures.22 cases.5.3 Screening of influence factorsWhether or not a fracture occurred was used as an outcome variable.Group Lasso model screening showed that the specific group variables included:bone mineral density.age,food type,height,menstruation,number of births,and liver and kidney yin deficiency(foot and foot irritation,night sweats,and legs).Soft,dazzling,blurred vision,dry eyes,bad heat,hair loss,teeth shaking,bitter mouth,irritability,afternoon hot flashes,insomnia,easy dreams,chest pain,fullness,lower limbs tendons).A total of six explanatory variables were obtained from the decision tree model.The indicators were,in descending order of importance,bone mineral density,dizziness,meat,production frequency,blurred vision,and fatigue.5.4 Construction of the model5.4.1 The model Based on Logistic RegressionBased on the group variables selected by the Group Lasso model,osteoporotic fracture prediction tools were derived based on Logistic regression models:P＝1.88 +0.437*BMD+0.289*age+0.023*rice noodles-0.007*dairy products-0.096*beans Products-0.128*Meat-0.084*Fish-0.007*Fresh Vegetables-0.018*Eggs + 0.047*Seaweed + 0.048*Height-0.035*Whether it Becomes Short-0.081*Menarche Age+ 0.171*Whether Menopause + 0.121*Menopause Number of years + 0.039*Number of Pregnancy + 0.192*Number of Production-0.056*Whether the Uterus and Ovary are removed + 0.05*Feet of Foot and Foot Illness-0.094*Night sweats + 0.008*Soft Legs + 0.15*Dizziness-0.048*Blurred Vision-0.045*Eyes Dry-0.089*Bad Heat+ 0.08*Hair Loss + 0.034*Tooth Shake-0.101*Bitterness + 0.004*Fury + 0.054*Afternoon Hot Flash-0.056*Insomnia + 0.019*Dreamy Starter-0.02*Chest Throat+ 0.137*The lower limbs transferred tendons,and the receiver operating characteristic curve was plotted against the predicted probability of the prediction model.The results showed that the area under the curve was AUC = 0.8775(95%CI = 0.8412-0.9138).5.4.2 The Model Based on Decision TreesAccording to the decision tree model,the main classification criteria for fracture populations are as follows:1 patients with abnormal bone mass and more than one delivery;2 patients with abnormal bone mass,those without a history of farrowing but less meat intake;3 those with normal bone mass,No symptoms on weekdays,but more than 1 childbirth and often fatigue;4 patients with normal bone mass,daytime with dizzy symptoms but no vision blurred;5 patients with reduced bone mass.The receiver operating characteristic curve was plotted against the prediction tree prediction model prediction probability.The results showed that the area under the curve was AUC =0.871(95%CI = 0.8226-0.9211).5.5 Comparison of Two Models5.5.1 Comparison of the ROC CurveBased on the decision tree classification model and logistic regression model,the ROC curve was plotted for the prediction probability of PMOP fracture.After calculation,the AUC value of the decision tree model is 0.871(95%CI=0.8226-0.9211);the AUC value of the Logistic regression model is 0.8775(95%CI=0.8412-0.9138).According to the comparison of AUC values,both models showed good predictive power.5.5.2 Comparison of sample sizeSample size is critical for both models.When the test sample size is small,the estimation of the regression coefficients in the Logistic model will have a large error,far from the standard value;similarly,the accuracy of the decision tree model is also affected by the size of the sample.From the perspective of theory and application,the theoretical basis of the Logistic regression model is relatively complete and has a wide range of applications.The model has no rigid requirements for whether the independent variables meet the normal distribution,and can perform regression modeling on the dependent variables of different data types,including categorical dependent variables,classified independent variables,continuous independent variables,and mixed variables.The application scope of the model is very high.6.ConclusionBased on the data of 1129 high-risk patients with PMOP fractures in 1129 cases of 40-65 years old in Beijing and Shanghai,a preliminary risk prediction tool for PMOP fractures was preliminarily constructed.Combining modern medical risk factors,clinical symptoms with traditional Chinese medicine features,and TCM syndrome factors information as a reference variable for the construction of predictive models can improve the accuracy and practicality of disease prediction models,and provide a reference for clinical practice and promotion.Based on the utility.