Objective To analyze the disease burden of type 2 diabetes mellitus(T2DM)attributable to high body mass index(BMI)in China from 1990 to 2019 in the context of rapid growth in high BMI rates.Methods Data was extracted from GBD 2019,and the disease burden of T2DM attributable to high BMI in China from 1990 to 2019 was analyzed for overall and subgroups defined by age and sex separately and jointly.The joinpoint regression models were used to analyze the trends of standardized death rate and standardized disability-adjusted life year(DALY)rate.Results From 1990 to 2019,the prevalence of T2DM increased from 2928.78 per 100000 to 6328.79 per 100000 in China.The number of T2DM deaths attributed to high BMI increased from 10500 to 47500 and the standardized death rate increased from 1.25 per 100000 to 2.39 per 100000.The attributed DALY increased from 771800 person-years to 3737600 person-years,and the standardized DALY rate increased from 80.21 per 100000 to 181.54 per 100000.Years of life lost(YLL)and years lived with disability(YLD)and their standardized rates also increased.From 1990 to 2019,the annual average percentage change of the standardized death rate and the standardized DALY rate of T2DM attributable to high BMI were 2.28%and 2.81%,respectively,which were statistically significant(P<0.05)and males were both higher than females.The standardized DALY rate and the standardized death rate of males exceeded that of females in 2010 and 2014,respectively.Age-stratified results showed that the burden of T2DM,which is attributed to a high BMI,is even greater in people over 50 years old.The YLD rate attributable to high BMI increased the most among the 15～49 age group,reaching 323.99%.Conclusion From 1990 to 2019,the disease burden of T2DM that can be attributed to high BMI increased significantly in China.It is necessary to strengthen prevention and control efforts,effectively manage population BMI,and adopt key interventions for high-risk groups to reduce the disease burden of T2DM.

Objective The BOIN12 design was improved based on conditional weighted likelihood method to solve the optimal dose exploration in the case of delayed toxicity and efficacy of drugs.Methods The conditional weighted likelihood method was used to estimate the toxicity rate and conditional efficacy rate.The improved BOIN12 design was compared with other phase Ⅰ-Ⅱ design through simulation to evaluate its performance.Results Simulation results show that the improved design still has excellent statistical performance in different scenarios and the trial duration is greatly shortened.Conclusion The improved design can addresses both toxicity and efficacy delays in phase Ⅰ-Ⅱ trials.

Objective To develop a subgroup analysis method that can simultaneously deal with longitudinal data containing measurement errors and dropouts.Methods Subgroup analysis was carried out based on a threshold regression model.A new generalized unbiased estimation equation is constructed by using the independence between repeated measurements to deal with measurement errors and introducing an inverse probability weighting matrix to deal with missing response.Results The computer stochastic simulation shows that the proposed estimation method is effective in dealing with measurement errors and dropouts,and has smaller bias and mean square error than the generalized estimation equation method without correcting measurement errors or dropouts.Conclusion In subgroup analysis,when there are measurement errors in covariables and missing values in response variables,it is usually necessary to deal with the measurement errors and missing values in order to obtain reliable parameter estimation.

Objective To explore the dose-response relationship between BMI and the prevalence of dyslipidemia using restricted cubic spline model.Methods Using data of Chronic Disease and Risk Factor Surveillance in Nanjing City from 2017 to 2018.A representative sample of 61 098 residents aged≥18 years was surveyed by face-to-face questionnaire survey,anthropometric measurements and laboratory examinations.A logistic regression model of complex sampling and restricted cubic spline model were used to analyze the dose-response relationship between BMI and the risk of dyslipidemia.Results Finally,60 283 subjects were included.Among them,there were 17 093 dyslipidemia patients with a standardized prevalence rate of 29.8%.After adjusting for confounding factors by multiple logistic regression,participants with overweight(OR=1.43,95%CI:1.36～1.49)and obesity(OR=1.97,95%CI:1.83～2.12)had a significantly higher risk of dyslipidemia compared to those with normal weight.The multiple restricted cubic spline model indicated a non-linear dose-response relationship between BMI and the risk of dyslipidemia in women,young,middle-aged and elderly population,and a linear dose-response relationship between BMI and the risk of dyslipidemia in men.Conclusion Weight control plays a very important role in the prevention of dyslipidemia.

Objective To construct a clinical prediction model for the risk of Alzheimer's disease based on ApoE,combined with risk factors and common clinical indicators.Methods There were 61 cases of Alzheimer's disease patients and 111 cases of fuzzy matching healthy physical examination from Nanjing Hospital of Chinese Medicine data platform from January 2019 to January 2021.Using LASSO regression screening of risk factors,constructing logistic regression forecasting model,10 fold cross verifies the degree of differentiation,validation the calibration of the bootstrap method.The clinical guidance of the prediction model was evaluated by the clinical decision curve,and finally,the clinical prediction model was visualized by nomogram.Results 12 variables were screened out and four risk factors were included,which are age,free triiodothyroxine(FT3),gender and ApoE.The AUC of ROC of the whole sample was 0.879,and the average AUC of ROC after 10 folded and 9 crossed training sets verification was 0.864.Bootstrap method and Hosmer-Lemeshow were used to test the calibration degree.Results χ2 =6.496,P=0.592>0.05.The threshold probability of clinical decision curve ranged from 1%to 88.6%.Conclusion Individualized evaluation of patients using clinical prediction models constructed by age,FT3,gender and ApoE can provide early warning of Alzheimer's disease,carry out early prevention intervention and slow down the development of the disease.

Objective To construct logistic regression,random forest and SVM models to predict the influencing factors of overweight and obesity in medical students,and the prediction performance of the three models was compared,so as to obtain the optimal model for the risk assessment of overweight and obesity.Methods Participants included 1 866 medical students from a city in Hebei Province from May to December 2020.The relevant data of overweight and obesity screening were collected through self-test questionnaire;three models of logistic regression,random forest and SVM are constructed by python.Results The test set showed that the accuracy of logistic regression,random forest and SVM models were 96.26%,98.66%and 98.13%respectively;the specificity were 99.77%,100%and 99.00%,respectively;and the AUC were 0.88,0.99 and 0.88 respectively.Random forest is the optimal prediction model;according to the random forest model results,subjective well-being,negative events and students'economic status are more than 10%of weight in the model.Conclusion Subjective well-being,negative events and students'economic status are the main factors affecting the incidence of overweight and obesity in medical students;the prediction performance of random forest model was better than logistic regression model and SVM model.

Objective To compare the principles,methods and applications of provider-to-population ratios method,gravity-based model,two-step floating catchment area(2SFCA)method and improved two-step floating catchment area method for evaluating the spatial accessibility to healthcare services.Methods Taking the spatial accessibility to maternal and child healthcare services in Nanning prefecture for example,we collected data on vector map,transportation network,population size,and the number of health professionals in maternal and child health institutions.Provider-to-population ratios method,gravity-based model,2SFCA method and improved 2SFCA method were applied to evaluate the spatial accessibility to maternal and child healthcare services,on the scales of county level,township level and village level,respectively.Results The four models showed that the spatial heterogeneity of spatial accessibility to maternal and child health services was significant,and the spatial accessibility was gradually decreased from urban areas of Nanning to rural areas.However,regions with high spatial accessibility and regions with low spatial accessibility,decreasing trend in spatial accessibility across regions,the median and inter quartile range of spatial accessibility were different across the four models.Conclusion The practical significance in evaluating spatial accessibility to healthcare services for provider-to-population ratios method,gravity-based model,2SFCA method and improved 2SFCA method was different,yet the spatial accessibility varied somewhat across the four models.Thus,integrating findings of the four models based on multi space scales is strongly recommended to evaluate the spatial accessibility to healthcare services comprehensively and robustly.

Objective To understand the trajectory and classification of adult body mass index(BMI)in Jiangsu Province.Methods Based on China Health and Nutrition Survey,this study used the linear mixed model tree to explore the trajectory and classification of BMI of people aged 18-65 in Jiangsu Province.Results The linear mixed model tree had 13 nodes and the depth was 6.The classification nodes were baseline BMI,average calorie intake and baseline age.Conclusion The linear mixed model tree can identify the trajectory of BMI and expand the research method of longitudinal data.

Objective The continuous and discrete Bayesian network models of serum uric acid and related metabolic indexes were established to explore the influencing factors of serum uric acid and compare the characteristics and advantages of the results of the two networks.Methods A total of 4646 patients with serum uric acid and metabolic diseases from chronic disease monitoring in Shanxi Province in 2015 were selected.IPCB algorithm was used to establish a continuous Bayesian network of serum uric acid.Meanwhile,the above indicators were discretized,and MMHC was used to establish a discrete Bayesian network of high uric acid.Results The discrete Bayesian network found 14 edges,in which triglyceride and diastolic blood pressure abnormalities were directly related to the occurrence of high uric acid,leading to the occurrence of high uric acid.Age was an indirect factor;Age,TG,LDL,HDL,SP and DP are directly related to uric acid level.With the increase of age,TG and LDL and the decrease of HDL,uric acid level increases,while the increase of uric acid level leads to the increase of SP and DP.TC is indirectly related to uric acid.Conclusion The two network models adapt to different data types,but the continuous Bayesian network has more direct correlation factors and better overall explanatory degree.

Objective To compare the existing evaluation methods of sample's representativeness and provide reference for selection of sample representativeness evaluation methods in clinical research.Methods Simulate the target population of lung cancer patients and select samples with different sample sizes and different degrees of deviation based on the distribution of traits of lung cancer patients in China and the actual situation of sample screening in domestic clinical studies.Calculate sample representativeness using the existing evaluation methods of sample's representativeness,and calculate estimation deviation(bias).By constructing the correlation model between the measured value of each method and bias,analyze the accuracy and stability of each method.Results The overall structural variance rate RV1、RV2、C-statistic based on propensity score、SGCR and K-S distance could well measure the degrees of deviation of different samples.Under different sample sizes,the R2 of RV2 and RV1 are greater than 0.90,and R2 of C-Statistic、SGCR and K-S distance were greater than 0.80.Conclusion The overall structural variance rate is more accurate and stable because the traits weight is taken into account.In particular,RV2 can better measure the representativeness of samples with different degrees of deviation and accurately reflect the estimation deviation.However,when it is difficult to obtain the feature importance information,the reliability of the representative measurement of SGCR as well as C-statistic and K-S distance used the propensity score-based method are acceptable.