| Objective Stroke is one of the most serious public health problems in the world.It is the second leading cause of death and the leading cause of adult disability worldwide.The prevention and control situation of stroke in China is extremely grim.Every year,for every 100,000 population,about 116 people die of stroke in urban areas and 111 people die of stroke in rural areas.In addition,the recurrence rate of stroke in China is still very high at 11.2%.The prevention and control results of stroke is not optimistic.Therefore,it is necessary to explore the stroke prevention and control program suitable for China’s national conditions.According to several guidelines on the prevention and treatment of stroke,the combination of conventional treatment and control of high-risk factors has a better results.However,clinical prevention and treatment decisions are based on evidence-based conclusion,and the current guidelines for the prevention and treatment of stroke are mainly based on some foreign evidence-based conclusion.The incidence of stroke varies by race and region.Therefore,it is necessary to establish a stroke prediction model suitable for China by combining the newly designed reasonable prospective epidemiological data and rigorous statistical methods.Based on the natural population cohort in rural areas of northeast China established in 2013,this study obtained the latest follow-up data from 2015 to 2018.LASSO method was used to simplify and integrate more than 20 stroke risk factors,and Logistic regression results were quantified,graphically and visualized by using nomogram.A risk prediction model for stroke was established for the first time.In addition,a nested case-control study was conducted to analyze the relationship between plasma non-targeted metabolomics and fatal stroke,so as to find new predictors of stroke,so as to update and optimize the prediction model in the future and improve the prediction ability of stroke risk.Methods Part Ⅰ: In 2013,stratified cluster random sampling method was adopted to randomly select 3 counties in the eastern,southern and northern regions of liaoning province.Then,one town was randomly selected in each county.Finally,8-10 villageswere randomly selected in each town.The current situation of resident villagers aged 35 and above in this area was investigated to obtain the incidence of stroke,and the variables of multivariate Logistic regression analysis were screened by LASSO method to establish the risk factor model of stroke.To analyze the risk factors related to stroke.Part Ⅱ: Follow-up was conducted in 2015-2018 based on the population cohort established in 2013.Then people in two of the counties were randomly selected to merge into the modeling queue,and the remaining people in one county were used as the verification queue.Univariate Cox regression analysis was used to initially screen the candidate factors of stroke incidence with P value less than 0.2.The independent variables included in the Cox regression model were further screened by LASSO method,and the independent variables included in the final Cox regression model were determined according to the results of LASSO method.According to the regression coefficient of the selected independent variables,the corresponding nomogram model was drawn.Bootstrap self-sampling method was adopted to internally verify the nomogram model,and then external validation was performed on the model with the validation queue.Add eGFR to build new model.By comparing the NRI and IDI of the two groups of models,the improvement degree of the model was evaluated.Finally,the applicability of the model is evaluated by decision analysis.Part Ⅲ: In the previously established cohort,eligible hypertension patients were selected,and 29 cases of fatal stroke due to hypertension were randomly selected.The matched hypertension patients with the same follow-up period and no stroke were matched according to 1:1 as the control group.Nontargeted metabolomics tests were performed on baseline plasma from all patients by using ultrahigh performance liquid chromatography-mass spectrometry in both positive and negative ionization modes.Was carried out on the original data after normalization processing,partial least squares discriminant analysis and orthogonal partial least squares discriminant analysis,by multivariate statistical analysis of the VIP value multiples and folding screen differences metabolites,through pathway enrichment analysis,screening is associated with hypertension fatal stroke onset of metabolites and pathways.Results Part Ⅰ: A total of 11,956 people were analyzed at baseline,including 1151 stroke patients,with a prevalence rate of 9.63%.Univariate analysis found 28 variabledifferences.LASSO method was applied to establish a multivariate Logistic regression model with 11 risk factors.Multivariate regression analysis showed that previous history of hypertension,diabetes and coronary heart disease increased the risk of stroke(hypertension OR: 1.58 95%CI: 1.36-1.84;Diabetes OR: 1.43 95%CI: 1.19-1.72;Coronary heart disease: OR: 1.30 95%CI: 1.03-1.63).People with a family history of hypertension and stroke have an increased risk of stroke(hypertension OR: 1.63 95%CI:1.39-1.89;Stroke OR: 1.63 95%CI: 1.39-1.92).Stroke risk was reduced in people with moderate to severe labor intensity(moderate OR 0.85,95%CI: 0.74-0.98).Severe OR0.55,95%CI: 0.47-0.65).High uric acid increases the risk of stroke(OR: 1.55 95%CI:1.22-1.95).Part Ⅱ: The baseline characteristics of the population in the modeling cohort(7608)and the validation cohort(3799)were compared,and it was found that there were certain differences between the two groups in age,smoking and drinking history,education level,previous history and family history.The background of the two groups of people is different,so it can be considered that the verification queue is an ideal external verification population.Using the LASSO method,lambda value was taken to be equal to lambda.0.5se,and a total of 7 independent variables were included.They are age,gender,the history of hypertension,stroke or TIA,marital status and physical activity intensity.The nomogram model showed that the score increased with age.Men scored 17.9.The scores of divorced and widowed were 13.51 and 13.93 respectively.The score of hypertension history was 25.64.The score of history of stroke and TIA was 26.90.The scores of moderate and mild physical activity were 9.25 and 11.01 respectively.High AST scores 15.36;With the increase of the overall score of the nomogram model,the risk of stroke increased in 1 year,3 years and 4 years.ROC curve was drawn for the modeling queue and validation queue to show the modeling queue: 1-year risk cutoff value =0.007736473,AUC=0.823;3-year risk cutoff value =0.02182493,AUC=0.787;4-year risk cutoff value =0.02827864,AUC=0.792.Validation queue: 1-year risk cutoff value =0.01054306,AUC=0.780;3-year risk cutoff value =0.02366501,AUC=0.756;4-year risk cutoff value =0.03019615,AUC=0.753.The nomogram model established by the newly added eGFR showed that scores increase with age;Men scored 19.69;The scores of unmarried,divorced and widowed were4.11,14.44 and 14.60 respectively.Hypertension history score was 28.11.History of stroke and TIA was 28.11.The scores of moderate and mild physical activity were 10.24 and 11.28 respectively.High AST scores 17.06;With the increase of the overall score of the nomogram model,the risk of stroke increased in 1 year,3 years and 4 years.The 1-year cut-off value of the modeling queue of GFR model =0.007804812,AUC=0.822;3-year cut-off value =0.02245559,AUC=0.789;4-year cut-off value =0.02910098,AUC=0.793.The one-year cut-off value of the verification queue of GFR model=0.009084268,AUC=0.773;3-year cut-off value =0.02612931,AUC=0.758;4-year cut-off value =0.03244035,AUC=0.755.NRI and IDI have little difference in the prediction ability of the old and new models.Part Ⅲ: The age,sex and baseline blood pressure grade composition of the case group and the control group were consistent.There were no differences in height,weight,waist circumference,total cholesterol,triglycerides,low-density and high-density lipoprotein,fasting blood glucose,smoking and drinking(all P < 0.05).The results of pls-da and opls-da analysis in the normalized pls-da mode showed that the grouping separation trend was obvious.Combined with the VIP values obtained by opls-da analysis,according to the screening threshold p-value < 0.05,|log2FC| > 0.263 and VIP > 1,55 differential metabolites,45 cationic modes and 10 anionic modes were finally obtained,among which 23 were up-regulated and 32 down-regulated.Enrichment pathway analysis screened 6 pathways with significance less than 0.05.It involves autophagy regulation,pathogenic escherichia coli infection,glycosylphosphatidylinositol anchor biosynthesis,d-glutamine and d-glutamine metabolism,caffeine metabolism and other pathways.Conclusions This study found that the prevalence of stroke in rural population in northeast China was 9.63%.For the first time,the LASSO method was used to screen variables for multivariate COX regression analysis,and the nomogram was used to construct the risk factor model.A quantitative,graphical and visualized stroke risk prediction model suitable for primary and secondary stroke prevention in China has been established.At the same time,it was found that many metabolites such as glutamate and ferulic acid may be involved in the pathogenesis of hypertensive stroke by using the metabolomics method,which provides a new direction for the prevention and pathogenesis exploration of hypertensive stroke. |
PDF Full Text Request