Construction And Verification Of High-risk Of Stroke Risk Prediction Model In Physical Examination Population

Objectives:1.Through questionnaire survey,the lifestyle status of the physical examination population was discussed.2.To screen the high-risk groups of stroke in the physical examination population,and analyze the influencing factors of the high-risk groups of stroke.3.To construct the high-risk of stroke risk prediction scoring system and decision tree model,which can provide screening tools for identifying high-risk objects of high-risk groups of stroke.Methods:From June 2021 to August 2022,a convenient sampling method was adopted in a physical examination center in Baoding City to conduct questionnaire survey,related physical examination and biochemical detection among people aged 40 and above who participated in physical examination.The "2015 Community Population Screening Table for Risk Factors of Cardiovascular and cerebrovascular Diseases"was used for stroke high-risk screening,and the general information questionnaire,food frequency questionnaire(FFQ),International Short Physical Activity Questionnaire(IPAQ),and Pittsburgh Sleep Quality Questionnaire(PSQI)were used to investigate the physical examination population.Measures include height,weight,blood pressure,blood sugar,blood fat and so on.Epidata 3.1 software was used for data entry,Excel 2016 software for data sorting,and SPSS 26.0 software for statistical analysis.Frequency,percentage,mean ± standard deviation were used for statistical description of the data,univariate analysis(χ2 test,nonparametric test)was used to screen the predictors,and binary logistic regression analysis was used to explore the influencing factors of stroke risk groups.The high-risk of stroke risk prediction scoring system and CART decision tree model were constructed and verified internally.Results:1.Lifestyle of the medical examination group94 smokers(12.82%);28 people(3.82%)drank alcohol every day;87 people(11.87%)had salty food and 97 people(13.23%)had greasy food.Five dietary patterns were extracted by exploratory factor analysis:vegetarian pattern of 159 people(21.69%),meat-egg pattern of 156 people(21.28%),processed meat-offal pattern of 150 people(20.46%),coffee-beverage pattern of 140 people(19.10%),milk tea pattern of 128 people(17.46%).The retired and elderly people tend to choose the diet pattern mainly based on animal food.The men are more inclined to choose the milk-tea diet pattern than the women.The people with higher education are more inclined to choose the processed meat-animal offal-aquatic food diet pattern than the people with lower education level.The level of physical activity was mainly medium and low,and the high level of physical activity only accounted for 15.96%,and the higher the age,the lower the detection rate of high level of physical activity.The detection rate of sleep disorder was 13.78%.2.High-risk groups of stroke screening conditionsThe high-risk groups detection rate of stroke was 37.79%(277 persons)among 733 subjects.The detection rate of moderate-risk groups was 8.32%(61 persons);53.89%(395 persons)were detected at low-risk groups.3.Influencing factors of high-risk groups of strokeThe results of logistic regression analysis showed that gender、marital status、work status、smoking years、dietary pattern、physical activity level、sleep disorder were independent influencing factors in high-risk groups of stroke.The risk of high-risk groups of stroke was higher in men than in women(OR=2.34);Unmarried/divorced/widowed people were higher than married(OR=2.74);Retirees were higher than unemployed(OR=3.42);Those who had smoked for 10～20 years(OR=4.15)and those who had smoked for≥20 years(OR=6.48)were higher than those who had smoked for<10 years;The coffee-beverage dietary pattern was higher than that of the vegetarian model(OR=2.31);Those with moderate and low levels of physical activity were higher than those with high levels of physical activity(OR=2.54、3.88);People with sleep disorder were higher than those without sleep disorder(OR=2.09).The decision tree model included 6 variables:smoking years、age、alcohol consumption、dietary pattern、physical activity level、sleep disorder.among which smoking years were the most important influencing factors,and there were obvious interactions between the influencing factors.4.Construction and verification of prediction models4.1 Scoring SystemThe score system based on the regression model has a total score of 0～79.Including gender、marital status、work status、smoking years、dietary pattern、physical activity level、sleep disorder were 7 predictive factors.Using this scoring system,people can bedivided into high-risk group(23～79 points)and low-risk group(0～22 points),that is,when the individuals cores≥23points,it can be judged as a high-risk objects of high-risk groups of stroke.The AUC values of the modeling set and validation set were 0.760 and 0.762,respectively.Hosmer-Lemeshowtest results and calibration curve showed that the model had good calibration degree.4.2 CART Decision Tree ModelThe decision tree model constructed based on CART algorithm has a tree depth of 5 layers and 7 terminal nodes.Including 6 influencing factors:smoking years、age、alcohol consumption、dietary pattern、physical activity level、sleep disorder,and a total of 7 rules were generated,of which smoking ≥10 years,with the highest risk of stroke(83.1%).The AUC value of the model is 0.751,both the modeling set and the verification set gain curves show that the model fits well.4.3 Comparative of the two modelsBoth the constructed scoring system and decision tree model have good differentiation and calibration,and the prediction performance of the scoring system is slightly better than that of the decision tree model.Conclusions:1.The dietary pattern of the physical examination group in Baoding was diverse,the physical activity was at a low level,and the detection rate of sleep disorders was at a high level.2.The detection rate of high-risk groups of stroke was at a high level of physical examination population in Baoding City.3.Gender,marital status,work status,smoking years,dietary pattern,physical activity level,sleep disorder were independent influencing factors in high-risk groups of stroke.4.The scoring system and decision tree model constructed in this study have good predictive performance,and the scoring system is slightly better than the decision tree model.This screening tools can be used to identify high-risk subjects of high-risk groups of stroke,and to provided more accurate information for nursing staff on stroke prevention at zero level.