Cerebral stroke onset risk assessment method and system

Abstract

The invention discloses a cerebral apoplexy onset risk assessment method and system, and the method comprises the following steps: S1, sample collection: collecting sample data, and updating a sample in a database in real time; s2, creating a risk prediction model according to the current database; s3, screening out a plurality of variables of which the association strength with the cerebral apoplexy onset risk reaches a predetermined requirement from the factors incorporated into the model; performing single factor analysis by using a Cox proportional risk regression model, preliminarily determining an adjustment factor, performing prediction factor screening by using Lasso regression, and achieving the purpose of variable screening by adjusting a penalty coefficient lambda; s4, inputting information of a to-be-evaluated user into the risk prediction model, and calculating a comprehensive score according to the input information of the to-be-evaluated user; and S5, predicting the risk level of the to-be-evaluated user according to the comprehensive score, and displaying the risk level through an interactive interface. Through the cerebral apoplexy onset risk assessment model, high-risk groups of cerebral apoplexy can be effectively identified, intervention measures can be taken as soon as possible, and occurrence and development of cerebral apoplexy can be controlled.

Description

Technical Field

[0001] The present invention belongs to the technical field of big data analysis and prediction, and specifically relates to a stroke risk assessment method and system for assessing and grading the stroke risk of a population. Background Art

[0002] Stroke is a disease caused by multiple causes. The stroke risk assessment model is a good tool for comprehensive assessment of multiple influencing factors. It establishes a quantitative relationship between risk factors and health status through statistical models, and predicts the probability of an individual suffering from a stroke within a certain period of time. It plays an important role in effectively identifying people at high risk of stroke and conducting early intervention.

[0003] The stroke risk assessment models published abroad are based on samples from Western populations, whose ethnicity and lifestyle are different from those of the Chinese population, and are all complex models. The cohorts based on the existing models in China are all established a long time ago and lack large cohort studies. The accuracy and applicability of stroke risk assessment models in my country need to be improved. Summary of the invention

[0004] In response to the problems existing in existing technologies, a nationally representative cohort study data was used to establish a high-sensitivity and high-specific stroke risk assessment model suitable for adults in my country, providing a basis for effectively identifying high-risk susceptible groups for stroke and scientifically formulating targeted intervention measures and strategies.

[0005] To achieve the above object, the present invention provides a method for assessing the risk of stroke, the method comprising the following steps:

[0006] S1. Sample collection, collect sample data and update samples in the database in real time;

[0007] S2. Create a risk prediction model based on the current database; the fitted prediction model is in the form of:

[0008]

[0009] Among them, λ(t|Xi) is the risk function of being exposed to a factor at time t; λ0(t) is the risk function of not being exposed to the factor;

[0010] S3. Select several variables whose correlation strength with stroke risk meets the predetermined requirements from the factors included in the model; use the Cox proportional hazard regression model to conduct univariate analysis, preliminarily determine the adjustment factors, and then use Lasso regression to screen predictive factors, and achieve the purpose of variable screening by adjusting the penalty coefficient λ;

[0011] S4. Input the information of the user to be evaluated into the risk prediction model, and calculate a comprehensive score based on the input information of the user to be evaluated;

[0012] S5. predict the risk level of the user to be assessed based on the comprehensive score and display it through the interactive interface;

[0013] The fitted model is:

[0014] The predicted probability value of stroke risk: P = 1-0.9999458^exp(0.072664627*age-0.078075715*education level (junior high school)-0.197020818*education level (high school and above)+0.007298872*systolic blood pressure+0.013844382*diastolic blood pressure+0.154268582*diabetes+0.006942786*waist circumference+0.425624388*self-rated poor health).

[0015] Further, in step S1,

[0016] The sample size was calculated using a special formula for calculating the sample size of a cohort study. The formula is as follows:

[0017]

[0018] Where: p 1 and p 0 are the expected incidence rates of the exposed group and the control group, is the average of the two incidence rates, q = 1-p, Z α and Z β is the boundary value of the standard normal distribution.

[0019] Assuming α = 0.05 (two-sided), β = 0.10, Z α =1.96,Z β =1.282, the incidence of stroke in the non-exposed group p 0 =0.56%, RR=1.6. Substituting into formula (2), the sample size is calculated to be 26,910 people.

[0020] The EPV strategy is adopted, and it is set that at least 10 events are required for each predictor variable to be included in the prediction model. The calculation formula is as follows:

[0021]

[0022] In the formula is the event rate studied by the prediction model, n var is the potential number of variables when fitting the model.

[0023] Furthermore, step S1 includes database cleaning and variable cleaning. Database cleaning: merge the baseline database and the follow-up database to form a general database; in the process of database merging, the baseline data and the follow-up data are first accurately matched, and the ID number, name, and gender are all consistent. For missing baseline or follow-up ID numbers, fuzzy matching can be performed using the date of birth, and the matching principle is that the date of birth, name, and gender are all consistent. After matching the date of birth and age and gender, if it still cannot be matched as the same person, the object will be eliminated; variable cleaning: 1) including variable validity cleaning: including checking whether there are logical errors, if there are, correct them; 2) checking whether there are outliers, correcting or eliminating observations with outliers; 3) and variable integrity cleaning: processing of missing values ​​of variables, for variables with less than 5% missing, use the mean for interpolation. .

[0024] Further, in step S2, the predicted probability value of stroke risk is:

[0025] P=1-S 0 (6)^exp(β1Xi1+…+βpXip) (4)

[0026] Among them, S 0 (6) is the 6-year basic survival probability; β1Xi1+…+βpXip is the linear prediction value;

[0027] The factors initially included in the model included: age, sex: 0=male, 1=female, education level: 1=primary school and below, 2=junior high school, 3=senior high school and above, marital status: 1=single, 2=married or cohabiting, 3=other, smoking: 0=no, 1=yes, drinking: 0=no, 1=yes, region: 1=eastern, 2=central, 3=western, systolic blood pressure, diastolic blood pressure, diabetes: 0=no, 1=yes, high-density lipoprotein cholesterol, self-rated poor health: 0=no, 1=yes, body mass index, waist circumference, 14 variables; among them, age, systolic blood pressure, diastolic blood pressure, high-density lipoprotein cholesterol, body mass index and waist circumference were continuous variables.

[0028] Further, the step S3 includes:

[0029] S3.1 Screening of predictors;

[0030] The Cox proportional hazard regression model was used for univariate analysis to preliminarily determine the adjustment factors. Lasso regression was used for predictive factor screening, and the penalty coefficient λ was adjusted to achieve the purpose of variable screening. The penalty coefficient λ that minimized the model estimation error was considered to be the optimal one.

[0031] S3.2. Predictors selected using Lasso regression.

[0032] Furthermore, the risk level is determined based on the calculated predicted probability value of stroke risk: low risk: <1.5%, general risk: 1.5%-7.9%, moderate risk: 8%-17.9%, high risk: ≥18%.

[0033] Furthermore, for continuous data, if the data satisfy normal or approximately normal distribution, the mean ± standard deviation is used. The t test or analysis of variance was used for comparison between groups; for skewed continuous data, the median [M(p 25 , p 75 )], and the rank sum test was used for inter-group comparison; for categorical data, frequency and constituent ratio [n(%)] were used to express them, and the chi-square test or rank sum test was used for inter-group comparison.

[0034] On the other hand, the present invention provides a stroke risk assessment system, which includes a sample collection module, a model construction module, a variable determination module, a prediction module and a display output module.

[0035] Furthermore, in the system, the fitted prediction model is in the form of:

[0036]

[0037] Among them, λ(t|Xi) is the risk function of being exposed to a factor at time t; λ0(t) is the risk function of not being exposed to the factor.

[0038] Further, wherein the sample collection module is used to collect sample data and update the basic database in real time;

[0039] The model building module is used to create a risk prediction model based on the current basic database; the factors initially included in the model include: age, gender: 0 = male, 1 = female, education level: 1 = primary school and below, 2 = junior high school, 3 = high school and above, marital status: 1 = single, 2 = married or cohabiting, 3 = other, smoking: 0 = no, 1 = yes, drinking: 0 = no, 1 = yes, region: 1 = east, 2 = central, 3 = west, systolic blood pressure, diastolic blood pressure, diabetes: 0 = no, 1 = yes, high-density lipoprotein cholesterol, self-rated poor health: 0 = no, 1 = yes, body mass index, waist circumference 14 variables; among them, age, systolic blood pressure, diastolic blood pressure, high-density lipoprotein cholesterol, body mass index and waist circumference are continuous variables;

[0040] The variable screening module is used to screen out several variables whose correlation strength with the risk of stroke meets the predetermined requirements from the factors included in the model;

[0041] The prediction module is used to input the information of the user to be evaluated into the risk prediction model, and calculate the comprehensive score based on the input user information to be evaluated; the information at least includes the final variables used in the risk prediction model;

[0042] The display output module is used to predict the risk level of the user to be assessed based on the comprehensive score and display it through the interactive interface.

[0043] Beneficial effects:

[0044] Stroke has become the second leading cause of death in the world. In my country, stroke has become the disease with the heaviest disease burden. The situation of stroke prevention and control in my country is severe, and effective intervention strategies and measures are urgently needed to prevent the increase in the number of stroke patients. Through the stroke risk assessment model, high-risk groups for stroke can be effectively identified, and intervention measures can be taken early to control the occurrence and development of stroke, improve the health level of residents, and reduce the disease burden. In 2013, the American College of Cardiology / Heart Association (ACC / AHA) released the Pooled Cohort Equations (PCE) model for 10-year atherosclerotic cardiovascular disease (ASCVD) risk prediction, but at the same time pointed out that the PCE model is based on cohort data of white and black people in the United States and may not be applicable to other populations. In addition, due to the differences in lifestyle between Chinese and foreign populations, some scholars have shown that the external validation results of foreign stroke prediction models show that there is an overestimation or underestimation phenomenon. This prediction model uses the latest follow-up data at that time to develop a stroke risk model suitable for the Chinese population, which is more accurate. This prediction model also optimizes the screening process of prediction factors. Based on similar models using literature methods and empirical methods, it adopts Lasso regression with better performance, which can compress the regression coefficient of the prediction factor to 0, thereby screening out valuable prediction factors and developing a more streamlined model. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 A flow chart of a method and system for assessing the risk of stroke according to the present invention;

[0046] Figure 2 It is a schematic diagram of database information of the present invention;

[0047] Figure 3 It is a schematic diagram of the relationship between the penalty coefficient λ and the variables of the present invention;

[0048] Figure 4 This is a schematic diagram of the optimal situation of the penalty coefficient λ of the present invention;

[0049] Figure 5 This is a nomogram used in the stroke risk assessment model of the present invention. DETAILED DESCRIPTION

[0050] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0051] In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and "third" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance.

[0052] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0053] The following combination Figure 1-Figure 5 The specific embodiments of the present invention are described in detail. It should be understood that the specific embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0054] like Figures 1 to 5 As shown, according to a method and system for assessing the risk of stroke of the present invention, the system includes a sample collection module, a model building module, a variable determination module, a prediction module and a display output module.

[0055] Among them, the sample collection module is used to collect sample data and update the basic database in real time;

[0056] The model building module is used to create a risk prediction model based on the current basic database; the factors included in the model include: age (continuous variable), gender: 0=male (reference), 1=female, education level: 1=primary school and below (reference), 2=junior high school, 3=senior high school and above, marital status: 1=single, 2=married / cohabiting (reference), 3=other, smoking: 0=no (reference), 1=yes, drinking: 0=no (reference), 1=yes, region: 1=eastern, 2=central (reference), 3=western, systolic blood pressure (continuous variable), diastolic blood pressure (continuous variable), diabetes: 0=no (reference), 1=yes, high-density lipoprotein cholesterol (continuous variable), self-rated poor health: 0=no (reference), 1=yes, body mass index (continuous variable), waist circumference (continuous variable) 14 variables.

[0057] The variable screening module is used to screen out several variables whose correlation strength with the risk of stroke meets the predetermined requirements from the factors included in the model.

[0058] The prediction module is used to input the information of the user to be evaluated into the risk prediction model, and calculate the comprehensive score based on the input information of the user to be evaluated. The information includes at least the final variables used in the risk prediction model, for example, in this embodiment, the information includes at least age, education level, systolic blood pressure, diastolic blood pressure, diabetes, waist circumference and self-assessed health.

[0059] The display output module is used to predict the risk level of the user to be assessed based on the comprehensive score and display it through the interactive interface.

[0060] Correspondingly, a method for predicting stroke risk according to the present invention comprises the following steps:

[0061] S1. Sample collection, collect sample data and update samples in the database in real time;

[0062] S2. Create a risk prediction model based on the current database; the fitted prediction model is in the form of:

[0063]

[0064] Among them, λ(t|Xi) is the risk function of being exposed to a factor at time t; λ0(t) is the risk function of not being exposed to the factor;

[0065] S3. Select several variables whose correlation strength with stroke risk meets the predetermined requirements from the factors included in the model; use the Cox proportional hazard regression model to conduct univariate analysis, preliminarily determine the adjustment factors, and then use Lasso regression to screen predictive factors, and achieve the purpose of variable screening by adjusting the penalty coefficient λ;

[0066] S4. Input the information of the user to be evaluated into the risk prediction model, and calculate a comprehensive score based on the input information of the user to be evaluated;

[0067] S5. Predict the risk level of the user to be assessed based on the comprehensive score and display it through the interactive interface.

[0068] Specifically, in step S1,

[0069] The sample size was calculated using a special formula for calculating the sample size of a cohort study. The formula is as follows:

[0070]

[0071] Where: p 1 and p 0 are the expected incidence rates of the exposed group and the control group, is the average of the two incidence rates, q = 1-p, Z α and Z β is the boundary value of the standard normal distribution.

[0072] Assuming α = 0.05 (two-sided), β = 0.10, Z α =1.96,Z β =1.282, the incidence of stroke in the non-exposed group p 0 =0.56%, RR=1.6. Substituting into formula (2), the sample size is calculated to be 26,910 people.

[0073] At the same time, based on the empirical estimation of statistical simulation, the sample size is evaluated to see whether it is sufficient. The strategy for considering the number of variables in the prediction model is the number of events required for each variable included in the model, that is, the EPV (Events Per Variable) strategy. Using 10EPV, it is required that at least 10 events are required for each predictor variable included in the prediction model. The simulation experiment results show that when the outcome is a time-event variable, the estimated model coefficient is less biased when the number of events in the sample size is 10 times or more than the independent variable. The calculation formula is as follows:

[0074]

[0075] In the formula is the event rate studied by the prediction model, n var is the potential number of variables when fitting the model.

[0076] The database of the present invention includes: a baseline database and a follow-up database (including a family database and a cardiovascular and cerebrovascular database).

[0077] The baseline database and the follow-up database are merged to form a master database. During the database merging process, the baseline database and the follow-up database are first accurately matched, and the ID number, name, and gender are all consistent. If the baseline or follow-up ID number is missing, the date of birth can be used for fuzzy matching. The matching principle is that the date of birth, name, and gender are all consistent. After matching the date of birth and age and gender, if it still cannot be matched to the same person, the subject will be eliminated.

[0078] After the database is merged, the variables in the data are cleaned. In principle, each variable in the data is cleaned. It mainly includes: (1) Variable validity cleaning: including checking whether there are logical errors, if there are, correct them; checking whether there are outliers (outliers are determined through literature review and expert consultation), and correcting or eliminating observations with outliers. (2) Variable integrity cleaning: mainly dealing with missing values ​​of variables. For some variables, such as missing age and gender, the ID card number information can be used to fill in. If the age is missing, the ID card number is used to understand the birth date of the research subject, and the age of the research subject is obtained by subtracting the birth date from the survey date. If the gender is missing, the second to last digit of the ID card number is used, with odd numbers representing males and even numbers representing females. At the same time, for some variables with less missing values, such as those with missing values ​​less than 5%, the mean is used for interpolation.

[0079] S2. Create a risk prediction model based on the current database; the fitted prediction model is in the form of:

[0080]

[0081] Among them, λ(t|Xi) is the risk function of being exposed to a factor at time t; λ0(t) is the risk function of not being exposed to the factor.

[0082] Prediction probability of 6-year stroke risk:

[0083] P=1-S 0 (6)^exp(β1Xi1+…+βpXip) (4)

[0084] Among them, S 0 (6) is the 6-year basic survival probability; β1Xi1+…+βpXip is the linear prediction value.

[0085] The factors included in the model included: age (continuous variable), sex: 0=male (reference), 1=female, education level: 1=primary school and below (reference), 2=junior high school, 3=senior high school and above, marital status: 1=single, 2=married / cohabiting (reference), 3=other, smoking: 0=no (reference), 1=yes, drinking: 0=no (reference), 1=yes, region: 1=eastern, 2=central (reference), 3=western, systolic blood pressure (continuous variable), diastolic blood pressure (continuous variable), diabetes: 0=no (reference), 1=yes, high-density lipoprotein cholesterol (continuous variable), self-rated poor health: 0=no (reference), 1=yes, body mass index (continuous variable), waist circumference (continuous variable), 14 variables.

[0086] S3. Select several variables whose correlation with the risk of stroke meets the predetermined requirements from the factors included in the model. Including:

[0087] S3.1 Screening of predictors

[0088] Univariate analysis was performed using the Cox proportional hazard regression model to preliminarily determine the adjustment factors. Since ischemic stroke and hemorrhagic stroke have some common risk factors, the population with hemorrhagic stroke as the follow-up outcome was excluded in the analysis of influencing factors of ischemic stroke, and vice versa. The proportional hazard hypothesis was verified using the Schoenfeld residuals method. For variables that did not meet the proportional hazard hypothesis, the interaction term between the variable and time was added to the model for adjustment. In the sensitivity analysis, deaths and stroke patients diagnosed within the first year were excluded, and the model adjustment factors were consistent with the multivariate analysis. The subjects were divided into different subgroups according to baseline characteristics for analysis. The likelihood ratio test was used for the interaction test of the subgroup analysis to compare whether the differences between the models with and without the interaction term were statistically significant.

[0089] Interaction analysis between factors, multiplicative interaction analysis using Cox regression, the interaction terms between variables were included in the model, and the existence of multiplicative interaction was determined by the size of the household value, and P < 0.05 was considered to be a multiplicative interaction. Additive interaction was calculated using the SAS program and Excel sheet provided by Andersson, and the estimated value and covariance matrix were calculated using SAS software, and then the additive interaction index was calculated using Excel. Three indicators were used to evaluate additive interactions: (1) relative excess risk due to interaction (RERI): reflects the relative risk caused by additive interaction. When the 95% CI of RERI does not include 0, it is considered statistically significant. (2) attributable proportion due to interaction (AP): reflects the proportion of the total effect of two factors attributable to the interaction when the two factors exist at the same time. When the 95% CI does not include 0, it is considered statistically significant. (3) the synergy index (S): reflects the ratio of the effect of the two factors existing at the same time to the sum of the independent effects of the two factors. When the 95% CI does not include 1, it is considered statistically significant.

[0090] The above statistical analysis was performed using SAS9.4 (SAS Institute Inc.) software. Two-sided test, P < 0.05 was considered statistically significant.

[0091] Thus, 14 variables were initially screened and included in the model, including age (continuous variable), sex: 0=male (reference), 1=female, education level: 1=primary school and below (reference), 2=junior high school, 3=senior high school and above, marital status: 1=single, 2=married / cohabiting (reference), 3=other, smoking: 0=no (reference), 1=yes, drinking: 0=no (reference), 1=yes, region: 1=eastern part, 2=central part (reference), 3=western part, systolic blood pressure (continuous variable), diastolic blood pressure (continuous variable), diabetes: 0=no (reference), 1=yes, high-density lipoprotein cholesterol (continuous variable), self-rated poor health: 0=no (reference), 1=yes, body mass index (continuous variable), and waist circumference (continuous variable).

[0092] Lasso regression is used to screen predictive factors. The relationship between the penalty coefficient λ and the number of variables (including dummy variables) is shown in Figure 3 , the lower x-axis is log(λ), the upper x-axis is the number of variables in the model, and the y-axis is the (standardized) regression coefficient of the variable. The variable screening purpose is achieved by adjusting the penalty coefficient λ, where the λ that minimizes the model estimation error is considered to be the optimal case.

[0093] Specific screening process:

[0094] 1) Put all the predictors into the model and establish Lasso regression. The Lasso regression model is as follows:

[0095]

[0096] Where arg min represents the variable value when the objective function takes the minimum value; RSS represents the residual sum of squares;

[0097] 2) Lasso regression introduces penalty coefficients λ and ∑ j |β j | is the L1 penalty function. As the penalty coefficient increases, the regression coefficients of the variables decrease, and some will be compressed to 0, indicating that these variables contribute little to the model under this penalty coefficient and can be deleted. Therefore, the penalty coefficient determines the degree to which the regression coefficient is compressed and thus determines the predictive factors that are finally included in the model;

[0098] 3) The selection of penalty coefficient λ is crucial in the Lasso modeling process and can directly affect the degree to which the coefficients in the model are compressed. Therefore, it is necessary to determine the optimal penalty coefficient value. Common methods include cross-validation and information standard methods. The basic idea of ​​cross-validation is to divide the data into K groups of equal size, then select an extreme value of λ (generally an upward extreme value), use Lasso to fit K-1 data, use the coefficients estimated in the previous step to predict the Kth data, and then repeat this process. Record the predicted values ​​under each K value, average the K values, and calculate all standard errors for a specific λ value. Usually, 100 λ values ​​are selected, and then the final parameter λ is determined by the size of the mean square error (MSE) value. There are two options for parameter λ in the cross-validation results, namely the λ value at the minimum mean square error and the λ value corresponding to the maximum value within one standard error of the minimum mean square error. The former may not be completely solved due to the relatively small degree of compression of the regression coefficient, and to a certain extent cannot meet the purpose of Lasso modeling. Therefore, some studies suggest that λ be selected as the maximum value within one standard error of the minimum mean square error.

[0099] In this example, cross validation is used to find the ideal model. When the results show the minimum mean square error, λ min =0.0008435731, that is Figure 4 Middle left dashed line; maximum value within one standard error of the minimum mean square error, λ 1se =0.01414531, that is Figure 4The middle right dotted line. When log(Lambda.min) = -7.077864 is optimal, 12 predictors are screened out, including age, gender, education level, smoking, drinking, region, systolic blood pressure, diastolic blood pressure, diabetes, high-density lipoprotein, waist circumference and self-rated health. Lambda.min still screened out many variables, which need further screening. However, when log(Lambda.1se) = -4.25837, two predictors were screened out, namely age and systolic blood pressure, and Lambda.1se screened out fewer variables. Therefore, predictor screening was carried out within the range of Lambda.min to Lambda.1se. After repeated screening and evaluation of the screened predictors, the evaluation principles include: 1) potential predictive ability, which is mainly based on professional knowledge and the frequency of occurrence in previous similar models; 2) predictors are not part of the outcome definition; 3) considering their cost, that is, the difficulty of obtaining the information, such as the cost, time, resources required and patient burden. Prioritize the easily available demographic data and medical history information, then consider the information of simple examinations, and finally consider the information of expensive examinations. Lambda is finally set to 0.003526, that is, log(Lambda) = -5.647591. In this example, seven predictive factors are selected: age, education level, systolic blood pressure, diastolic blood pressure, diabetes, waist circumference, and self-rated health.

[0100] Lasso regression was used to screen out seven predictors (age, education level, systolic blood pressure, diastolic blood pressure, diabetes, waist circumference and self-rated health), and the Cox proportional hazards model was used to fit the prediction model.

[0101] S3.2. Based on the prediction factors screened by Lasso regression, the model fitted by the present invention is:

[0102]

[0103]

[0104] Compared with the previous fitting model that divided the educational level into two categories (junior high school and below, senior high school and above), the current fitting model shows that it is more reasonable to divide the educational level into three categories (primary school and below, junior high school, senior high school and above), that is, to divide it into low, medium and high according to the highest educational level. People with junior high school education and people with primary school education and below have the same age, systolic blood pressure, diastolic blood pressure, diabetes, waist circumference and self-rated health. Figure 5 The nomogram shows that there is a gap of more than 1.5 points between the two, which suggests that based on the principle of a simple model, this classification will be more accurate.

[0105] The 6-year stroke risk prediction probability value of the present invention is:

[0106] P = 1-0.9999458^exp(0.072664627*age (continuous variable, years)-0.078075715*education level (junior high school)-0.197020818*education level (high school and above)+0.007298872*systolic blood pressure (continuous variable, mmHg)+0.013844382*diastolic blood pressure (continuous variable, mmHg)+0.154268582*diabetes (0=no, 1=yes)+0.006942786*waist circumference (continuous variable, cm)+0.425624388*poor self-rated health (0=no, 1=yes)).

[0107] The risk level is divided according to the predicted probability value of the population without outcome events. Since P does not conform to the normal distribution, it conforms to the normal distribution after log transformation. First, the mean after log transformation is calculated. and standard deviation (s), calculate and The calculated value is taken as the inverse natural logarithm as the cut-off value for grading. Risk level: low risk: <1.5%, general risk: 1.5%-7.9%, medium risk: 8%-17.9%, high risk: ≥18%.

[0108] S4. Input the information of the user to be evaluated into the risk prediction model, and calculate the comprehensive score based on the input information of the user to be evaluated. The information includes at least the final variables used in the risk prediction model, for example, in this embodiment, the information includes at least 7 predictors of age, education level, systolic blood pressure, diastolic blood pressure, diabetes, waist circumference and self-rated health.

[0109] S5. Predict the risk level of the user to be assessed based on the comprehensive score and display it through the interactive interface, where: low risk: <1.5%, general risk: 1.5%-7.9%, medium risk: 8%-17.9%, high risk: >18%.

[0110] In a specific embodiment, the database includes 36,195 subjects from the 2010 baseline survey as baseline data, and 27,436 subjects were actually followed up from 2016 to 2017. Baseline cancer patients, subjects with missing baseline waist circumference, subjects with missing follow-up outcomes, subjects with incorrect outcome diagnosis dates, and subjects with abnormal baseline BMI and waist circumference were excluded, and finally 27,112 subjects were included in the analysis. Among them, 576 people died (2.1%), 12,259 men (45.2%), and 14,853 women (54.8%).

[0111] For continuous data, if they meet normal or approximately normal distribution, the mean ± standard deviation is used. The t-test or analysis of variance was used for comparison between groups. For continuous data with skewed distribution, the median (lower quartile, upper quartile) [M (p 25 , p 75 )], and the rank sum test was used for inter-group comparison. For categorical data, frequency and constituent ratio [n(%)] were used to express, and the chi-square test or rank sum test was used for inter-group comparison.

[0112] According to the distribution and coordination of monitoring sites in the eastern, central and western regions of the 2010 China Chronic Disease and Risk Factor Monitoring Project, 60 monitoring sites in 11 project provinces were selected as follow-up sites for this study. Based on the inclusion and exclusion criteria of this study, a total of 36,195 research subjects were finally included as the follow-up population for this cohort study, including 16,531 males (45.7%) and 19,664 females (54.3%). A follow-up survey was conducted in 2016-2017, and 8,759 subjects were lost to follow-up. A total of 27,436 subjects were actually followed up (follow-up rate 75.8%), including 12,402 males (45.2%) and 15,034 females (54.8%). After removing 68 patients with baseline cancer, 149 subjects with missing baseline waist circumference information and missing follow-up outcomes, 52 subjects with incorrect outcome diagnosis dates, and 55 subjects with abnormal baseline BMI and waist circumference, a total of 27,112 subjects were finally included in this analysis, including 576 deaths. Figure 2 shown.

[0113] A total of 27,112 subjects were included in the analysis, including 12,259 males (45.2%) and 14,853 females (54.8%), with an average age of (47.8±13.9) years. Among the subjects, 5,658 (20.9%) were 60 years old and above, 10,563 (39.0%) were urban residents, and people from the eastern, central and western regions accounted for 31.6%, 48.2% and 20.2%, respectively. 7,527 (27.8%) were current smokers, 7,015 (57.2%) were current male smokers, and 512 (3.4%) were current female smokers. 9,874 (36.4%) were drinkers, 7,445 (60.7%) were male drinkers, and 2,429 (16.4%) were female drinkers. The obesity rate was 13.3%, 11.8% for men, and 14.6% for women. The prevalence of prediabetes was 15.3%, 15.6% for males and 15.1% for females. The prevalence of diabetes was 10.2%, 10.7% for males and 9.8% for females. The prevalence of hypertension was 39.3%, 40.3% for males and 38.5% for females. The prevalence of dyslipidemia was 48.5%, 52.5% for males and 45.1% for females. Among the subjects, 21.4% of the subjects consumed too much red meat, and 80.4% of the subjects consumed insufficient amounts of vegetables and fruits.

[0114] The variable assignment table is as follows:

[0115]

[0116] Based on 27,112 research subjects, this study excluded the deceased (576 people), the people with less than 1 year of follow-up time (80 people), the people under 35 years old and over 75 years old (5,919 people), and finally included 20,537 research subjects to establish the model, including 1,174 people with first stroke. According to the median follow-up time and the maximum follow-up time, the prediction period of the prediction model is limited to 6 years.

[0117] In step S4, the stroke prediction model of the present invention is displayed using a nomogram (see Figure 5 ). The implementation principle of the nomogram of the present invention is to integrate multiple predictive factors based on the prediction model analysis, that is, to calculate each score based on the value of the individual predictive variable, and to obtain a total score by summarizing, and then calculate the risk probability of stroke based on the total score. The advantage of the nomogram is that it can be used outside of computers or in areas with limited network. The specific implementation process of the nomogram is as follows:

[0118] S4.1 constructs a prediction model, and assigns points to each value level of each predictor according to the contribution of each predictor in the model to the outcome variable, thereby obtaining a score for each predictor. The contribution degree is the value of the regression coefficient. The larger the regression coefficient, the greater the contribution degree and the higher the score.

[0119] S4.2 Add the scores of each predictor to get the total score;

[0120] S4.3 Calculate the risk probability of the individual outcome event through the functional conversion relationship between the total score and the probability of occurrence of the individual outcome event;

[0121] S4.4 uses scaled line segments to plot scores and risk probabilities according to a set ratio to express the relationship between the various predictive factors in the prediction model.

[0122] The present invention uses a nomogram to display the model, such as Figure 5 The horizontal axis is the score scale from top to bottom, age (years), education level, systolic blood pressure (mmHg), diastolic blood pressure (mmHg), diabetes, waist circumference (cm) and self-assessed health predictors, total score and 6-year stroke risk probability:

[0123] Usage example: A resident is 60 years old (score: 62.5 points), has an elementary school education or below (score: 6.5 points), systolic blood pressure: 180 mmHg (score: 30 points), diastolic blood pressure: 100 mmHg (score: 32.5 points), has no diabetes (score: 0 points), waist circumference: 110 cm (score: 13.5 points), and self-rated health is poor (score: 13.5 points). The total score of the resident is 62.5+6.5+30+32.5+0+13.5+13.5=158.5 points, and the corresponding risk of stroke is approximately 20%, which is a high risk. The stroke risk prediction probability calculation formula is used: P = 1-0.9999458^exp(0.072664627*age-0.078075715*education level (junior high school)-0.197020818*education level (high school and above)+0.007298872*systolic blood pressure+0.013844382*diastolic blood pressure+0.154268582*diabetes+0.006942786*waist circumference+0.425624388*self-rated poor health), and the resident's risk probability is calculated to be 18.69%, which is greater than the threshold of 18% and is a high risk.

[0124] Any process or method description in the flowchart of the present invention or described in other ways herein can be understood as a module, segment or part of a code including one or more executable instructions for implementing the steps of a specific logical function or process, which can be implemented in any computer-readable medium for use by an instruction execution system, device or equipment, and the computer-readable medium can be any medium containing storage, communication, propagation or transmission programs for use by execution systems, devices or equipment, including read-only memories, magnetic disks or optical disks, etc.

[0125] In the description of this specification, the description with reference to the terms "embodiment", "example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. In addition, those skilled in the art can combine or combine different embodiments or examples described in this specification and the features therein without causing any contradiction.

[0126] Although the above content has shown and described the embodiments of the present invention, it can be understood that the above embodiments are exemplary and cannot be understood as limitations of the present invention. Ordinary technicians in the field can change, modify, replace, modify and other update operations on the above embodiments within the scope of the present invention.

Claims

1. A method for assessing the risk of stroke, characterized in that: The method comprises the following steps: S1. Sample collection, collect sample data and update samples in the database in real time; S2. Create a risk prediction model based on the current database; the fitted prediction model is in the form of: Among them, λ(t|Xi) is the risk function of being exposed to a factor at time t; λ0(t) is the risk function of not being exposed to the factor; S3. Select several variables whose correlation strength with stroke risk meets the predetermined requirements from the factors included in the model; use the Cox proportional hazard regression model to conduct univariate analysis, preliminarily determine the adjustment factors, and then use Lasso regression to screen predictive factors, and achieve the purpose of variable screening by adjusting the penalty coefficient λ; S4. Input the information of the user to be evaluated into the risk prediction model, and calculate a comprehensive score based on the input information of the user to be evaluated; S5. predict the risk level of the user to be assessed based on the comprehensive score and display it through the interactive interface; The fitted model is: The predicted probability value of stroke risk: P = 1-0.9999458^exp(0.072664627*age-0.078075715*education level (junior high school)-0.197020818*education level (high school and above)+0.007298872*systolic blood pressure+0.013844382*diastolic blood pressure+0.154268582*diabetes+0.006942786*waist circumference+0.425624388*self-rated poor health).

2. The method for assessing the risk of stroke according to claim 1, characterized in that: In step S1, the sample size is calculated using a special formula for calculating the sample size of a cohort study, and the formula is as follows: Where: p1 and p0 are the expected incidence rates of the exposed group and the control group, respectively. is the average of the two incidence rates, q = 1-p, Z α and Z β is the standard normal distribution boundary value; Assuming α = 0.05 (two-sided), β = 0.10, Z α =1.96,Z β =1.282, the incidence of stroke in the non-exposed group p0 = 0.56%, RR = 1.6; Substituting into formula (2), the sample size is calculated to be 26910 people; The EPV strategy was adopted, and it was set that at least 10 events were required for each predictor variable to be included in the prediction model; the calculation formula was as follows: In the formula is the event rate studied by the prediction model, n var is the potential number of variables when fitting the model.

3. The method for assessing the risk of stroke according to claim 2, characterized in that: The step S1 includes database cleaning and variable cleaning. Database cleaning: merge the baseline database and the follow-up database to form a master database; in the process of database merging, the baseline data and the follow-up data are first accurately matched, and the ID number, name, and gender are all consistent; if the baseline or follow-up ID number is missing, the date of birth is used for fuzzy matching, and the matching principle is that the date of birth, name, and gender are all consistent; after matching the date of birth and age and gender, if it still cannot be matched as the same person, the object is eliminated; the method of variable cleaning is as follows: 1) variable validity cleaning: including checking whether there is a logical error, if there is, correct it; 2) check whether there is an outlier, and correct or eliminate the observation with an outlier; 3) Variable integrity cleaning: For variables with missing values ​​less than 5%, the mean is used for interpolation.

4. The method for assessing the risk of stroke according to claim 1, characterized in that: In step S2, the predicted probability value of stroke risk is: P=1-S0(6)^exp(β1Xi1+…+βpXip) (4) Among them, S0(6) is the 6-year basic survival probability; β1Xi1+…+βpXip is the linear prediction value; The factors initially included in the model included: age, sex: 0=male, 1=female, education level: 1=primary school and below, 2=junior high school, 3=senior high school and above, marital status: 1=single, 2=married or cohabiting, 3=other, smoking: 0=no, 1=yes, drinking: 0=no, 1=yes, region: 1=eastern, 2=central, 3=western, systolic blood pressure, diastolic blood pressure, diabetes: 0=no, 1=yes, high-density lipoprotein cholesterol, self-rated poor health: 0=no, 1=yes, body mass index, waist circumference, 14 variables; among them, age, systolic blood pressure, diastolic blood pressure, high-density lipoprotein cholesterol, body mass index and waist circumference were continuous variables.

5. The method for assessing the risk of stroke according to claim 1, characterized in that: The step S3 includes: S3.1 Screening of predictors; The Cox proportional hazard regression model was used for univariate analysis to preliminarily determine the adjustment factors. Lasso regression was used for predictive factor screening, and the penalty coefficient λ was adjusted to achieve the purpose of variable screening. The penalty coefficient λ that minimized the model estimation error was considered to be the optimal one. S3.

2. Predictors selected using Lasso regression.

6. The method for assessing the risk of stroke according to claim 4, characterized in that: The risk level is determined based on the calculated predicted probability value of stroke risk: low risk: <1.5%, general risk: 1.5%-7.9%, medium risk: 8%-17.9%, high risk: ≥18%.

7. The method for assessing the risk of stroke according to claim 1, characterized in that: For continuous data, if they meet normal or approximately normal distribution, the mean ± standard deviation is used. The t test or analysis of variance was used for comparison between groups; for skewed continuous data, the median [M(p 25 , p 75 )], and the rank sum test was used for inter-group comparison; for categorical data, frequency and constituent ratio [n(%)] were used to express them, and the chi-square test or rank sum test was used for inter-group comparison.

8. A stroke risk assessment system, characterized in that: The system is used to implement the stroke risk assessment method according to any one of claims 1 to 7, and the system includes a sample collection module, a model building module, a variable determination module, a prediction module and a display output module.

9. The stroke risk assessment system according to claim 8, characterized in that: In the system, the fitted prediction model is in the form of: Among them, λ(t|Xi) is the risk function of being exposed to a factor at time t; λ0(t) is the risk function of not being exposed to the factor.

10. The stroke risk assessment system according to claim 8, characterized in that: in, The sample collection module is used to collect sample data and update the basic database in real time; The model building module is used to create a risk prediction model based on the current basic database; the factors initially included in the model include: age, gender: 0 = male, 1 = female, education level: 1 = primary school and below, 2 = junior high school, 3 = high school and above, marital status: 1 = single, 2 = married or cohabiting, 3 = other, smoking: 0 = no, 1 = yes, drinking: 0 = no, 1 = yes, region: 1 = east, 2 = central, 3 = west, systolic blood pressure, diastolic blood pressure, diabetes: 0 = no, 1 = yes, high-density lipoprotein cholesterol, self-rated poor health: 0 = no, 1 = yes, body mass index, waist circumference 14 variables; among them, age, systolic blood pressure, diastolic blood pressure, high-density lipoprotein cholesterol, body mass index and waist circumference are continuous variables; The variable screening module is used to screen out several variables whose correlation strength with the risk of stroke meets the predetermined requirements from the factors included in the model; The prediction module is used to input the information of the user to be evaluated into the risk prediction model, and calculate the comprehensive score based on the input user information to be evaluated; the information at least includes the final variables used in the risk prediction model; The display output module is used to predict the risk level of the user to be assessed based on the comprehensive score and display it through the interactive interface.

Images