CN103198211B - Quantitative analysis method for influences of attack risk factors of type 2 diabetes on blood sugar - Google Patents

Abstract

The invention relates to a quantitative analysis method for the influence of type 2 diabetes risk factors on blood sugar, and belongs to the fields of biological information processing and medicine. The present invention first uses the C4.5 and EM clustering algorithms to realize the selection of important risk factors for morbidity; Realize the quantitative analysis of the influence of multiple factors on blood sugar. Compared with a large number of existing statistical methods, the present invention adopts a data mining method, and while fully considering the mutual influence among multiple factors, realizes the quantitative analysis of the impact of multiple factors on blood sugar in the refined population, greatly improving the accuracy of quantitative analysis. Accuracy rate, and can provide a judgment method for detailed intervention of individual disease. The invention can carry out intervention guidance on the onset of individual type 2 diabetes, not only can prevent or delay the onset, but also the method can be extended to the quantitative analysis of other disease risk factors.

Description

Quantitative analysis method for the influence of risk factors of type 2 diabetes on blood sugar

technical field

The invention relates to a quantitative analysis method for the influence of multiple factors on blood sugar, which belongs to the fields of biological information processing and medicine.

Background technique

Type 2 diabetes has become a major health problem worldwide. It is estimated that by 2025, 380 million people worldwide will suffer from diabetes. At present, my country has become the second largest country with diabetes after India. According to the survey by the Ministry of Health, there are about 3,000 new cases of diabetes every day in my country, and about 1.2 million new cases every year, of which about 95% are type 2 diabetes patients. Type 2 diabetes has become the third chronic disease that seriously affects human health after cancer and cardiovascular and cerebrovascular diseases. Its etiology is the result of the interaction of environmental factors, genetic factors, and lifestyle. At present, the risk factors that have gained consensus include aging, obesity and overweight, blood lipids, abnormal blood pressure levels, family history of diabetes, etc. Multiple factors work together to affect the increase in blood sugar levels and lead to disease.

Since type 2 diabetes is difficult to cure once it develops, if the risk factors are intervened before the onset, the incidence rate can be effectively reduced and the quality of life can be improved. Most of the relevant studies use statistical methods such as multiple regression, meta-analysis, and cox regression, and use relative risk to study the relationship between risk factors and the onset of disease. Studies by Hu F B from Harvard University have shown that overweight and obesity are the most important factors for the occurrence of type 2 diabetes. By comparison, it was found that 3.4% of women in the low-risk group had a relative risk of diabetes of 0.09, and 91% of the cases were caused by unhealthy living habits. Mhurchu C N et al. used cox regression method to report the relationship between body mass index and diabetes in the Asia-Pacific population, and found that reducing body mass index in this region can effectively reduce the incidence of diabetes. Or use multiple regression algorithm and meta-analysis. The research usually uses the relative risk to explain whether a certain factor is a risk factor related to type 2 diabetes, and gives a qualitative conclusion. The present invention adopts BP neural network algorithm to calculate sensitivity, quantitatively measures the impact of risk factors on blood sugar changes, reflects the impact of risk factors changes on blood sugar changes through sensitivity, and uses sensitivity comparison to illustrate the quantitative influence degree of risk factors on blood sugar changes , is to explore the process-related factors of the characteristics and regularity of blood sugar changes, and is used to guide corresponding intervention measures, control the rising trend of blood sugar as soon as possible, and achieve the purpose of preventing and controlling the occurrence of diabetes.

Contents of the invention

The purpose of the present invention is to propose a quantitative analysis method based on BP neural network in order to solve the problem of quantitative analysis of the influence of multiple factors on blood sugar.

The design principle of the present invention is: use C4.5 and EM clustering algorithm to screen out the main risk factors, in order to determine the object of quantitative analysis; To the national sample population physical examination data that does not suffer from type 2 diabetes, carry out according to gender and age Population division; BP neural network algorithm was used to quantitatively analyze the influence of risk factors on blood sugar changes. The present invention refines the population while screening out the risk factors, quantitatively analyzes the influence of multiple factors on blood sugar, and gives a quantitative representation of the influence of multiple factors on blood sugar in the refined population, and quantitatively ranks the multi-factors of different refined populations Different, provide a judgment method for individual fine-grained intervention.

Technical scheme of the present invention is realized through the following steps:

Step 1: Obtain the population physical examination data to form a national sample population physical examination data source S without type 2 diabetes.

The specific method is as follows: In order to obtain a complete and available data source through the actual physical examination data from 2001 to 2008, preprocess the physical examination data, firstly through data cleaning, fill in vacant values, identify isolated points, eliminate noise and correct inconsistencies in the data; Then data transformation includes data format transformation and data semantic transformation; finally, to ensure that the information is not lost, duplicate factors and factors with too many vacant values are deleted through data specification, and the data source of the national sample population physical examination S={s ₁ , s ₂ , s ₃ ,..., s _k }, where k is the total number of people undergoing physical examination after pretreatment.

Step 2, on the basis of step 1, screen the main risk factors. The specific process is as follows:

Step 2.1, data processing experiment parameter setting module. According to the data source S, select the algorithm for screening the main risk factors, and set the parameters of the algorithm.

Step 2.2, EM clustering algorithm module.

The specific method is as follows: conduct clustering experiments on the data source S to cluster P or q clusters, change the number and types of risk factors participating in the experiment, observe the experimental results, obtain clustering results that can better reflect the characteristics of the population, and record the participants. Clustered risk factors.

Step 2.3, EM clustering, C4.5 classification combination experiment.

The specific method is as follows: the participating factors of the EM clustering experiment are the best clustering factors obtained from the above clustering experiment, and the clustering experiment of clustering P or q is carried out, and the data source S is separated according to the health characteristics of different groups of people. Groups with different health characteristics were analyzed using the C4.5 algorithm, and the classification participation factors were all l-dimensional risk factors. The calibration thresholds of the classification experiments were R, V, T, and Z, respectively, and the corresponding classifications of groups with different health characteristics were obtained. decision tree.

In step 2.4, the experimental results are counted to obtain the main risk factors of the c dimension, and further screened to obtain the main risk factors of the u dimension according to medical knowledge. In step 3, according to gender and age, the data source S of the national sample population physical examination obtained in step 2 is divided to generate a refined population.

The specific method is as follows: first, divide by gender to obtain male population and female population; then divide by age older than e years and less than or equal to e years respectively, and obtain group d of refined population.

Step 4, use the refined population obtained in step 3 to train the BP neural network model, and then calculate the sensitivity of different risk factors to blood sugar, and use the sensitivity to achieve quantitative analysis.

Step 4.1, under the given main risk factor dimension u, use d groups of refined population training to generate d BP neural network models, and the generation method of each model is:

In step 4.1.1, the u-dimensional risk factors of the processed training data are selected as the input of the model, blood glucose is used as the output of the model, and the forward propagation of information and the back propagation of errors are used for training to generate a BP neural network model. The input risk factors are calculated and transmitted from the input layer to the output layer layer by layer through the hidden layer. Each layer of neurons only affects the state of the next layer of neurons. If the output layer does not get the expected output, the error change value of the output layer is calculated. Then carry out backpropagation, through the network, the error signal is transmitted back along the original connection path to adjust the weights of each neuron. After multiple iterations, until the average relative error is less than σ, the BP neural network model is generated by training, and the model output is calculated. mean relative error.

In step 4.1.2, input the verification data into the generated BP neural network model, calculate the output blood glucose value, and obtain the average relative error of the verification data through error calculation.

In step 4.2, the sensitivity of multiple factors to the influence of blood sugar is calculated through the BP neural network model. Sensitivity is to determine the contribution rate or degree of influence of the model parameters on the model output by analyzing the influence of different parameter combinations on the simulation effect of the model.

Set nL-1 forward network (n is the number of input variables of BP neural network model, L is the number of hidden layers of BP neural network model, 1 is the number of model output variables), the network output has the following form: y =f(x ₁ ,...,x _n ) (x is the input of the BP neural network model, and y is the output of the BP neural network model). Taking two input risk factors as an example, the sensitivity of the two input variables to the output variable was investigated by calculating the second-order partial derivative of the formula. Let the hidden layer activation function of the neural network be a logarithmic S-type function

via the Jacobian matrix

In the formula: T is the transposition operation of the matrix, m is the sample number of the data source used, and n is the number of input variables. Relating the jth input x _j change to the jth output y _j = f(x _j ) change means that the sensitivity of the network output depends on small perturbations of the input. For a neural network with n inputs, a hidden layer with L neurons and an output layer, the sensitivity of the input variables x _i and x _k to the output variable y on the tth sample is

In the formula: S ₁ is the first derivative of the output layer activation function to its input, and S ₂ is the second derivative of the output layer activation function to its input. is the response of the j-th hidden layer neuron on the t-th sample, v _j1 is the weight between the output neuron and the j-th hidden layer neuron, w _ij is the i-th input neuron and the j-th hidden layer neuron The weight between neurons, w _kj is the weight between the kth input neuron and the jth hidden layer neuron. Through the sensitivity analysis of different risk factors, the quantitative analysis of each risk factor on blood sugar changes was obtained.

Beneficial effect

Compared with a large number of statistical analysis methods based on linear regression, meta-analysis, etc., the present invention adopts the data mining method of BP neural network to realize the quantitative analysis of blood sugar changes, and has the characteristics of high accuracy.

Compared with the group analysis, the present invention adopts the group division technology, which has a higher accuracy rate, more targeted analysis of blood sugar changes, and provides judgment basis for individual fine-grained intervention, so as to prevent or delay the occurrence of type 2 diabetes . The present invention can be applied to quantitative analysis of other disease risk factors, and can also be applied to a virtuous cycle of factor intervention-judgment-factor intervention, thereby effectively improving the health level of individuals.

Description of drawings

Fig. 1 is the schematic diagram of the method for quantitative analysis of the influence of multiple factors on blood sugar of the present invention;

Fig. 2 is a schematic diagram of data preprocessing in a specific embodiment;

Fig. 3 is the flow chart of clustering experiments in the specific embodiment;

Fig. 4 is the flow chart of clustering, classification combined experiments in the specific embodiment;

Fig. 5 is the crowd division method in the specific embodiment;

Fig. 6 is a schematic diagram of generating a BP neural network model in a specific embodiment;

Fig. 7 is a histogram of the sensitivity of risk factors in the specific embodiment.

detailed description

In order to better illustrate the purpose and advantages of the present invention, the implementation of the method of the present invention will be further described in detail below in conjunction with the accompanying drawings and implementation examples.

Taking 9,632 pieces of physical examination data from the physical examination center in 2007 and 2008 as input, design and deploy 9 groups of refined crowd verification: (1) verify the data of the whole population, (2) verify the data of males, (3) verify the data of Validate female data, (4) verify data for people older than 50 years old, (5) verify data for people older than 50 years old, (6) verify data for men who are older than 50 years old, (7) verify data for men who are younger than or equal to 50 years old Data verification, (8) data verification for women older than 50 years old, (9 data verification for women less than or equal to 50 years old.

The training data comes from the actual physical examination data from 2001 to 2008. A total of 59,839 undiseased physical examination data are used as input, including 34,377 male data, accounting for 57.4%, and 25,462 female data, accounting for 42.6%. World Health Organization (WHO) standard judgment. The gender and age distribution of the examinees is shown in Table 1. The verification data uses 9632 pieces of physical examination data from the physical examination center in 2007 and 2000.

Table 1. Gender and age distribution statistics of training data

Perform data preprocessing on the training data as shown in Figure 2 and crowd division as shown in Figure 3 to obtain nine groups of people, train and generate nine BP neural network models as shown in Figure 4, and calculate the average relative error:

In the formula: E is the average relative error of the model output, m is the number of samples of the data source used, y' _i is the blood glucose value of the ith sample output by the model, and y _i is the actual blood glucose value of the ith sample, which are calculated separately Average relative error for each model.

Perform the same data preprocessing and population division on the verification data to obtain nine groups of people, input the corresponding models respectively, calculate the output blood glucose value of each model, and then calculate the average relative error through the error to verify the accuracy of the model.

Table 2 The average relative error table output by nine groups of models

Through nine BP neural network models, the sensitivities of nine groups of risk factors to blood sugar changes are obtained, as shown in Table 3, and the corresponding sensitivity histograms of risk factors are shown in Figure 5, from left to right for all Sensitivity of different risk factors among the population, male population, female population, population over 50 years old, population less than or equal to 50 years old, male population >50 years old, female population >50 years old, male population ≤50 years old, and female population ≤50 years old .

Table 3 Sensitivity of disease risk factors to changes in blood sugar in the nine groups of people

Through the comparative analysis of the sensitivity of risk factors in different populations, the following results can be drawn:

1. Effect of body weight on blood sugar changes

Body weight is the most likely factor that causes changes in blood sugar. The sensitivity of body weight to blood glucose changes in the whole population was 0.2449. The degree of influence of body weight on blood sugar changes is not only reflected in the sensitivity of body weight ranking first in the sensitivity calculation of the whole population; Both rank first, and only rank second and third among people over 50 years old and women over 50 years old, respectively. The latter two groups are also considered to be due to the characteristics of the female population older than 50 years old.

2. Effect of blood lipid level on blood sugar changes

Changes in cholesterol levels are second only to body weight in affecting changes in blood sugar. The sensitivity to change of cholesterol level in the whole population was 0.2294. In people older than 50 years old and women, the effect of cholesterol level on blood sugar is higher than that in people younger than 50 years old and men. Among women over 50 years old, the influence of cholesterol level on blood glucose changes was 28% higher than that of men of the same age group (0.2538vs.0.1985). In the whole population, the effect of triglyceride level on blood sugar changes ranks fourth (0.1227), which is 47% lower than that of cholesterol level (0.2294vs.0.1227). However, in the male population, especially the male population older than 50 years old, the effect of triglyceride levels on blood sugar changes (0.1970) increased significantly by 60%.

3. The effect of age on blood sugar changes

In the whole population, the sensitivity of age to blood glucose changes was 0.1657, ranking third. Men are 4.7 times more sensitive to age than women (0.2192vs.0.0383).

In the whole population, the sensitivity of the above three factors to the influence of blood sugar changes has reached 64% of the current detection of risk factors. Considering that the sensitivity of triglycerides to blood sugar changes is 0.1227, the three factors of body weight, blood lipid level (cholesterol and triglycerides) and age can affect 77% of blood sugar changes, and the three factors are about 25% and 35% respectively. % and 17%. Overall, body weight, blood lipid levels (cholesterol and triglycerides), and age were the main factors that influenced changes in blood sugar.

4. Effect of gender on blood sugar changes

In the whole population, the sensitivity of gender is only 0.0091, and the gender factor has little effect on blood sugar changes in the whole population, which is also consistent with the phenomenon that the prevalence of diabetes has little difference between genders. However, after considering the age factor, the impact of gender on blood sugar changes still has a certain effect. In people older than 50 years, the effect of sex on blood glucose levels increased by nearly 2.5 times, and the sensitivity increased to 0.0315.

The specific impact of gender factors on blood glucose changes in different age groups is as follows:

1) In the male and female groups over 50 years old, the impact of changes in blood lipid levels is different. Men were more sensitive to changes in triglyceride levels (0.1970 vs. 0.1659), while changes in cholesterol and high-density lipoprotein levels had a greater effect on women's blood sugar (0.2538 vs. 0.1985; 0.1974 vs. 0.1437).

2) In the group of men younger than 50 years old, the weight factor showed the highest degree of influence (0.2911), which was 19%, 18% and 32% higher than the influence degree of the whole population group, the pure male group and the men over 50 years old group respectively %. The effect of changes in cholesterol and high-density lipoprotein levels on female blood sugar still existed in the group of women younger than 50 years old, and the effect of gender factors on the level of high-density lipoprotein was more obvious than that in the group older than 50 years old (0.1515vs.0.0371).

The following conclusions are further obtained: the quantitative analysis method based on BP neural network uses sensitive quantification to measure the influence degree of risk factors on blood sugar changes, realizes the transformation from qualitative analysis to quantitative calculation, and obtains the influence of different risk factors on blood sugar changes. sensitivity. (1) Changes in body weight are the most likely to cause changes in blood sugar, followed by cholesterol, age and triglycerides. Their sensitivity to blood sugar changes has reached 77% of the current detection of risk factors. Body weight, blood lipid levels (cholesterol and triglycerides) esters) and age accounted for about 25%, 35% and 17%, respectively. (2) The sensitivity of age to blood glucose changes was 0.1657, ranking third. Men are 4.8 times more sensitive to age than women (0.2192vs.0.0383). (3) The effect of gender on blood glucose changes was not significant in the whole population, with a sensitivity of 0.0091, but after considering age, gender had a certain effect on blood glucose changes. The effect of gender on blood glucose level is more obvious in people over 50 years old, and the sensitivity is 0.0315. Among people over 50 years old, the impact of triglycerides on men's blood sugar levels is 19% higher than that of women; the impact of body weight on men's blood sugar levels is 14% higher than that of women, and the impact of women's cholesterol levels on blood sugar changes is 28% higher than that of men; less than or equal to In the 50-year-old population, the protective effect of high-density lipoprotein on blood sugar changes is obvious.

Claims (3)

1. A method for quantitative analysis of the impact of risk factors for type 2 diabetes on blood sugar, characterized in that the method comprises the following steps: Step 1. Perform data cleaning on the actual physical examination data from 2001 to 2008, fill in vacancies, identify outliers, eliminate noise and correct inconsistencies in the data; then perform data transformation, including data format conversion and data semantic conversion; Under the condition that the information is not lost, the repeated factors and the factors with more blank values are deleted through data reduction, and the data source S of the national sample population without type 2 diabetes is formed; Step 2, use the EM clustering algorithm for the rough selection of risk factors on the data source S, and then use the risk factor selection method that combines EM clustering and C4.5 algorithm to screen the main risk factors that cause type 2 diabetes; Step 3. According to gender and age, divide the data source S of the national sample population physical examination obtained in step 1, and train the BP neural network model for the 9 groups of refined populations based on the risk factors obtained in step 2. Based on the weight of the BP neural network, use A sensitivity calculation method under the comprehensive action of multiple factors, which calculates the sensitivity of different risk factors to the influence of blood sugar, and then realizes quantitative analysis; Wherein, the sensitivity calculation method under the comprehensive action of multiple factors is as follows: nL-1 forward network is provided, where n is the number of input variables of the BP neural network model, and L is the number of hidden layers of the BP neural network model , 1 is the number of model output variables, and the network output has the following form: y=f(x ₁ ,…, x _n ), where x is the input of the BP neural network model, y is the output of the BP neural network model, through Find the second-order partial derivative of this formula to examine the sensitivity of the input variable to the output variable, and set the hidden layer activation function of the neural network as a logarithmic S-type function $\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; e</span>}}^{<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; x</span>}}}$ via the Jacobian matrix $\frac{\mathrm{<span\; class="notranslate">\; d</span>}\mathrm{<span\; class="notranslate">\; the\; y</span>}}{{\mathrm{<span\; class="notranslate">\; dx</span>}}^{\mathrm{<span\; class="notranslate">\; T</span>}}}<span\; class="notranslate">\; =</span>{<span\; class="notranslate">\; (</span>\frac{<span\; class="notranslate">\; \&part;</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}}{<span\; class="notranslate">\; \&part;</span>\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; )</span>}_{\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; no</span>}}$ In the formula: T is the transposition operation of the matrix, m is the number of samples of the data source used, n is the number of input variables, change the jth input x _j and the jth output y _j = f(x _j ) The connection means that the sensitivity of the network output depends on the small disturbance of the input. For a neural network with n inputs, a hidden layer with L neurons and an output layer, the input variables _xi and x _k on the tth sample The sensitivity to the output variable y is ${\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; L</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; j</span>}}{\mathrm{<span\; class="notranslate">\; v</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; )</span>\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; L</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}\mathrm{<span\; class="notranslate">\; j</span>}}{\mathrm{<span\; class="notranslate">\; v</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; L</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; j</span>}}{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}\mathrm{<span\; class="notranslate">\; j</span>}}{\mathrm{<span\; class="notranslate">\; v</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 2</span>}{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; )</span>$ In the formula: S ₁ is the first-order derivative of the output layer activation function to its input, S ₂ is the second-order derivative of the output layer activation function to its input, is the response of the j-th hidden layer neuron on the t-th sample, v _j1 is the weight between the output neuron and the j-th hidden layer neuron, w _ij is the i-th input neuron and the j-th hidden layer neuron The weight between neurons, w _kj is the weight between the kth input neuron and the jth hidden layer neuron. 2. The method according to claim 1, characterized in that, the data source S adopts the EM clustering algorithm to carry out the rough selection of risk factors, and then adopts the risk factor selection method of fusion EM clustering and C4.5 algorithm , the steps of screening the main risk factors for type 2 diabetes include: Step 2.1, select the algorithm for screening the main risk factors according to the data source S, and set the parameters of the algorithm; Step 2.2: Carry out a clustering experiment of clustering P or q on the data source S, change the number and type of risk factors involved in the experiment, observe the experimental results, obtain a clustering result that can better reflect the characteristics of the population, and record the participating clustering results. Class risk factors to achieve the purpose of rough selection of risk factors; In step 2.3, the participating factors of the EM clustering experiment are the best clustering factors obtained from the above clustering experiment, and the clustering experiment of clustering P or q is carried out, and the data source S is separated according to the health characteristics of different groups of people. C4.5 algorithm is used to analyze the groups with healthy characteristics, and the classification participation factors are all l-dimensional risk factors. The calibration thresholds of the classification experiments are A, B, C and D respectively, and the classification decisions corresponding to groups with different health characteristics are obtained. Tree; Step 2.4: Statisticalize the experimental results to obtain the main risk factors of the c-dimension, and further screen to obtain the main risk factors of the u-dimension according to medical knowledge, so as to achieve the purpose of selecting risk factors. 3. The method according to claim 1, characterized in that, according to age and sex, the national sampling population physical examination data source S obtained through step 1 is divided, and the BP neural network model is trained respectively for the refined crowd, based on the BP neural network The weight value calculates the sensitivity of different risk factors to the influence of blood sugar, and then realizes the quantitative analysis. The specific method is as follows: Step 3.1, according to age and gender, refine the population in step 1, first divide by gender, obtain male population and female population; then divide by age greater than e years and less than or equal to e years, and obtain a total of d group of refined populations ; Step 3.2, under the given main risk factor dimension u, use n groups of refined population training to generate n BP neural network models, the generation method of each model is: Step 3.2.1, select the u-dimensional risk factors of the processed training data as the input of the model, blood glucose as the output of the model, use the forward propagation of information and the back propagation of the error to train and generate the BP neural network model, and input the risk factors from The input layer is calculated layer by layer through the hidden layer and passed to the output layer. Each layer of neurons only affects the state of the next layer of neurons. If the output layer does not get the desired output, calculate the error change value of the output layer, and then reverse Propagation, the error signal is transmitted back along the original connection path through the network to adjust the weight of each neuron, and after multiple iterations, until the average relative error is less than σ, the BP neural network model is trained to generate the BP neural network model, and the average relative error of the model output is calculated; Step 3.2.2, input the verification data into the generated BP neural network model, calculate the output blood glucose value, and obtain the average relative error of the verification data through error calculation; Step 3.3, the sensitivity is to analyze the influence of different parameter combinations on the simulation effect of the model, determine the contribution rate or influence degree of the model parameters to the model output, and analyze the sensitivity of different risk factors to obtain the effect of each risk factor on blood glucose Quantitative analysis results of changes.