Research on the Influence Factors that Possibly Lead to Diabetes
Pengzhou Xu
School of Mathematical Sciences, Inner Mongolia University, Hohhot, 010000, China
Keywords: Diabetes, Pathogenic Factors, Binary Logistic Model.
Abstract: Diabetes has become a serious public health problem worldwide. Previous studies have found that diabetes is
related to family genetics, age and high blood pressure, but there are other unknown factors worth
investigating. In this study, a Binary Logistic Model was used to process data from the American Behavioral
Risk Factor Surveillance System (BRFSS), published in 2015. The data included 30,691 men and women of
all income levels and age groups. The study concluded that while diabetes was not associated with Vegetable
Consumption, it had significant positive effects on Age, Gender, BMI, High Blood Pressure, High Cholesterol,
Smoking, Stroke and Difficulty Walking. There were significant negative effects on Exercise, Fruit
Consumption, Alcohol and Education. Among the factors closely related to diabetes, Stroke, Age, Difficulty
Walking, did not appear in previous studies. The research not only provides some new perspectives for
relevant medical personnel to study the pathogenesis of diabetes, but also helps diabetic patients to treat
diabetes in a timely manner. At the same time, it also plays a positive role in diabetes prevention.
1 INTRODUCTION
China's economic development has led to changes in
Chinese people's living habits, and the incidence of
diabetes has increased year by year (Wang et al.,
2021). In addition, from 2013 to 2020, the mortality
rate of urban diabetic patients in China has increased
significantly, and diabetes has become an important
public health problem in China (Zhu et al., 2020 & Li
et al., 2020). Therefore, understanding the causes of
diabetes is of great significance to control the
development and treatment of the disease and reduce
the mortality. The purpose of this paper is to study the
potential factors that lead to diabetes to help people
assess their own risk of diabetes, and to take a series
of protective and treatment measures
.
Diabetes is a metabolic disease, which is usually
caused by hereditary and long-term external
influences, causing organ lesions, resulting in lower
insulin secretion than normal levels (Bai, 2018 &
Robinson and Pickering 2024)). The occurrence
factors of diabetes are complex, and some scholars
have found that diabetes has a certain correlation with
age, BMI, overweight and hypertension (Zhang et al.,
2022). In addition, Du et al. found that smoking and
living conditions were related to diabetes (Du et al.,
2022). Based on these research results, this paper will
study whether 14 factors (Age, Gender, BMI, High
Blood Pressure, High Cholesterol, Smoking, Stroke,
Exercise, Fruit Consumption, Vegetable
Consumption, Alcohol, Difficulty Walking,
Education, Income) whether these factors are related
to diabetes. In a similar direction, Ye et al. used a
multi-factor logistic regression model and a mixed
graph model (Ye et al.,2024). Logistic regression
analysis model is a classic model with high efficiency
and simplicity. In the field of medical research, case-
control studies are needed to establish multiple paired
groups. The literature indicates that unhealthy
lifestyle (Smoking, Low Physical Pabor, Alcohol,
BMI) is closely related to diabetes, but the screening
population in the literature is mainly Chinese elderly,
and the possibility of participating in screening in
other age groups is not fully considered. Li et al.
established a structural equation model (SEM) (Li et
al.,2023), but the computational complexity of this
model was high, and a large amount of data needed to
be collected to ensure the accuracy of the calculation.
Moreover, the literature also did not fully consider the
possibility of other age groups participating in
screening. Duan et al. also adopted a Logistic
regression model, but only considered the
relationship between sleep time and diabetes, and did
not investigate more factors affecting the onset of
diabetes (Duan et al., 2019). Deng et al. used the PSO-
BP neural network model which combines Particle
Swarm Optimization (PSO) algorithm and
Backpropagation (BP) neural network, but this model
Xu, P.
Research on the Influence Factors that Possibly Lead to Diabetes.
DOI: 10.5220/0012983100004601
Paper published under CC license (CC BY-NC-ND 4.0)
In Proceedings of the 1st International Conference on Innovations in Applied Mathematics, Physics and Astronomy (IAMPA 2024), pages 31-37
ISBN: 978-989-758-722-1
Proceedings Copyright © 2024 by SCITEPRESS – Science and Technology Publications, Lda.
31
is sensitive to parameter setting. Improper Settings
will cause the algorithm to fail to converge (Wang et
al., 2019)
.
In short, the research on the influencing factors of
diabetes has attracted many scholars. This paper will
mainly use the binary Logistic regression model to
analyse the influencing factors of diabetes, and put
forward suggestions on the prevention of diabetes
according to the results.
2 METHODS
2.1 Data Sources
The data in this literature comes from the Kaggle
website, and the data is compiled by Alex Teboul
based on the American Behavioral Risk Factor
Surveillance System (BRFSS) in 2015, with a total of
30691 samples
.
2.2 Variable Selection
The data used in this paper count a total of 30691
people, including those who have and do not have
diabetes, of whom 13949 are male and 16742 are
women. The data contains 14 variables (Age, Gender,
BMI, High Blood Pressure, High Cholesterol,
Smoking, Stroke, Exercise, Fruit Consumption,
Vegetable Consumption, Alcohol, Difficulty
Walking, Education, Income) There are 21
features(Age, Gender, BMI, High Blood Pressure,
High Cholesterol, Cholesterol Tests, Heart Disease,
Health Insurance, Medical Exam Costs, Health
Status, Mental Health, Physical Health, Smoking,
Stroke, Exercise, Fruit Consumption, Vegetable
Consumption, Alcohol, Difficulty Walking,
Education, Income) in the original data, and since this
literature does not study the clinical significance in
depth, the sample features with strong clinical
significance are discarded
.
T
able 1 shows the number of people with this disease
and the number of people with diabetes. As shown in
Table 1, the symbols of each factor are shown above
for ease of writing. The data sample was 30691
people, of whom 15345 had diabetes.
Table 2 shows the number of females and males,
and diabetes patients in each education group. Among
the characteristics of education, 1-6 means never
receiving an education until graduating from college.
The education of people in the data is mainly
concentrated in technical school graduation and
university graduation
.
Table 1: Logogram and numbers of the 14 factors.
Elements Lo
g
o
g
ra
m
Numbe
r
Diabetes
Age
x
30691 15345
Gender
x
30691 15345
BMI
x
30691 15345
High Blood Pressure
x
17272 11586
High Cholesterol
x
16261 10380
Smoking
x
14600 7981
Stroke
x
1913 1422
Exercise
x
21594 9767
Fruit Consumption
x
18903 9117
Vegetable Consumption
x
24252 11625
Alcohol
x
1325 369
Difficulty Walking
x
7624 5614
Education
x
30691 15345
Income
x
30691 15345
Table 2: Gender distribution by education group.
Education 1 2 3 4 5 6
Number 35 714 1425 8281 8684 11552
Female 21 408 855 4621 4994 5843
Male 14 306 570 3660 3690 5709
Diabetes 21 504 936 4623 4516 4745
IAMPA 2024 - International Conference on Innovations in Applied Mathematics, Physics and Astronomy
32
Table 3: Gender distribution by income group.
Income 1 2 3 4 5 6 7
8
Number 1528 1891 2303 2888 3454 4435 4967
9225
Female 1075 1269 1507 1781 2030 2373 2506
4201
Male 453 622 796 1107 1424 2062 2461
5024
Diabetes 996 1285 1464 1739 1907 2275 2337
3342
Table 4: Gender distribution by age group.
Age [18-34] [35-49] [50-64] [65-79] [80-100]
Number 1929 4769 10992 10594 2362
Female 1043 2693 5990 5679 1337
Male 886 2121 5002 4915 1025
Diabetes 240 1454 5692 6579 1380
Table 5. Binary Logistic regression prediction accuracy.
Accuracy rate Recall rate Precision rate F1 AUC
0.728 0.728 0.729 0.728 0.803
Table 3 shows the number of females and males, and
diabetes patients in each income group. Among the
income characteristics, 1 to 8 respectively under the
annual revenues of 10000 USD to the annual
revenues of 75000 USD or above. In the data, people's
annual income is mainly distributed in more than
75000 USD
.
Table 4 shows the number of females and males,
and diabetes patients in each age group. In the data
used in this paper, people are mainly distributed in the
age range of 50-64 and 65-79 years old
.
2.3 Research Protocol
In this paper, the binary Logistic model was used for
analysis. Whether diabetes mellitus was a dependent
variable (Y), and 14 characteristics were independent
variables (X), with 0 representing no and 1
representing yes. Next, this paper uses SPSSAU and
SPSSPRO to analyse the relationship between the
effect of X on Y, the relationship between the 14
factors on diabetes
.
2.4 Model Principle
Binary Logistic regression is a generalized linear
regression, which is used to solve the study of
influencing factors in binary classification problems.
The vector form of the linear regression model is:
𝑓
𝑥
=𝑤
𝑥+𝑏 (1)
Unit step function is needed to transform the
results of regression model into 0 and 1 classification
results, and Sigmod function continuity and function
value distribution can well replace unit step function.
Its formula is as follows:
𝑓
𝑥
=
(2)
The independent variable x in the Sigmod
function is replaced by the linear regression equation
(1), and the prediction function of the Logistic model
is finally obtained as follows:
𝑃=
(3)
2.5 Model Testing
Since the quality of model fit is measured by model
prediction accuracy, F1 represents the harmonic
average of precision and recall, and AUC represents
a measure of the ability of the classifier to distinguish
classification. It can be seen from Table 5 that the
fitting quality of the model is acceptable
.
Research on the Influence Factors that Possibly Lead to Diabetes
33
Figure 1: The proportion of factors affecting diabetic patients.
3 RESULTS AND DISCUSSION
3.1 Correlation Analysis
Figure 1 shows the proportion of a range of factors
that contribute to diabetes in people with diabetes.
Factors that lead to diabetes include High blood
pressure, High cholesterol, Smoking, Stroke, No
exercise, No eating fruits, No eating vegetables,
Alcoholism, Difficulty walking, etc. The proportion
of diabetes samples was identified from the samples
and classified according to different influencing
factors to determine the probability of each risk
leading to diabetes
.
It can be seen from Figure 1 that stroke has the
greatest impact on diabetes at 74.33%, followed by
difficulty walking at 73.64%, and alcohol abuse has
the smallest impact on diabetes at 27.85%. Therefore,
of the characteristics covered in this dataset, stroke is
the most risk factor for developing diabetes, with
difficulty walking a close second.
Before the regression analysis, the correlation
between independent variables and dependent
variables was analysed. Table 6 shows the strength of
the correlation between diabetes mellitus and 14
factors (Age, Gender, BMI, High Blood Pressure,
High Cholesterol, Smoking, Stroke, Exercise, Fruit
Consumption, Vegetable Consumption, Alcohol,
Difficulty Walking, Education, Income) using
Pearson correlation coefficient. Whether these factors
are associated with diabetes can be preliminarily
judged by observing whether the p-value is greater
than 0.05. The p-value of all 14 factors were less than
0.01, indicating that 14 factors were significantly
related to diabetes. In the further study, Pearson
correlation coefficient was observed to analyse
whether the correlation was positive or negative.
Exercise, Fruit Consumption, Vegetable
Consumption, Alcohol, Education, Income were
significantly negatively correlated with diabetes,
while Age, Gender, BMI, High Blood Pressure, High
Cholesterol, Smoking, Stroke, Difficulty Walking,
were significantly positively correlated with diabetes
(Table 6).
Table 6: Pearson correlation analysis.
Factors
p
-value Correlation coefficient
x
0.000 0.279
x
0.000 0.041
x
0.000 0.286
x
0.000 0.388
x
0.000 0.294
x
0.000 0.089
x
0.000 0.125
x
0.000 -0.147
x
0.000 -0.045
x
0.000 -0.080
x
0.000 -0.094
x
0.000 0.272
x
0.000 -0.153
x
0.000 -0.208
All in all, these 14 factors are correlated with
diabetes. Although the correlation coefficients of
Gender, Smoking, Stroke, Exercise, Fruit
Consumption, Vegetable Consumption, Alcohol, and
IAMPA 2024 - International Conference on Innovations in Applied Mathematics, Physics and Astronomy
34
Education are less than 0.2, their p-values are all less
than 0.01, so the correlation exists
.
3.2 Model Results
As shown in Figure 2, the study introduced various
factors that could be related to diabetes into the
model, these include Age, Gender, BMI, High Blood
Pressure, High Cholesterol, Smoking, Stroke,
Exercise, Fruit Consumption, Vegetable
Consumption, Alcohol, Difficulty Walking,
Education, Income. Through calculation, this study
obtained the final linear regression equation:
P = 3.874 + 0.147x
+ 0.257x
+⋯+
0.510x
0.057x
0.086x
(4)
Table 7 is the final result of the model. By
observing whether the value of p is greater than 0.05,
those factors can be preliminarily judged to have an
impact on diabetes. First of all, the p value of
Vegetable Consumption is greater than 0.05, which
has no effect on diabetes. Age, Gender, BMI, High
Blood Pressure, High Cholesterol, Smoking, Stroke,
Exercise, Fruit Consumption, Alcohol, Difficulty
Walking, Education and Income were all less than
0.01, indicating that these factors had a particularly
strong relationship with diabetes. In the further study
of the influencing factors, the marginal effect
combined with the regression coefficient can
specifically reflect the influence of these 13 factors on
diabetes. It is reflected in: Age, Gender, BMI, High
Blood Pressure, High Cholesterol, Smoking, Stroke,
Difficulty Walking for each additional unit, the
changes (increases) of Diabetes were 115.8%,
125.3%, 108.3%, 248.6, 202.3%, 108.1% and
142.3%, respectively. Because these values are large,
these factors have a significant impact on developing
diabetes.
Figure 2: Variable-related schematic.
Table 7: Model results.
Elements regression coefficient standard error z value p-value 95% CI marginal effect
x
0.147 0.005 26.937 0.000 1.146 ~ 1.171 0.027
x
0.257 0.028 9.237 0.000 1.225 ~ 1.366 0.047
x
0.079 0.002 34.382 0.000 1.078 ~ 1.088 0.014
x
0.910 0.029 31.504 0.000 2.349 ~ 2.630 0.181
x
0.704 0.028 25.484 0.000 1.916 ~ 2.135 0.134
Research on the Influence Factors that Possibly Lead to Diabetes
35
x
0.078 0.028 2.825 0.005 1.024 ~ 1.142 0.014
x
0.352 0.060 5.864 0.000 1.264 ~ 1.600 0.064
x
-0.087 0.031 -2.777 0.005 0.862 ~ 0.975 -0.016
x
-0.027 0.029 -0.942 0.346 0.920 ~ 1.030 -0.005
x
-0.096 0.035 -2.768 0.006 0.849 ~ 0.972 -0.017
x
-0.790 0.072 -11.044 0.000 0.395 ~ 0.522 -0.142
x
0.510 0.035 14.519 0.000 1.555 ~ 1.785 0.095
x
-0.057 0.015 -3.822 0.000 0.917 ~ 0.973 -0.010
x
-0.086 0.007 -11.546 0.000 0.904 ~ 0.931 -0.016
Constant -3.874 - - - - -
Compared with other similar studies, they were
more likely to look at the effects of age-specific and
known factors on diabetes, such as overeating in the
elderly, high blood pressure in the elderly, and
sedentary adolescents. In contrast, the features
contained in this paper are more detailed and
comprehensive, which can not only avoid some errors
caused by univariate analysis, but also bring some
new perspectives to the study of diabetes. Not only
will this direction for relevant medical staff pointed
out that more treatment, can also be found in a timely
manner and prevent diabetes.
4 CONCLUSION
Based on the current study, which focuses on the
influencing factors that may contribute to the
development of diabetes, the conclusions are: Age,
Gender, BMI, High Blood Pressure, High
Cholesterol, Smoking, Stroke, and Difficulty
Walking may be associated with diabetes. Some of
these factors have been overlooked in previous
studies.
The research method used in this paper is binary
Logistic regression, which can better deal with
classification problems than the single factor analysis
method used in other studies. However, it is
undeniable that the classification accuracy of binary
logistic regression may not be high when processing
high-dimensional data, which may cause errors in the
research results of this paper. At the same time, binary
Logistic regression is more sensitive to the selection
of features, and may have low precision problems
when dealing with data sets with complex
relationships. In addition, due to the limited amount
of data and the fact that the sample did not cover all
age groups and races, the accuracy of the model
results may also be affected. However, this study still
has great advantages and values. First of all, the
selection of data features was screened, and after
excluding the features that had little relevance to this
study, Pearson correlation analysis was used to ensure
that all 14 features were relevant to Diabetes. After
ensuring the correlation between Diabetes and 14
characteristics, the graphical method is used to
visually show and compare the difference in the
proportion of different influencing factors in diabetic
patients, which makes the experimental process more
intuitive and the experimental results clearer. On the
other hand, it has a positive effect on the prevention
and treatment of diabetes. In addition to the known
factors associated with diabetes such as High blood
pressure, High blood lipids and Age, there are many
other factors associated with diabetes, such as Stroke
and Difficulty walking. Overall, although the
relationship between diabetes and these new possible
influencing factors needs further study, it is meant to
point the way for future research. And it can improve
the efficiency of prevention and treatment of diabetes
patients, timely detection of diabetes, improve the
survival rate of patients.
REFERENCES
Wang L, et al. 2021 Prevalence and Treatment of Diabetes
in China, 2013-2018. JAMA 326(24) 2498–2506.
Zhu W, et al. 2019 Mortality trend of diabetes mellitus in
China from 2006 to 2020. Journal of Binzhou
Medical College 46(04) 299-303+314.
Li Y, et al. 2020 Prevalence of diabetes recorded in
mainland China using 2018 diagnostic criteria from
the American Diabetes Association: national cross-
sectional study. BMJ.
Bai Y 2018 Introduction to the cause of diabetes. Science
and technology 2 168.
Robinson G N and Pickering R J 2024 Melanocortins and
their potential for the treatment prevention and
IAMPA 2024 - International Conference on Innovations in Applied Mathematics, Physics and Astronomy
36
amelioration of complications of
diabetes. Diabetology 5(1) 69-84.
Zhang X, et al. 2022 Chinese adults, early onset status and
influencing factors of study. Chinese journal
epidemiology.
Gao Y Gong L and Liang G 2022 Beijing's Daxing district
residents diabetes prevalence and influence factors
analysis. Chinese journal of clinical doctors 50(11)
1322-1325.
Du J, et al. 2022 Study on influencing factors of diabetes in
Chinese elderly people based on health ecology
model. Chronic disease prevention and control in
China 30(6) 457-460 +464.
Ye T T, et al. 2024 Association between unhealthy lifestyles
and hypertension, diabetes and dyslipidemia in old
adults in China. Chinese Journal of Epidemiology
45(03) 385-392.
Li X, et al. 2023 Prevalence of diabetes in rural elderly in
Yunnan Province and its influencing factors based on
structural equation model. The Chinese journal of
disease control 27(5) 546-550.
Duan S, et al. 2019 The relationship between nocturnal and
daytime sleep patterns and the prevalence of diabetes
mellitus in middle-aged and elderly people in China.
Ningxia Medical Journal 43(8) 703-707.
Wang S, et al. 2019 Risk factor analysis and screening
model construction of type 2 diabetes based on BP
neural network based on particle swarm optimization
algorithm. Western Chinese Medicine.
Research on the Influence Factors that Possibly Lead to Diabetes
37
