Using logistic regression models to determine factors affecting diabetes in the red sea state

Diabetes is a disease influenced by some factors related to individual's behavior and the surrounding environment. This study aimed to identify the factors influencing diabetes infection in the Red Sea State for the period (2015-2018). The study depended on data collected from a diabetes treatment center in Port Sudan city eastern of Sudan. The sample size was (202). Data were analyzed using chi-square test, simple Binary logistic regression, and multiple logistic regression. Chi-square test results showed that there was a relationship between diabetes infection and the predictor variables gender, sex, family history, and kinship. Application of the binary logistic regression revealed that there was a significant association between the predictor variables: sex, age, and diabetes infection. However, educational level, income, and blood sugar were found to be statistically insignificant. The study recommended increasing the number of diabetes screening and treatment centers in all cities in the Red Sea State and developing existing centers.

Illustrations explain the paper; charts are replotted from the paper's own table values.

AI-generated illustration

↗ Enlarge

Significant Risk Factors for DiabetesClose ×

AI-generated conceptual illustration (Human adults of diverse ages and genders). It explains the idea and does not show study data.

Open full-size image ↗

Replotted from source data

↗ Enlarge

Chi-square: Gender vs age vs Family history vs KinshipClose ×

Replotted from Table 2 (page 3) of the original paper. Every bar is the paper's own table value.

Open full-size image ↗

Replotted from source data

↗ Enlarge

Odds ratio: (11-20) vs (21-30) vs (31-40) vs (41-50) vs (51-60)Close ×

Replotted from Table 3 (page 4) of the original paper. Every bar is the paper's own table value.

Open full-size image ↗

Introduction

Sudan is the one of largest countries in Africa. It is located in the North -east of the African continent and shares borders with nine countries. Its population, which is multi-ethnic, is 40 million with an annual growth of 1.7% in rural areas and 4% in urban areas. Sudan is comprised of 18 states, and there are 253 government hospitals and many smaller health units most of which are poorly prepared (MOH, 2001) [15] . This situation makes it difficult for these hospitals and health units to cope with various medical cases including diabetes. Diabetes is a sever chronic disease occurring either when the pancreas generates little amount of insulin or when the body cannot adequately benefit from the insulin it makes. The incidence of diabetes has been steadily increasing over the past few decades. Globally, 422 million adults were infected with diabetes in 2014 in comparison to 108 million in 1980 (WHO, 2016) [22] . It was reported that 282 million were diabetic worldwide and 80% were living in low and middle-income countries (Guariguata, 2014) [12] . In 2014, the global pervasiveness of diabetes was estimated to be 9% among adults aged above 18. In 2012, 1.5 million death cases resulted from diabetes worldwide whereas it is not only ranked the 8 th major cause of death among both sexes but it is also regarded the 5 th leading cause of death among women (Gadsby, 2002) [11] . According to Sudan Centre of Bureau of Statistics (CBS, 2010) [8] , the incidence of diabetes in Sudan increased from 9% in 2010 to 10.6% in 2013. Diabetes was formerly regarded a rare disease in Sudan, but this scene has changed dramatically in the previous 40 years when the tendency in infection have started to rise. As in many other low-and middle-income countries (LMICs), diabetes popularity is estimated to be 8%, but could it be even higher in some northern states of the country, reaching figures of about 19% (Elmadhoun, 2016) [10] . Diabetes type 1 was reported to infect (83.8%) of Sudanese people whereas type 2 infections are (85%) of the total diabetes conditions (Noor, (2017) [18] . Diabetes increased incidence in Sudan is in line with the emerging global pandemics of metabolic syndrome and obesity (Popkin, 2012) [19] . In Sudan, the prevalence of fatty liver is thought to be around 20% in individuals without diabetes and risk factors were related to obesity and an increase in age. However, among individuals with type 2 diabetes the prevalence was around 50.3%.

The risk factors were overweight, obesity, central obesity, high triglyceride level and low HDL-c level. (Almobarak, 2015) [3] .

Objective of the Study

The study aimed to identify the factors affecting diabetes infection in the Red Sea State for the period (2015-2018).

Method of data analysis 3.1 Source of data

The study conducted in the Red Sea State in eastern of Sudan based on data from diabetes center in Port Sudan, mainly in Salalab District. The data was collected from the patients who visited the center depending on cross-section data for the years (15-18). The sample size was (202) members who entered the diabetes center for checking or diagnosis.

Dependent variable

The response variable of our study is diabetes occurrence or not. Hence, the response variable yi for the diabetic can be explained as: 1 for those entering the center and infected with diabetes, and 0 for those who were not diabetic.

Independent Variables

Firstly, we focused on a group of variables such as gender, age, tall, height, income, mass body index, infection, period of infection, kinship, and blood sugar. However, the information was not available for the entire variables; it either did not exist or incomplete. Therefore, we adopted the explanatory variables or independent variables for the study such as sex, age, educational level, income, kinship, and blood sugar.

Binary logistic regression

With binary logistic regression, we estimated the probability of a dichotomous response (which of course is also its mean) for various values of explanatory variables proposed by Willy (1996) [14] . We fit a following model of the form:

The term of the right side of the equation, called a logistic function. In which, the relationships between p and x are nonlinear and can be linearized. Let be the linear predictor where is defined by the transformation. This transformation is called the logit transformation of the probability p and the ratio called the Odds. Hence, multiple logistic regression model can be written as:

Fitting the logistic model

Logistic regression uses a maximum likelihood estimation technique to estimate the model parameters. In other words, ML finds the best values for n=ßo+ß1x1+……..+ßkxk Observed values of say, are the n independent random observations corresponding to the random variables (Y1, Y2, ….., Yn). Since Yi is a Bernoulli random variable with functional form and hence, the likelihood function is given by:

The estimation of parameters require the maximization of the likelihood function or equivalently the maximization of the natural logarithm of the likelihood function.

Literature Review

The previous studies related to the topic were summarized as follows: Suwannaphant and others (2017) [21] , aimed to determine the association between socioeconomic status and diabetes mellitus. They revealed that the prevalence of Diabetes Mellitus was 3.70% in 2010 and 8.11% for the year 2012. There was an association between socioeconomic status and diabetes mellitus. Respondents who were Females, old aged and of low educational attainment were vulnerable to diabetes mellitus. In addition, senthilevel (2012) [20] , concluded that the influencing factors of Diabetic retinopathy, one can protect patients from the development of diabetic retinopathy or reduce their probability of getting diabetic retinopathy. In the same line, Niyikora (2015) [17] , found that age, alcohol consumption, cholesterol level, occupation status, and hypertension were associated with the outcome of having diabetes. Explanatory variables like gender; smoking, family history of diabetes had weak association with being infected with diabetes. On the other side, Isnini (2018) [13] , concluded that attributes that impact on diabetes were nine including age, hemoglobin, sex, blood sugar pressure, white cell count, urea, total cholesterol, and BMI. Eliadarous (2017) [9] summarized that both self-rated health and health Utilities Index were lower in people with diabetes, compared with those without diabetes, and they were associated principally with pain, visual impairment and negative emotions. Ahmed and others (2001) [1] found that the prevalence of diabetes was (3.4%). Also, the problems of diabetes care in Sudan included shortage in equipped diabetes care centers, scarcity of trained specialists, expensive medications, weak coping with medicines and nutrition and illiteracy and inaccurate beliefs. Balla and others (2014) [6] revealed that the prevalence of diabetes is increasing in rural population of Sudan and knowledge about diabetes is low. Bakhet (2015) [4] showed that age, height, gender, sugar percentage in blood influence the level of diabetes. Alameen (2010) [2] , concluded that the independent variables of the study: the duration of the disease, the highest level of sugar in the blood, the lowest level of blood sugar, type, weight, blood pressure, follow-up, injury of a family member associated with diabetes. According to Balaji (2011) [5] , there were lower attendance rates in clinics (55%) and non-compliance to dietary requirements (79%), since most of the patients never receive diabetes education and care.

Statistical Data Analysis 6.1 Descriptive Statistics Results

The descriptive statistics for the variables in the current study showed that, for the gender there120 (59.4%) of the respondents were males and 82 (40.6%) were females. In terms of the participants' age, it was found that the highest number and percent for the age group was 68(33.7%) for the age group (41-50) year. The high number and percentages of educational level was (44.6%) for those who had secondary education, and 20 (9.9%) of the respondents were illiterate.

Regarding the income of the respondents, the high number and percentage were for the group (1000-3000) Sudanese pounds with 122 (60.4%). The variable infection of diabetes shows that there were 130 (64.4%) of the participants were diabetic and 72 (35.6%) of them were not infected with diabetes. The variable period of infection showed that there were 49 (24.3%) of the sample whose infection period was (11-20)

Chi-square test Results

The results of chi-square test presented in the table below, show an association between diabetes infection and the predictor variables such as gender, sex, family history, and kinship. Whereas the predictors; education, income, blood sugar found to be insignificant with diabetes infection.

Multiple Logistic Regression Results

Table (3) displays the results of the binary logistic regression, which shows the coefficients, standard error, Wald test for the predictors significant, odds ratio and coefficients of the factors associated with each of the predictor variables including the logistic regression model. It is obvious that there was a significant relationship between the variables: sex, age and diabetes infection. However, education level, income, and blood sugar were found to be statistically insignificant.

Goodness of Fitting the Logistic Regression Model

To check the adequacy of the model, Likelihood ratio (LR) tests, R 2 statistics, Hosmer-Lemeshow test and classified table were used. LR test revealed a significant difference between the likelihood ratios for the final model and the likelihood ratio for the model without predictor or reduced model. The results showed -2 Log Likelihood value for the reduced model and final model was 263.154 and 156.331, respectively. Chisquare statistics was 106.810 with (p-value = 0.000). This indicates that the final mode fitted well since the predictor variables had statistically significant impact on diabetes at 5% level of significance.

Model Summary of Binary Logistic Regression

Nagelkerke square was estimated at (0.564) indicating that the effect of the predictor variables was 56.4% in the dependent variable diabetes. i.e., 56.4% of the variance in patients who visited the diabetes centers whether they were positively diabetic, or they showed negative results was due to change in the explanatory variables.

Table 5: Model Summary

Step -2loglikelihood Cox and Snell R Square Negelkerke R Square 1 156.331 0.411 0.564

Testing the Goodness Fit of the Model

As shown by Hosmer-lemeshow test results, chi-square test statistics was 6.104 with p-value 0.636. Therefore, the study model was quite a good fit. This fact confirms that the estimated model fitted the data well.

Discussion

Descriptive statistics showed that 120 (59.4%) of the sample member were males and 82 (40.6%) were females. This result contradicts with the report of the Ministry of Health in the State of Khartoum in 2017 [16] , which revealed that women exported the list of diabetes, where the percentage of women with diabetes (85) thousand cases, while the incidence of diabetes for males were 73 thousand cases. The results obtained from the application of binary logistic regression showed that the factors gender, age, kinship have significant effects on diabetes infection at 5% level of significance. However, educational level, income, blood sugar were found to be statistically insignificant. This result is different from Alameen (2010) [2] study who found that blood sugar was associated with diabetes infection. On the other hand, the results revealed that having positive diabetes is significantly associated with the gender of those who entered the diabetes center in Port Sudan city and males more than females. This fact is confirmed by the odds ratio of the variable gender which was (0.273) implying that males were 0.273 times more likely to be infected with diabetes than females. Diabetes is associated with age, age group (51-60) presents odds ratio (1.662), this result indicates that those who were between 51-60 years old were 1.662 more likely to have diabetes than the other age groups. This result indicates that as the age of those who entered the diabetes center increases the possibility of being infected with diabetes increases. This finding is consistent with the finding by (CDC, 2011) [7] which found that in the age group 45-64 years, 13.7% had diabetic; and the highest percentage of 26.9% was found in the age group of ≥ 65 years. The same feature was also found in England, where the prevalence of diabetes was increasing with age. This result is in line with that result of Suwannaphant (2017) [21] and Isnini (2018) [13] . In addition, the kinship of the respondents was another factor that was significantly associated with diabetes; this is reflected in the result of the odds ratio (3.031), which indicates that individuals whose parents are diabetic; they are 3.031 more likely to be infected with diabetes than other classes of kinship such as brother, sister, grandfather/mother. This finding is consistent with the scientific fact that genetics has a relative effect on diabetes; especially a parent with diabetes that would expose children to diabetes. On the other side, the results showed that the educational level and blood sugar of the respondents had no significant effect on diabetes infection. This result contradicts with Bakhet's (2015) [4] study, which found that there was association between blood sugar and diabetes infection.

Conclusion

Diabetes in Sudan has been increasing in recent years. The study, conducted at the Center for Diabetes Therapy in Port Sudan, in the Red Sea State, concluded that the factors affecting diabetes are age, gender, and kinship. While variables such as educational level, income, blood sugar, do not have a relation with diabetes infection. Males were more infected with diabetes than females.

Recommendations

According to the results of the study, we propose the following recommendations: 1. A good information system must be provided to organize the information of those who attend the centers of diabetes in the Red Sea State. 2. Increasing the number of diabetes screening and treatment centers in all cities in the Red Sea State and developed existing centers. 3. Awareness of and knowledge about diabetes risk factors are needed to be raised. 4. More in-depth studies to further investigate factors that affect diabetes are needed to be conducted.

Figures and tables from the paper

Variable	classification	Number	Percentage
Gender	Male	120	59.4
	Female	82	40.6
Age	11-20 year	2	1
	21-30 year	11	5.4
	31-40 year	23	11.4
	41-50 year	68	33.7
	51-60 year	62	30.7
	More than 60	36	17.8
Education	Illiterate	43	21.3
	Primary	30	14.9
	Secondary	90	44.6
	University	19	9.4
	Post graduate	20	9.9
Income	Less than 1000 pound	63	31.2
	1000-3000 pound	122	60.4
	40000-100000 pound	17	8.4
Infection of diabetes	Negative	72	35.6
	Positive	130	64.4
Period of infection	1-5 year	9	4.5
	6-10 year	22	10.9
	11-20 year	49	24.3
	More than 20 year	43	21.3
	Missing	79	39.1
Infected family (history)	Yes	125	61.9
	No	77	38.1
Kinship	Parents	127	62.9
	Brother	25	12.4
	Sister	21	10.4
	Grandfather/mother	11	5.4
	other	18	8.9
Blood sugar	70-120 mg/dl	6	3
	121-150 mg/dl	67	33.2
	151-200 mg/dl	74	36.6
	More than 200 mg/dl	55	27.2

Table text extracted from the original PDF; superscript letters are shown inline. The original table image is below.

View original table image

↗ Enlarge

Table 1Close ×

Respondents characteristics

Open full-size image ↗

Original paper tableSource PDF · page 3 ↗

Predictor variable	Response variable	Chi-square	p-value	result
Gender	Diabetes infection	8.54	0.003	significant
age	Diabetes infection	31.29	0.000	significant
Education	Diabetes infection	8.43	0.077	insignificant
Income	Diabetes infection	3.90	0.142	insignificant
Family history	Diabetes infection	64.51	0.000	significant
Kinship	Diabetes infection	38.92	0.000	significant
Blood sugar	Diabetes infection	6.63	0.85	insignificant

Table text extracted from the original PDF; superscript letters are shown inline. The original table image is below.

View original table image

↗ Enlarge

Table 2Close ×

Results of Chi-square test

Open full-size image ↗

Original paper tableSource PDF · page 3 ↗

Variable	B	S.E	Wald	df	Sig	Odds ratio	95% C.l for Odds ratio	—
							Lower	Upper
Sex (1)	-1.566	0.541	8.362	1	0.004	4.786	1.656	13.831
Age			9.788	5	0.008
(11-20)	-19.78	27215.3	0.000	1	0.999	0.000	0.000	-
(21-30)	-2.05	1.129	3.294	1	0.07	0.129	0.014	1.178
(31-40)	-0.904	0.796	1.289	1	0.256	0.405	0.085	1.928
(41-50)	-0.803	0.664	1.463	1	0.227	0.448	0.122	1.646
(51-60)	0.508	0.679	0.531	1	0.466	1.662	0.424	6.520
Education			3.095	4	0.466
Illiterate	23.739	12272.6	0.000	1	0.998	20405814482	0.000
Primary	23.713	12272.6	0.000	1	0.998	19887752345	0.000
Secondary	23.202	12272.6	0.000	1	0.998	11920786610	0.000
University	22.157	12272.6	0.000	1	0.999	4194057486	0.000
Income			2.934	2	0.231
Less than 1000	-21.62	12272.6	0.000	1	0.999	0.000	0.000
1000-3000 p	-22.58	12272.6	0.000	1	0.999	0.000	0.000
Kinship			20.737	4	0.000
Parent	1.109	0.699	2.517	1	0.113	3.031	0.770	11.922
Brother	-1.488	0.940	2.506	1	0.113	0.226	0.036	1.425
sister	-1.107	0.904	1.501	1	0.221	0.330	0.056	1.943
Grand f/m	-21.83	11167.3	0.000	1	0.998	0.000	0.000
Blood sugar			0.126	3	0.989
70-120 mg/dl	20.791	15214.4	0.000	1	0.999	1070590768	0.000
121-150 mg/dl	-0.142	0.595	0.057	1	0.811	0.867	0.270	2.783
151-200 mg/dl	-0.181	0.517	0.123	1	0.726	0.834	0.302	2.300
Constant	-0.886	1.125	0.620	1	0.431	0.412

Table text extracted from the original PDF; superscript letters are shown inline. The original table image is below.

View original table image

↗ Enlarge

Table 3Close ×

Results of binary logistic model

Open full-size image ↗

Original paper tableSource PDF · page 4 ↗

Step Block step1 model	106.810	19	0.000
	106.810	19	0.000
	106.810	19	0.000

Table text extracted from the original PDF; superscript letters are shown inline. The original table image is below.

View original table image

↗ Enlarge

Table 4Close ×

Result of omnibus tests of model coefficients

Open full-size image ↗

Original paper tableSource PDF · page 4 ↗

↗ Enlarge

Table 5Close ×

Model Summary

Open full-size image ↗

Original paper figureSource PDF · page 4 ↗

↗ Enlarge

Table 7Close ×

Results of Classification Tablea

Open full-size image ↗

Original paper figureSource PDF · page 4 ↗

↗ Enlarge

Table 6Close ×

Result of hosmer-lemeshow test

Open full-size image ↗

Original paper figureSource PDF · page 4 ↗

1. Ahmed M, Ahmed NH. Diabetes mellitus in Sudan: The size of the problem and the possibilities of efficient care. Practical Diabetes international, 2001, 18(9).
2. Al-Ameen F. The distinctive function of distinguishing foot problems in diabetics, Omdurman Islamic University, unpublished master thesis, Khartoum, Sudan, 2010.
3. Almobarak AO, Barakat S, Suliman EA. Prevalence of and predictive factors for nonalcoholic fatty liver disease in Sudanese individuals with type 2 diabetes: is metabolic syndrome the culprit? Arab Journal Gastroenterol. 2015; 16:54-8.
4. Bakhet S. Using the Linear Discriminative function to indicate the most important effective factors of getting diabetes among mature people: A case study of GIAD Specialized Hospital, Khartoum, Sudan, 2015.
5. Balaji A. Quality of care among type 2 diabetes mellitus patients residing in an urban slum of Chennai Corporation --a community-based cross-sectional study. Journal of the Indian Medical Association. 2011; 109(7):462-468.
6. Balla SA, Ahmed HA, Awadelkareem MA. Prevalence of diabetes, knowledge and attitude of rural population towards diabetes and hypoglycemic event. American Journal of Health Research. 2014; 2(6):356-360.
7. Centers for Disease Control and Prevention. National diabetes fact sheet: national estimates and general information on diabetes and prediabetes in the United States. U.S. Department of Health and Human Services, Centers for Disease Control and Prevention: Atlanta, 2011.
8. Central Bureau of Statistics, Sudan. Sudan household survey. In statistics, editor: Khartoum, 2010.
9. ELaidarous H. Exploring the impact of diabetes in Sudan: Out-of-pocket expenditure and social consequences of diabetes on patients and their families. Department of Health Science, Karolinska Institute: Stockholm, Sweden, 2017.
10. Elmadhoun WM, Noor SK, Ibrahim AA, Bushara SO, Ahmed MH. Prevalence of diabetes mellitus and its risk factors in urban communities of north Sudan: Populationbased study. Journal of diabetes. 2016; 8(6):839-46.
11. Gadsby R. Epidemiology of diabetes. Advanced drug Delivery Reviews. 2002; 54(9):1165-72.
12. Guariguata L, Whiting D, Hambleton I, Beagley J, Linnenkamp U, Shaw J. Global estimates of diabetes prevalence for 2013 and projections for 2035.Diabetes research and clinical practice. 2014; 103(2):137-49.
13. Isnini Y, Badriyah T, Syarif I. Influencing of Logistic Regression Models for prediction and analysis of Diabetes risk factors. International journal of Engineering Technology. 2018, 6(1).
14. John Willy, Sons Inc. An introduction to categorical data analysis. 3rded. New York, 1996.
15. Ministry of Health. Sudan Annual Statistical Report. Khartoum: Sudan, 2001.
16. Ministry of Health, State of Khartoum. Annual Report, Khartoum, Khartoum State, Sudan. 2017.
17. Niyikora S. Multiple Logistic Regression modeling on risk factors of diabetes: A case study of Gitwe Hospital (2011-2013). Unpublished master thesis, Applied Statistics Department, Jumbo Kenyatta University, 2015.
18. Noor SK, Elmadhoun WM, Bashara SO. Glycaemic control in Sudanese individuals with type 2 diabetes: Population based study. Diabetes Metab Syndr. 2017; 11(1):S147-51.
19. Popkin BM, Adair LS, Ng SW. Global nutrition transition and the pandemic of obesity in developing countries. Nutr Rev. 2012; 70:3-21.
20. Senthilevel V, Radhakrishnan R, Sathiyamoorthi R. Prediction of diabetes retinopathy among diabetics using binary logistic regression approach, Indian Journal of Medical Specialist. 2012; 3(1):18-20.
21. Suwannaphant K, Laohasiriwong W, Puttanapong N, Seangsuwan J, Phajan T. Association between socioeconomic status and diabetes mellitus: The national socioeconomic survey, 2010 and 2012. Journal of Clinical and Diagnosis Research. 2017; 11(7):LC18LC22.
22. World Health Organization. Global report on diabetes. www.who.int/diabetes/global-report. 2016.