## Introduction

The epidemic of the century-Diabetes Mellitus (DM), often referred as simply diabetes, is a set of metabolic disorder or syndrome recognized as chronic hyperglycemia (presence of high blood sugar) occurs due to imperfections in insulin action, insulin secretion, or both1,2. It is now prevalent across the world with multiple complications3,4,5. In 2014, around 422 million people were affected by DM and 1.5 million deaths were reported in 20126,7. According to WHO report, 2016, 8% (12.88 million) of total population of Bangladesh was affected by diabetes whereas 3% of total deaths of all-ages occurred due to diabetes8. An increasing rate in the prevalence of DM among the Bangladeshi populace is also observed over time9.

Diabetes may result in a wide range of physiological as well as psychological problems including sexual disorder10. The lower sexual functions or dysfunctions termed as loss of libido may be observed in both females and males as a consequence of DM11,12,13,14,15,16,17,18. In addition, severe vision loss, acute renal diseases which may require dialysis or kidney transplant, myocardial infarction otherwise known as heart attack, cerebrovascular diseases like stroke, and hypertension are markedly observed19,20,21,22,23,24,25,26,27. Due to the intensity of the adverse effects of diabetes, it is important to find out the determinants to address the issue in order to contribute to improving country health situation.

Diabetes mellitus (DM) is a disease that relies on many factors and may vary over time and region. Accordingly, it requires to be screened on a constant basis. Previously, many studies explored the risk factors of DM by utilizing several statistical models. Among these models, a single level logistic model was very popular. However, the single level model depends on some strict assumptions which may not be possible to follow all the time, specifically, when someone is dealing with a dataset having hierarchical formation. Alternatively, one can apply a two-level regression model. Thus, our study intends to explore the prevalence of DM and its possible risk factors in Bangladesh by applying a two-level regression model since the necessary information was obtained from Bangladesh Demographic and Health Survey (BDHS) 2011 that follows a hierarchical structure.

## Materials and Methods

### Data Collection

This study utilized the BDHS 2011 dataset, which was collected by the National Institute of Population Research and Training (NIPORT) with the collaboration of Mitra and Associates (Bangladesh) and ICF international (USA)28. In this survey, a total of 83,731 household members were taken from 17,141 households by applying a two-stage stratified cluster sampling28. Although BDHS 2014 data are available, the necessary information for conducting this study such as fasting plasma glucose (FPG) is not available in BDHS 2014 and is very much accessible in BDHS 2011 survey. In 2011 survey, there were 3734 men and 3831 women aged from 35 years and above who participated in FPG measurement28.

### Dependent variable

For measuring fasting plasma glucose (FPG), the BDHS 2011 utilizes the cut-off points recommended by WHO29. If FPG ranges from 3.9 mmol/l (70 mg/dl) to 6.0 mmol/l (108 mg/dl), it was said to be normal29. The FPG value within the range between 6.1 to 6.9 mmol/l (110–124 mg/dl) was considered as pre-diabetic, and the FPG greater than or equal to 7.0 mmol/l (126 mg/dl) was said to be diabetic29. In this study, a participant was identified as a diabetic patient if his/her FPG value was greater than 6.1 mmol/l (110 mg/dl)30.

### Independent variables

A set of categorical explanatory variables was selected to fit the two level regression model. The selected variables were divided into two parts: level-1 variables and level-2 variables. At the level-1, we considered: sex (male, female), level of education (no education, primary education, secondary education, above secondary education), wealth index (poor, middle, rich), BMI (body mass index) [thin, if BMI is less than 18.5, normal, if BMI ranges from 18.5 to 24.9, overweight, if BMI is greater than 24.9] and working status (desk-work, physical work). The working status was used as a proxy variable for measuring physical activity31. Respondents were considered to be engaged in physical activities if his/her work liability includes physical activity related works31. This group includes agricultural worker, farmer, fisherman, cattle raising, poultry raising, rickshaw driver, road building, brick breaking, construction worker, domestic servant, boatman, beggar and factory worker31. The physically inactive group consists of land owner, unemployed/student, carpenter, tailor, doctor, nurse, dentist, lawyer, accountant, teacher, family welfare visitor, businessman, housewife, religious leader and retired person31. The level-2 variable includes region of residence which was classified as urban and rural.

### Statistical analysis

The relationship between chosen covariates and DM was analyzed by performing both bivariate and multivariate investigations. In the bivariate arrangement, chi-square test was utilized to identify the relationship between response and explanatory variables. Weighted prevalence of diabetes among Bangladeshi people was also calculated. In the multivariate arrangement, we applied a two-level logistic regression model to examine the risk factors of DM by reducing the regional effect that exists in the dataset. However, intra-class correlation coefficient (ICC) should be calculated prior to the application of any two-level model32. The mathematical form to calculate the ICC is

$$ICC=\frac{var({V}_{0j})}{var({V}_{0j})+({\pi }^{2}/3)}$$

Where, $$\,var({V}_{0j})$$ represents the variance of random intercept (variance of level-2). The range of ICC varies from 0 to 1. A two-level regression model is applicable, if the ICC is greater than 033.

The term $$var({V}_{0j})$$ should be calculated from two-level empty model. The functional form of the two-level empty model is-

$$logit({\pi }_{ij})={\alpha }_{0}+{V}_{0j}$$

in which $$logit\,({\pi }_{ij})=P({Y}_{ij}=1)$$, $${\alpha }_{0}$$ represents the intercept of fixed effect and $${V}_{0j}$$ represents the level-2 residual33. Finally, the mathematical form of the two-level random intercept model is-

$$logit({\pi }_{ij})={\alpha }_{0}+{\beta }_{10}{x}_{ij}+{\beta }_{01}{X}_{j}+{V}_{0j}$$

where $${x}_{ij}\,$$and Xj represent the level-1 and level-2, respectively, whereas β10 and β01 represent the fixed effects of the level-1 and level-2 variable, respectively33.

### Ethical Approval

This study depends on secondary dataset collected by NIPORT (Bangladesh) and MEASURE DHS. Every legitimate strategy was performed including human participants as per the moral norms of the national research committee and with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards28. The information records are unreservedly accessible from the website: www.dhsprogram.com. We got approval from the DHS program for utilizing the applicable datasets for this investigation. Ethical clearance for the BDHS 2011 data collection was taken from the ICF International’s IRB (Institutional Review Board)28. This survey confirmed international ethical standards of anonymity, confidentiality, and informed consent28. A letter of data authorization was taken from the Demographic and Health Surveys (DHS) Program, ICF International. The authority of Statistics Discipline in Khulna University, Bangladesh additionally gave us the chance to direct the examination.

### Informed consent

Informed consent has been taken from all participants associated with this research.

## Results

The background characteristics of study participants by diabetes status and weighted prevalence of diabetes mellitus with 95% confidence interval were displayed in Table 1. It was observed that the prevalence of diabetes increases with increasing age of the participants. The prevalence was almost equal among female (32.4%) and male (32.5%). Considering the education level, the higher rate of DM was observed among higher educated participants. Subsequently, the rich group had highest prevalence of diabetes (36.7%) than their counterparts. Prevalence of diabetes was higher in urban area (35.9%) than the rural area (30.8%). In our study, the prevalence rate of DM was higher among thin (31.6%), overweighed (48.2%), participants having hypertension (23.2%) and not engaged in physical work (32.8%). Except the sex of the participants, all the selected covariates were found to be statistically significant (p < 0.05).

In multivariate analysis, at first, we estimated a two-level empty model to obtain the value of intra-class correlation (ICC) within the divisions. The Table 2 is showing the value of ICC which is 0.019, suggesting that the two-level regression model can be applied to identify the risk factors for diabetes mellitus. The estimates of two-level model were displayed in Table 3. From this table, it was identified that the intercept of random part was statistically significant (p < 0.001), representing that the DM varies among divisions of Bangladesh. From fixed effects, we observed that the chance of having diabetes increases with increasing age of the participant (OR = 1.919, p < 0.01). It was also interesting to find that females were more likely to have diabetes (OR = 1.468, p < 0.001) than males. The occurrence of diabetes was 62% higher for higher educated participants (OR = 1.617, p < 0.001). Compared to the participants from middle income family, the odds of developing diabetes was 42% higher for the individuals who came from the rich family (OR = 1.420, p < 0.01). It was also found that the chance of developing diabetes among overweighed individuals was almost double (OR = 2.121, p < 0.001) than normal weighted people. Moreover, the chance of developing diabetes was higher among persons with hypertension (OR = 1.629, p < 0.001). However, the individuals engaged in physical work had less chance (OR = 0.972, p < 0.05) to develop diabetes compared to their counterpart.

## Discussion

In our study, we observed that the prevalence of DM was more among old age group. Different studies also found that the probability of diabetes was higher with increased age34,35,36,37,38,39,40. This may occur due to deficiency in insulin secretion following weaker pancreatic function of old people41. In addition, the demand of insulin may increase in the human body in certain context or improper utilization of insulin by the body among aged people42,43. The suggestive actions for older adults were maintaining blood pressure, blood sugar through improving lifestyle44,45,46,47,48,49,50,51.

Relation between gender and prevalence of diabetes mellitus showed a mixed result and geographical location has been found as a key determinant. Though we found almost same prevalence of diabetes in both sex, Gale and Gillespi (2001) found that females with non-European origin and males from European origin were more prone to have diabetes52. Hilawe et al. (2013) mentioned that the prevalence of DM was relatively higher in men from Middle or Eastern Africa whereas in women from Southern Africa or among respondents from Indian ethnicity53. Aregbesola et al. (2017) found a higher prevalence of diabetes risk in males54. Another study found that the risk of developing diabetes was higher among Caribbean women55. A study from China also found a higher prevalence of DM among older women compared with men56. In US context, the prevalence of DM was relatively lower among women compared to men between 2013–201657.

Lower educational attainment puts both male and female at a high risk of developing diabetes58,59,60. An inverse association between educational attainment and development of diabetes was observed in different studies61,62,63. A study revealed that low educational level was significantly related to poor glycaemic control for people having Type 2 DM in Bangladesh64. Other studies also demonstrated that higher educational level was independently related to better diabetes knowledge and having optimal glycaemic control through attending diabetes education65,66,67. However, Ali et al. (2019) found a somewhat reverse causality that hypertension, diabetes and comorbidity were higher among higher educated individuals68. Our study also showed similar result.

Prevalence of diabetes was higher amongst higher socio-economic status groups69,70,71,72. Thus, higher wealth is an independent risk factor for diabetes73. The poor socioeconomic status leaves one with a higher risk of remaining undiagnosed for diabetes74. Conversely, the richest have higher probability of being diagnosed with diabetes75. Our study also showed higher risk of developing DM among rich income group that supports other findings as they usually are less involved in physical work in Bangladesh.

DM is usually a hereditary disease. In addition, there are many causal and confounding factors that contributed to develop DM such as obesity. Obesity may result in diabetes which is well documented in different studies76,77,78. Hypertension is common among patients with diabetes and DM helps facilitating hypertension79,80,81,82. Diabetes and high blood pressure complement each other due to the fact that they both contain common physiological traits83. Diabetic patients experience increased peripheral artery resistance causing elevated systemic blood pressure.

Epidemic of diabetes is associated with decreasing levels of physical activity and an increasing prevalence of obesity. Priority should be given to promote physical activity in this regard84,85,86. Physical activity reduces the risk of diabetes which is evident in different studies87,88,89. Moderate to high level of physical activity is associated with substantially lower morbidity and mortality in people with diabetes90. Physical activity can reduce the risk of diabetes complications91. It is important for diabetes management as well92. Regular physical activity (PA) is recommended for diabetic patients93. If needed, counseling on PA by physicians should be considered in this regard94.

## Conclusion

Diabetes has become a major public concern across the world due to its pandemic nature. Bangladesh is not an exception. This study attempted to identify the socio-demographic determinants that condition the prevalence of diabetes mellitus among individuals in Bangladesh. It was observed that age, gender, educational attainment, possession of wealth, obesity, hypertension and level of physical activity were some significant predictors of the prevalence of diabetes mellitus among Bangladeshi adults. This study calls for greater attention of government and other concerned entities to come up with appropriate policy interventions to lower the prevalence of diabetes and associated risks.