Prevalence of diabetes and prediabetes among Bangladeshi adults and associated factors: Evidence from the Demographic and Health Survey, 2017-18

Aim: To estimate the age-standardized prevalence of diabetes and prediabetes, and to identify factors associated with these conditions at individual, household, and community levels. Methods: Data from 11, 952 Bangladeshi adults aged 18+ available from the most recent nationally representative Bangladesh Demographic and Health Survey 2017-18 were used. Anthropometric measurements and fasting blood glucose samples were taken as part of the survey. Prevalence estimates of diabetes and prediabetes were age-standardised, and risk factors were identified using multilevel mix-effect Poisson regression models with robust variance. Results: The overall age-standardized prevalence of diabetes was 12.8% (95%CI 11.2-14.3) (men: 12.8%, women: 12.7%), and prediabetes was 14.0% (95%CI 12.6-15.4) (men: 12.1%, women: 16.5%). Among people with diabetes, 61.5% were unaware that they had the condition, and only 35.2% were receiving treatment regularly. Factors associated with an increased risk of having diabetes were increasing age, male, overweight/obesity, hypertension, highest wealth quintile and living in Dhaka division. Individuals being underweight or currently working were less likely to develop diabetes. Conclusion Diabetes and prediabetes affect a substantial proportion (over one-quarter) of the Bangladeshi adult population. Continuing surveillance and effective prevention and control measures, with a particular focus on obesity reduction and hypertension management, are urgently needed. Keywords: Diabetes, prediabetes, prevalence, Bangladesh


Introduction
Diabetes mellitus remains a significant contributor to the global burden of disease [1]. People with diabetes have an increased risk of developing a number of serious life-threatening micro and macro vascular complications resulting in higher medical care costs, reduced quality of life and increased mortality [2]. The International Diabetes Federation (IDF) has estimated that 463 million adults live with diabetes worldwide in 2019, with a projected increase to 700 million by 2045 [3]. Seventy-nine per cent of those with diabetes live in low-and middleincome countries (LMICs) [4]. It is projected that diabetes cases will increase by 74% in Southeast Asian countries in the next two decades, from 88 million in 2019 to 153 million by 2045 [4].
In Bangladesh, there were 8.4 million adults living with diabetes in 2019, and projected to almost double (15.0 million) by 2045 [4]. It is also estimated that another 3.8 million people had prediabetes in Bangladesh in 2019 [4]. Studies, including a systematic review and metaanalysis, and national survey reports showed that the prevalence of diabetes among adults has increased substantially in Bangladesh, from ~5% in 2001 to ~14% in 2017 [5][6][7][8].
The economic costs of diabetes are high due to the treatment of complications associated with the disease. Annual global health expenditure on diabetes was estimated to be USD 760 billion in 2019 and is projected to reach USD 845 billion by 2045 [3]. In Bangladesh, these costs have been estimated to US$1.5 billion annually [9]. The true costs may be even higher when economic productivity loss, rather than just treatment expenditure, are included. expenditure is spent in these settings [4]. Health expenditure attributable to diabetes will increase in the 21 st century if no effective preventive measures are adopted [10].
It is thought that recent increases in time-variant risk factors for diabetes and prediabetes, including body mass index (BMI) and hypertension, account for the increasing prevalence in Bangladesh [6,7,11]. Overweight/obesity has consistently been associated with diabetes and prediabetes [12], but is modifiable through diabetes prevention lifestyle interventions [13].
Hypertension is an independent predictor of incident diabetes [14] and shares common aetiology and metabolic pathways with diabetes [15]. In countries with an ageing population and increasing urbanization, the overall burden of blood pressure-related complications such as nephropathy, neuropathy, and cardiovascular events is rising [16] To our knowledge, the age-standardized prevalence of, and risk factors for, diabetes and prediabetes in Bangladesh using the latest Bangladesh Demographic and Health Survey (BDHS) 2017-18 data have not yet been estimated. We, therefore, aimed to estimate the agestandardized prevalence of diabetes, and prediabetes in Bangladeshi adults aged 18 years and older using the latest BDHS. We also investigated factors associated with diabetes and prediabetes in Bangladeshi adults. Results are examined in detail according to individual, household and community-level characteristics. and construction materials) [20]. Community-level factors included were the place of residence and administrative divisions of the country.

Statistical analysis
The crude prevalence of diabetes and prediabetes were estimated by adjusting survey sampling weight and complex survey design. We used a multilevel Poisson regression model with a robust variance to identify factors associated with diabetes and prediabetes, and results were presented as prevalence ratio (PR) with 95% CIs. We used this model, firstly, the odds ratio estimated using logistic regression from a cross-sectional study may significantly overestimate relative risk when the outcome is common [21,22]. Secondly, in the BDHS, individuals were nested within the household; households were nested within the PSU/cluster. Our multilevel mixed-effects Poisson regression model account for these multiple hierarchies and dependency in data, and the problem of overestimation [23]. Four models were run separately for diabetes and prediabetes for different sets of confounding factors at the individual, household, and community levels with progressive model-building technique. Model 1 was run without confounding factors to determine the cluster level variation of diabetes and pre-diabetes in Bangladesh. Model 2 and 3 were adjusted for individual, and individual plus household level All rights reserved. No reuse allowed without permission. perpetuity.
preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in

Results
Of 12,100 weighted sample who provided FBG , 11,952 were included in the analysis ( Figure   1). The median (Q1-Q3) age of the participants was 36 (26-50) years ( preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for this this version posted January 26, 2021. ; perpetuity.
The age-standardized prevalence estimates of prediabetes were also higher in people those were obese, in the highest wealth quintile, living in urban areas, and in the Dhaka division (Table 2). Three out of five (61.5%, 95%CI 57.9-64.9) people living with diabetes did not know that they had it and only a third (35.2%, 95%CI 32.0-38.5) were receiving appropriate treatment.
All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for this this version posted January 26, 2021. ; Diabetes is defined as a fasting blood glucose level ≥7.0 mmol/L or self-reported diabetes medication use. Prediabetes is defined as a fasting blood glucose level from 6.1 mmol/L to 6.9 mmol/L, without medication.
All rights reserved. No reuse allowed without permission. perpetuity.
preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in In each of the four mixed-effects multilevel Poisson models to identify factors associated with diabetes and pre-diabetes, we compared intra-class correlation (ICC), Akaike's information criterion (AIC), and Bayesian information criterion (BIC) to select the best fitting model. The preferred model was the one that had the smallest ICC, AIC and BIC. According to these indicators, Model 4 (including individual, household and community-level factors) had the best model fit. The first (the null model), which included none of the covariates, produced an ICC of 30.52% and 42.54% variance in diabetes and prediabetes, respectively ( preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for this this version posted January 26, 2021. ;   perpetuity.
preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in

Summary of key findings
Diabetes and prediabetes affect a substantial proportion of the Bangladeshi population.
Based on data from the latest BDHS 2017-18, over one-quarter of individuals aged 18 years and older had diabetes or prediabetes in Bangladesh, representing in excess of 25 million individuals in 2020. Factors associated with diabetes were age, sex, BMI, wealth quintile, employment status, hypertension and administrative division of the country but not the place of residence (urban /rural) or level of education. They confirm a continuing high burden of diabetes and prediabetes in Bangladesh.

Comparison with existing literature
The prevalence of diabetes we have found is higher than the overall age-adjusted diabetes prevalence of 11.3% in the Southeast Asian region, estimated by the IDF in 2019 [4]. The IDF has identified that the countries with the largest numbers of adults with diabetes aged 20-79 years in 2019 in the region are China (116 million cases) and India (77 million cases) [4]. In 2019, the IDF ranked Bangladesh 10 th of countries with the highest number of adults (20-79 years) with diabetes (8.4 million cases), and it is expected to be ranked 9 th in 2030 and 2045 [4]. Our data confirm that cases in Bangladesh are very high. In our analysis, over 1 in 10 adults (18+) had diabetes, representing an estimated 14 million individuals in 2020. This large number of cases in Bangladesh indicate that it is one of the leading countries for diabetes burden in Southeast Asian region and underlines the urgent need for policies supporting the rollout of diabetes prevention in this country.
All rights reserved. No reuse allowed without permission. perpetuity.
preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for this this version posted January 26, 2021. ; https://doi.org/10.1101/2021.01.26.21250519 doi: medRxiv preprint Our study undertook an identical methodological approach, (e.g. anthropometric measurements and fasting blood samples), to the 2011 study and demonstrated that the prevalence of diabetes has increased markedly in Bangladesh over 7 years [5]. Similar increasing trends of diabetes have been observed in other Southeast Asian countries [25].
However, the extent to which changes in traditional diabetes risk factors can explain the increasing trends in prevalence of diabetes in this setting needs further investigation. A higher diabetes prevalence suggests that despite greater global awareness of diabetes and interventions for improved non-communicable disease management in primary health care [26], diabetes in Bangladesh is increasing. It suggests that health promotion may be failing in the face of dietary and lifestyles patterns that result in diabetes. Thus, more resources are needed to be invested in primary health care to address the prevention of diabetes in Bangladesh.
Our estimates suggest that the prevalence of prediabetes has decreased in Bangladesh in the 7 years between 2011 and 2017/18 [5]. Prediabetes is important as it is a time when microvascular complications are occurring, often without people knowing they are glucose intolerant. The literature shows that up to 40.5% of individual with prediabetes convert to diabetes during follow-up [27]. A high conversion rate of prediabetes to diabetes is indicative of the potential for an uncontrolled increase in prevalence of diabetes. The observation that diabetes prevalence has increased but prediabetes has decreased in 7 years may indicate higher than expected conversion rates possibly due to rapidly changing environmental conditions [27]. Changes in insulin resistance or beta cell activity (or both) may contribute that need further investigation in the Bangladeshi population. Further declines in prediabetes may be achievable as it is a reversible condition through lifestyle modification [28]. The policy All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for this this version posted January 26, 2021. ; implication is that diabetes prevention programmes need to continue targeting those with prediabetes to help delay this preventable disease early in its pathogenesis, reducing disease burden [29].
There was a positive association between increasing age and risk of diabetes in Bangladesh, consistent with well-established relationships in the literature [4,30]. The implication of this is that with increasing life expectancy in Bangladesh (current life expectancy at birth 72.3 years ) [31] the increasing numbers of older people will result in even more cases and a higher burden of diabetes.
Consistent with our study, world literature shows that obesity is a leading risk factor for type 2 diabetes [32]. However, the association between obesity and diabetes is complicated as obesity is also related to socioeconomic status. We observed a significantly higher prevalence of diabetes in the highest wealth quintile than the lowest wealth quintile. A possible explanation is that those in the highest wealth quintile in developing economies use disposable income to purchase western, high energy food ('nutrition transition') and avoid physically demanding tasks as symbols of status [33]. This results in obesity which in turn is associated with diabetes. Increasing obesity in Bangladesh [11] may also be due to reduced physical activity associated with changing traditional agricultural/domestic works replaced by technology, watching television and using internet. Conversely, in high-income countries, high socioeconomic status is not associated with obesity [34]. A reason may be that in higherincome countries, high socioeconomic status individuals use their resources to eat healthy diets and take regular exercise (e.g. personal trainers). Irrespective of the explanation, increasing obesity in Bangladesh suggests that diabetes will increase further with its All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for this this version posted January 26, 2021. ; https://doi.org/10.1101/2021.01.26.21250519 doi: medRxiv preprint strengthening economy. A further policy implication is that interventions for diabetes prevention in Bangladesh need to focus on obesity in particular, by reducing the consumption of unhealthy diets and increasing physical activity. This should be given priority as there is evidence that even modest weight reduction of 5-7% in high risk individuals result a declines in the incidence of diabetes [28].
In Bangladesh, adults' fat distribution rather than BMI per se, may result in a pattern of metabolic abnormalities associated diabetes. Central obesity and insulin resistance are higher in the people from South Asian (India, Pakistan, Bangladesh) origin than other ethnic groups of similar BMI [35]. Further, the risk may accumulate over the life-course as these patterns have been found to be in South Asians from birth [36]. Recent studies suggest that South Asians not only have high rates of insulin resistance, but an early decline in β-cell function as well a phenomenon with genetic underpinnings [34,37]. These observations need further investigation in the Bangladeshi population to see if they contribute to the substantial diabetes prevalence we have observed in 2017-18, in addition to traditional risk factors.
Our data also show that the prevalence of diabetes is higher in people with hypertension, which is in agreement with other studies [38,39]. Hypertension is exacerbated by other risk factors such as obesity, advanced age and significantly contributes to micro and macrovascular complications resulting in renal failure [40] and cardiovascular disease [38].
Pathways through which these complications may occur include insulin resistance, inflammation and obesity [15]. The implication of the strong association of hypertension and diabetes is that efforts are needed in Bangladesh to delay or prevent comorbid hypertension in diabetes through frequent follow-up and aggressive management. All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in Three out of five people living with diabetes in Bangladesh did not know that they had it, and only a third were receiving appropriate treatment. Renewed efforts are needed to increase awareness, treatment, and control to improve diabetes outcomes. Obesogenic environments such as walkable neighbourhoods and encouraging healthy food choices in schools and cafeterias should be developed as interventions to address non-communicable diseases, including diabetes, in Bangladesh. However, there is evidence that the Bangladesh health care system, which is orientated to communicable diseases [41], is struggling to address noncommunicable disease prevention adequately [42]. This is due to lifestyle changes resulting from the epidemiological transition and rapid urbanisation [42]. The Bangladesh government is making progress in achieving universal health coverage; however, barriers include large outof-pocket expenditure [43]. It places a financial burden on households, and has the effect of preventing people from accessing care (health care is viewed as a 'luxury' not a 'necessity') or seeking alternative providers who are cheaper, but untrained and cause adverse effects [44].
These barriers may undermine diabetes prevention efforts. However, future health promotion policy needs to prioritise affordable, high quality public health programmes for the prevention and management of diabetes.

Methodological considerations
The strengths of the study are that it used a large, nationally representative dataset suggesting the findings have external validity. A further strength of the study is that clinical variables including FBG, blood pressure and body weight and height were measured using high-quality techniques. The WHO criteria for the classification of diabetes and prediabetes were used while hypertension was defined using the seventh report of the Joint National Committee on Prevention, Detection, Evaluation, and Treatment of high blood pressure All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for this this version posted January 26, 2021. ; https://doi.org/10.1101/2021.01.26.21250519 doi: medRxiv preprint criteria. Our multilevel mixed-effects Poisson regression corrects the problem of overestimation of effects size produced by conventional logistic regression employed in crosssectional studies, and increased the precision of the findings. However, this was a crosssectional study, which limits our ability to infer causal relationships among the risk factors. A further limitation was that diabetes medication use was self-reported. Dietary and physical activity data were not collected and so could not be controlled for in the analysis. In future, collection of diet and physical activity data, and measures of central obesity, insulin resistance and beta-cell function will help to elucidate causal mechanisms accounting for the increasing prevalence of diabetes, but declining prevalence of prediabetes, in Bangladesh.

Implications of findings
This study implies that efforts to control diabetes and prediabetes in Bangladesh need to be strenghten. Further investment of resources is needed to manage these conditions. Diabetes and prediabetes are preventable diseases through modification of diet and physical activities.
Given that these are preventable diseases, Bangladesh needs to redouble efforts to implement diabetes prevention. This may require the health care system changes in which non-communicable disease prevention is prioritized and household medical care payments reviewed to reduce out-of-pocket expenses. These measures will be worth the investment as they will maximize access to high-quality public health programmes. A further implication of our analysis is that diabetes prevention should focus of reducing obesity and managing hypertension, particularly in men and those with disposable income. These factors were strongly associated with diabetes, suggesting that their management will bring the greatest benefits. Without effective preventive measures, diabetes will continue to increase in Bangladesh.
All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for this this version posted January 26, 2021. ; https://doi.org/10.1101/2021.01.26.21250519 doi: medRxiv preprint

Conclusions
Diabetes and prediabetes affect a substantial proportion (over one-quarter) of Bangladeshi adult population. These conditions remain a significant public health burden in Bangladesh.
Continuing surveillance and effective prevention and control measures, with a particular focus on obesity reduction and hypertension management, are urgently needed.