Non-communicable diseases (NCDs)—including cardiovascular diseases, diabetes, cancer and chronic respiratory diseases—are the leading causes of premature death in low-income and middle-income countries (LMICs).1 2 In particular, South Asia has recorded a higher number of life-years lost due to premature NCDs than any other global region, a situation which reflects both the region’s large population and the relatively young age at which NCD deaths tend to occur in this region.1 3 Furthermore, while NCD mortality rates have decreased during recent decades in most countries, they remain at a high level in the South East Asia region.4 5 Better understanding of the multidimensional determinants of NCDs in South Asia should, therefore, inform the development of appropriately tailored strategies for disease prevention and control.

There is, however, limited evidence available on NCD determinants for many South Asian populations.2 6 For example, Bangladesh, a country with 170 million people, is one of the least studied most densely-populated countries with regard to NCDs, despite steep and sustained increases in NCDs risk factors and incidence during recent decades.7 8 In 2021, 74% of all adult deaths in Bangladesh were attributed to NCDs.7 9 10 The high burden of NCDs in Bangladesh is not just of local public health concern, as mortality due to NCDs has been reported to be more than two times higher among Bangladeshis living in western regions compared with host populations.11 An important challenge is, therefore, to establish informative epidemiological resources in a rigorous manner to evaluate risk factors among Bangladeshis.

The present report provides a description of the methods used in the establishment of the BangladEsh Longitudinal Investigation of Emerging Vascular and nonvascular Events (BELIEVE) study. It also describes the baseline characteristics of the study population recruited so far and outlines the rationale for the study’s further development.

Cohort descriptionStudy design and participants

Between January 2016 and March 2020, the BELIEVE prospective cohort study recruited 73 883 participants from 30 817 households across three different settings in Bangladesh: urban (Mirpur-Dhaka), rural (Matlab-Chandpur) and urban-poor (‘slum’; Bauniabadh-Dhaka) (figure 1). Households were initially identified through complete surveys that had previously mapped households’ information across the three study sites. Individuals living in the identified households were eligible for enrolment into the study if they (1) were aged 11 years or older; (2) had lived in their current household for at least 3 years; (3) were intending to reside in the study site for at least a further 5 years and (4) were willing to provide written informed consent and prospective follow-up information. For all participants, the recruitment procedure first involved trained study personnel visiting all identified households to provide written and verbal information about the study. On completion of the household visit, all eligible household members were invited to attend a local study clinic to complete an individual-level baseline assessment (figure 2).

Figure 1

Location of the BELIEVE study sites. BELIEVE, BangladEsh Longitudinal Investigation of Emerging Vascular and nonvascular Events.

Figure 2

Study recruitment and follow-up procedures.

Questionnaire administration and physical measurements at baseline assessment

The study questionnaire was adapted to the Bangladesh context based on validated questionnaires used in previous large-scale studies.12–14 The initial version of the structured questionnaire underwent preliminary testing with a small group of individuals from the target population to evaluate clarity, relevance and cultural appropriateness. Following this, a larger group of individuals was involved for further validation and refinements, and adjustments were made to enhance readability and ensure logical flow, before finalising the questionnaire for the main study. To ensure both linguistic accuracy and cultural relevance, the questionnaire was translated into Bengali using a rigorous translation and back-translation process. During the household visit, trained personnel administered the structured questionnaire to collect household-level information and family structure characteristics (table 1) using a bespoke Android interface operated through handheld touchscreen tablet devices. During the clinic visit, trained personnel used a computer-based prepiloted epidemiological questionnaire to collect self-reported information on >300 items related to socioeconomic and demographic characteristics, consanguinity, behaviours (eg, tobacco and alcohol consumption, dietary intake, physical activity), female health, self-reported personal and family medical history and medication use (table 1; a copy of the study questionnaire is available in online supplemental appendix). Direct computer entry by participants, rather than interviews, was employed to enhance privacy when answering sensitive questions, with the option to skip these questions.

Table 1

Questionnaire-based information, physical measurements and biological samples collected at baseline in the BELIEVE study

Self-reported information on medical history and medication use was supplemented by medical records, drug prescriptions and by asking participants to bring their medications to the clinic visit. To assess local dietary patterns, benchmark food-frequency questionnaires previously developed and validated in Bangladesh were adapted for BELIEVE.15 These questionnaires estimated the standard portion size assigned to each food item and drinking water source. Participants’ phone numbers (or, for participants younger than 18 years, a contact number of a legal parent, guardian or caretaker) were routinely collected at the baseline visit for future contact and to collect follow-up information.

Standardised procedures and equipment were used to assess height, weight, waist and hip circumference, body composition, systolic and diastolic blood pressure, heart rate and upper body strength (online supplemental table 1). Briefly, blood pressure was measured using an automated device (Omron HEM 7130) on the right arm twice with a 3 min interval between each measurement, with the patient remaining in a sitting position for at least 5 minutes before the first evaluation. Height was measured using the ShorrBoard ICA Measuring Board to within 1 cm and weight was measured using the Tanita HD-661 scale to within 0.1 kg. Waist circumference was assessed using a non-stretchable soft standard measuring tape over the abdomen at the widest diameter between the costal margin and the iliac crest, and hip circumference at the level of the greater trochanters (ie, the widest diameter around the buttocks). Anthropometric measurements were performed in a standing position. Body composition was assessed by bioimpedance using the Tanita MC-780MA analyser and handgrip strength using a Jamar Plus Digital Hand Dynamometer.

To enhance consistency in the collection of data, we trained staff extensively and adopted standardised approaches, validated instruments and electronic data collection methods with built-in validity checks and queries. For example, the paper-free digital data collection platform involved extensive computerised checks to restrict missing values, duplications, inconsistencies and outliers. Information was transferred daily in a secure manner to the study’s central database at Cambridge University, with a copy also kept at the local recruitment centres for additional checks and queries. Reports were generated by data managers and reviewed by study Principal Investigators (PIs) and coordinators on a regular basis.

Collection, storage and initial use of biological samples

Participants provided non-fasting blood samples for research with the time of the last meal recorded. Up to 23 mL of non-fasted whole blood samples were collected from each participant aged 18 years and above in two tubes (11 mL EDTA and 12 mL serum). For participants aged 11–17 years, 12 mL of non-fasted whole blood samples were collected in two tubes (6 mL EDTA and 6 mL serum). Samples were centrifuged within 45 min of venipuncture (at 6000 rpm for 15 min). Isolated serum, EDTA plasma and whole blood samples were stored at −20°C before being transferred to a −86°C freezer at the end of each working day (online supplemental figure 1). To enable the measurement of additional potential risk factors (eg, metal contaminants), finger-nail and toe-nail clippings were collected in a subset of participants, kept separately in plastic bottles and stored at room temperature. Biological samples have been stored in long-term dual repositories located in Bangladesh and the UK to enable further assays. A range of biochemical and genomic analyses on the stored samples is currently ongoing.

Follow-up procedure and outcome ascertainment

Study participants are being actively followed up indefinitely for cause-specific mortality and incidence of selected non-fatal health outcomes (figure 2). Follow-up involves contact with participants by the study team at regular intervals (eg, every 18–24 months), during which trained personnel use an electronic questionnaire to collect information on medical events reported since the last contact. Based on the responses to the questionnaire (or if phone contact is unsuccessful), a household visit is scheduled to collect additional information and medical documents. During the household visit, trained personnel collect further information for selected major categories of non-fatal events (including myocardial infarction, diabetes, stroke and common cancers) using a structured questionnaire and a bespoke Android electronic interface operated through handheld touchscreen tablet devices.

As the majority of individuals in the study settings who have received medical attention tend to have paper copies of medical documents in their homes, study personnel visiting households collect digital photographs of any relevant medical documents available (including discharge summaries, diagnostic test reports, medications and death certificates) using tablet devices. If medical documents are unavailable, written informed consent is obtained to retrieve these from the relevant hospitals. Nearly all deaths in the study areas will have involved some form of medical attention, with their underlying causes certified by a medical doctor. Verbal autopsy is conducted in all deaths by trained personnel to help determine the most likely cause from symptoms or signs described by family members.16 To help adjudicate and classify health outcomes (eg, into definite, probable and possible categories), the information collected described above is periodically reviewed by medically trained personnel.

The quality and completeness of both mortality and morbidity follow-up data in each study site are regularly checked during the study period by the study coordinating centres. This involves monitoring the number of people who have died or are lost to follow-up, assessment of overall mortality patterns in the study cohort, levels of diagnostic criteria for individual diseases and the proportion of deaths with unknown causes.

Patient and public involvement

In keeping with this study’s principles of cocreation and community engagement, we have implemented a participatory strategy, collaborating closely with local residents and key stakeholders (some of whom were study participants). Community consultations were conducted in communal spaces to communicate the study’s objectives and actively seek feedback from community members. Invitations were extended to local leaders for these sessions, facilitating a deeper understanding of the community’s needs and concerns and allowing us to tailor the study accordingly. In addition, we employed local community members as field team personnel, who were trained to collect data and assist study participants. This approach not only provided employment opportunities for local people but also enhanced the cultural appropriateness of the study approaches we used. As well as cocreating the study design, study participants were provided information about the study’s purpose, with the assurance that their information would be exclusively used for research purposes.

Sample size considerations and statistical analysis

Sample size considerations have been guided by a combination of pragmatic constraints (eg, the availability of resources) and statistical power calculations. Assuming 5% of the population develops a disease condition during follow-up, a study of 70 000 individuals should provide 80% power to detect an HR of about 1.1 per 1 SD higher value at 5% significance level. In this report, continuous variables were summarised as mean (SD) or median and IQR, and categorical variables were summarised as frequencies and percentages. A complete case analysis was used to handle missing data. Variables were summarised and compared across relevant population subgroups (eg, study site, sex and age group) using the t-test for continuous variables and χ2 test for categorical variables. The prevalences of chronic conditions at baseline were summarised as overall crude prevalences, and by site, sex and 20-year age group. To make comparisons between sites analysis was conducted restricted to data from participants aged between 20 and 79 years to compare prevalences of conditions standardised according to the Bangladesh 2022 population structure.17 All statistical tests were two sided, and the significance level was set at p<0.05. Robust SEs were estimated to allow for the clustering of participants by household. Analyses were performed using STATA (V.16, StataCorp) and R (V.4.0.1, R Foundation for Statistical Computing). Future statistical analyses will be developed following relevant guidelines for cohort data with household clustering (eg, Strengthening the Reporting of Observational Studies)18 and will be presented elsewhere.

Findings to dateDemographic, behavioural and physical characteristics in the overall study population

Overall, 59 846 (81%) participants were recruited from Mirpur-Dhaka (urban site), 5332 (7%) from Bauniabadh-Dhaka (slum site) and 8705 (12%) from Matlab-Chandpur (rural site). 43 470 (59%) participants were female; the mean (SD) age at recruitment was 39 (15) years; 4122 participants (5.6%) were aged between 11 and 17 years inclusive (table 2). At the time of the baseline survey, 38 848 (52%) participants had either no formal education or primary level education only; 17% were current smokers; 23% were current users of chewable tobacco products and only 1.3% of the participants reported consuming alcohol regularly.

Table 2

Baseline characteristics of the 73 833 BELIEVE study participants

Several characteristics varied by study site. For example, 55% of participants in the urban site had a secondary or higher level of education compared with only 16% and 18% in slum and rural sites, respectively (online supplemental table 2). Whereas the percentages of participants self-reporting as current smokers were similar across the sites (17%, 17% and 13% in urban, slum and rural sites, respectively), reported use of chewable tobacco and alcohol was higher in the slum site (62% and 12%, respectively, vs 18% and 0.4% in the urban site, and 34% and 0.4% in the rural site). Additional conventional risk factor values differed by study site, with a tendency towards more adverse risk factor profiles in urban participants (online supplemental table 2). For example, mean (SD) systolic blood pressure was 125 (21) mm Hg in participants from the urban site vs 112 (20) and 120 (20) mm Hg in the slum and rural sites, respectively. Mean (SD) body mass index (BMI) was 26 (5) kg/m2, 23 (5) kg/m2 and 23 (5) kg/m2 in participants from the urban, slum and rural sites, respectively, with other measures of adiposity (including, waist-to-hip ratio and body fat percentage) following a similar trend. As the trend towards urbanisation accelerates in Bangladesh, such differences in risk factor profiles portend implications for striking increases in NCDs.19 20 This contrast in risk factor profiles across settings also suggests the need for context-specific solutions to help better prevent and control NCDs. We plan to contribute towards this goal through longitudinal surveillance of risk factors and NCD incidence in the BELIEVE cohort, as well as the repurposing of the cohort framework to support applied health research studies.

Certain baseline characteristics also differed between males and females (online supplemental table 3). Females were, on average, younger (38 vs 41 years), had spent fewer years in education (45% vs 50% had a secondary or higher level of education), were less likely to report any tobacco smoking (<1% vs 41%) and consuming alcohol (<1% vs 3%). Systolic blood pressure was higher among males (127 vs 121 mm Hg), whereas females had a higher BMI (26 vs 24 kg/m2).

Overweight, obesity and chronic health conditions in adults aged 20–79

Among adult participants aged between 20 and 79 years (online supplemental table 4–6) and using WHO-defined BMI categories, 39% of individuals were considered overweight (BMI 25–29 kg/m2) and 16% obese (ie, BMI≥30 kg/m2online supplemental table 4). After standardisation of the sex and age distribution of the Bangladeshi population for the same age range, the corresponding proportions were 37% and 14%, respectively (online supplemental figure 2). Proportions of overweight and obese participants were higher in the urban site, compared with the slum and rural sites, and also higher in females compared with males (online supplemental figure 2–3). There are likely to be multiple factors contributing to excess adiposity, including dietary patterns, sedentary lifestyles and urbanisation.1 21 Additionally, socioeconomic disparities and lack of awareness regarding the importance of diet and physical activity may play roles. As overweight and obesity are associated with multiple chronic health conditions, including cardiovascular diseases, diabetes and certain cancers,22–24 addressing this challenge is necessary to improve the health of the Bangladeshi population and alleviate healthcare system burdens.25 26

The most prevalent self-reported chronic health conditions recorded at baseline among adults aged 20–79 were hypertension, type 2 diabetes and hypercholesterolaemia (online supplemental table 6), with standardised prevalences of 21%, 15% and 11%, respectively (figure 3). The prevalence of these conditions was higher among participants recruited in the urban site compared with those in slum and rural sites. For example, the standardised prevalence of hypertension was 23%, 16% and 14% in urban, slum and rural sites, respectively. Corresponding prevalences were 16%, 10% and 5.7% for type 2 diabetes, and 12%, 8% and 1.5% for hypercholesterolaemia. The overall prevalence of type 2 diabetes in this study was higher than those previously reported for Bangladesh19 and varied by age and sex, with the highest prevalences among older participants and females (online supplemental figure 4–6). These findings may suggest potential age-specific and sex-specific differences in susceptibility to type 2 diabetes or disparities in healthcare access and management or both. We plan to contribute towards understanding such potential age-specific and sex-specific differences in determinants of chronic diseases, as it should inform the development of targeted preventive measures and tailored interventions to address the specific risk factors affecting different populations.

Figure 3

Standardised prevalences of selected chronic conditions at recruitment in BELIEVE for participants aged 20–79. Prevalence was standardised using the sex and age distribution of the Bangladesh 2022 population aged 20–79. BELIEVE, BangladEsh Longitudinal Investigation of Emerging Vascular and nonvascular Events.

Strengths and limitations

Although Bangladesh is experiencing substantial increases in NCDs,7 there is limited evidence about the multidimensional risk factors and drivers of NCDs in the country, thereby preventing a deeper understanding of the web of causation for NCDs and limiting development of optimal prevention and control approaches. The BELIEVE study has been established as a long-term epidemiological bioresource to help address this gap, including an investigation of Bangladesh’s contrasting urban and rural settings. Configured to investigate biological, behavioural, environmental and broader determinants of NCDs, the BELIEVE study is a household-based prospective cohort study that spans urban, slum and rural settings. To our knowledge, it represents the largest bioresource for NCDs and related traits in Bangladesh, involving active longitudinal follow-up of participants for new-onset health outcomes.

The BELIEVE study has demonstrated the feasibility of employing modern epidemiological methods at scale in a population-based study located across multiple different settings in Bangladesh. The study has used electronic data collection methods, using study forms implemented through bespoke software applications (‘apps’), allowing comprehensive recording of questionnaire-based data, as well as physical measurements. The study has also collected a range of biological samples, including serum, plasma, whole blood and nail samples stored in long-term repositories. The assay of these samples (which already includes genotyping, sequencing, multiple molecular ‘omics, and soluble clinical biomarkers) enables the study of many molecular factors, laying the foundations for an advanced understanding of disease pathways and potential therapeutic targets for the treatment and prevention of NCDs in Bangladesh and beyond.

The potential limitations of the BELIEVE study merit consideration. Within each of the three study settings, every household was invited to participate in the cohort. As the household participation rates have been high (>80%), they suggest that study participants should be broadly representative of the source population in the participating sites. However, the participants, households and sites included in the study were not necessarily representative of the general Bangladesh population. Nevertheless, it should be reasonable to infer that findings from this cohort can be generalised to other similar settings in Bangladesh because the study involves participants with a wide range of diverse characteristics (eg, in relation to age, sex, socioeconomic status, education level and occupation), recruited from a variety of different settings (urban, slum and rural areas) characteristic of contemporary Bangladesh.

A further potential limitation relates to the scope for misclassification of risk factors and health outcomes recorded at baseline and during study follow-up because of inaccuracies in participant self-reports. To help limit the effects of such potential misclassification, the BELIEVE study is supplementing self-reported data with the use of objective measurements (eg, assay of low-density lipoprotein -cholesterol (LDL-C), glycated haemoglobin (HbA1c) and arsenic metabolites), inspection of health records kept by participants during household follow-up visits, use of previously validated ‘verbal autopsy’ methods, and exploration of emerging potential linkages of study participants with digital health records kept at community healthcare and hospital levels.

Finally, there is potential in this prospective study for underestimation of the strength of the associations observed between risk factors and incident health outcomes due to fluctuations in risk factor levels within individuals over time (ie, ‘regression dilution’ bias). To help limit such bias, the BELIEVE study plans to conduct periodic resurveys of risk factor levels in randomly selected subsets of the study participants.