Data sources and sampling strategies
Data for this study were extracted from the Bangladesh National Household Survey of Persons with Disabilities (NSPD), 2021. It is a nationally representative cross-sectional survey conducted by the Bangladesh Bureau of Statistics. The study employed a two-stage stratified random sampling method to select respondents. In the first stage of sampling, 800 primary sampling units (PSUs) were randomly selected from a list of 293,579 PSUs. The PSUs were generated by the Bangladesh Bureau of Statistics as part of the 2011 Bangladesh Census and included an average of 120 households19. In the second stage of sampling, 45 households from each previously selected PSU were selected through systematic random sampling method. This resulted in a list of 36,000 households, of which data were collected from 35,493. All respondents who were regular residents of the selected households were included in the study. The study involved 14,659 children aged 0–4 years, 39,513 children aged 5–17 years, and 100,853 adults aged 18–95 years. Details of the survey sampling procedures have been published elsewhere16.
Analysis sample
The survey included individuals aged 2 years and older (excluding those under 2 years old, as identifying the need for assistive devices and detecting disabilities is particularly difficult in this age group in Bangladesh). As intended, we focused on individuals with different types of disabilities. Of all survey participants, a total of 4,270 reported having a disability and were included in the analysis.
Outcome variable
Our main objective was to investigate the occurrence of morbidity (the presence of a disease or health condition) in people with disabilities (functional limitations or restrictions in carrying out activities due to impairments). Relevant data were collected by presenting two independent sets of questions. First, questions aimed at determining the status of disabilities were administered, followed by questions on morbidity. To achieve this, data collectors carefully assessed the status of all household members by asking if all reported family members use assistive devices such as hearing aids to lead a normal life. Next, a set of questions was asked of the identified respondents or their actual caregivers (in the case of children under 18 years of age) to ascertain whether the reported difficulties qualified as disabilities. For this purpose, the Washington Group questions for children and adults were adapted. The selected group of respondents or their actual caregivers were also presented with two additional questions to determine the presence of pre-existing morbidity. First, they were asked: “Do you (or the name of the child under 18 years of age selected for the disability-related questions) have any health or physical problems other than your disability?” Those who answered this first question in the affirmative were then asked: “What health-related or physical problems do you have?” Respondents were provided with a comprehensive list of medical conditions to choose from, such as blood pressure, diabetes, asthma, epilepsy, heart disease, physical/mobility disabilities, and other health and physical issues. They also had the option to indicate if their health condition was not listed. These responses were used to create a simple classification to determine whether people with disabilities had a medical condition or not, which was considered as the outcome variable.
Explanatory variables
Several explanatory variables were included. They were selected in three stages. In the first stage, a list of relevant variables was generated through an exhaustive review of relevant literature9,10,20,21,22 covering LMICs and Bangladesh. Then, the availability of these selected variables was checked in the second stage of the survey. Finally, the available variables were considered in this study and categorized into individual, household and community level factors. Individual level factors included respondent’s age (children aged 2–17 years, adults aged 18–34 years, late adults aged 35–59 years, and elderly aged 60 years and above), gender (male vs. female), respondent’s employment status (agriculture, manual laborer, business, services, student, housewife, unemployed, and other), educational attainment (uneducated, primary education, secondary education and above), religion (Muslim vs. other), and marital status (married, unmarried, widowed/divorced/separated). Household wealth (poorest, poorer, middle, richer, and richest) was considered as a household level factor. Household wealth index was constructed by the survey agency using principal component analysis that considered several variables related to household wealth, such as ownership of a radio and television and type of roof on the house. Place of residence (rural vs urban) and administrative division (Barisal, Chattogram, Dhaka, Khulna, Mymensingh, Rajshahi, Rangpur and Sylhet) were included as community-level factors.
Statistical analysis
Descriptive statistics were used to explore the background characteristics of respondents. Pearson’s chi-square test was used to identify significant differences in the prevalence of disability with individual-, household-, and community-level explanatory variables. A two-level (household, cluster) multilevel logistic regression model was used to explore the association between the prevalence of disability and individual-, household-, and community-level factors. The nested structure of the NSPD data, where individuals are nested within households and households are nested within clusters, necessitated the use of multilevel modeling.23 Previous studies have found that multilevel modeling provides relatively better results than simple logistic regression models for such structured data.24 Two separate models were run for children aged 2–17 years and adults aged 18–95 years. Each model was adjusted for the explanatory variables considered. Multicollinearity was checked before running each model. Sampling weights were taken into account in all analyses. Results were recorded as adjusted odds ratios (aORs) and their 95% confidence intervals (95% CIs). All statistical analyses were performed using Stata software version 14.0 (StataCorp.org, College Station, TX, USA). All methods will be performed in accordance with the guidelines.
Ethical approval
The survey analyzed was conducted by the Bangladesh Bureau of Statistics. We obtained ethical approval from the Bangladesh Medical Research Council, Government of Bangladesh, and our own Internal Review Board before conducting the survey. We obtained informed consent from all respondents before conducting the interviews. In cases where respondents were unable to provide informed consent, we obtained consent from their legally authorized representatives, including the husbands of the women surveyed. Based on our interest in conducting this study, they kindly provided us with de-identified data for this study. As we will only obtain data in de-identified form, no further ethical approval is required.