We analysed data from the most recent Bangladesh Demographic and Health Survey (BDHS) conducted in 2017/18. The survey was conducted by the DHS program of the USA. The National Institute of Population Research and Training of the Ministry of Health and Family Welfare of Bangladesh worked as a local partner. The survey was conducted by following the two-stage stratified random sampling method. At the first stage, 675 primary sampling units (PSUs) were selected randomly from a list of 293,579 PSUs, created as part of the most recent 2011 Bangladesh Population and Housing Census. Of them, the survey was not conducted in three PSUs due to extreme flooding there, leaving a total of 672 PSUs. In the second stage, 20,160 households were selected for data collection with 30 households from each selected PSU. Of them, the interview was completed in 19,457 households with a 96.5% inclusion rate. There were 20,376 eligible women in these selected households, and data were collected from 20,127 women. Informed consent was obtained from all participants. The contraception data was collected from each of these women. However, diabetes and hypertension-related data were collected from one-fourth of the selected households (7 to 8 households per PSU), which generated 4,864 households. Unlike contraception data, diabetes and hypertension data were collected from both males and females in the selected households aged 18 years and more. There were 14,704 respondents in these selected households, 12,924 of them (men: 5,583; women: 7,341) had blood pressure measured, and blood glucose tested. Details of the sampling procedure have been published elsewhere .
Data of 3,947 women who met the following inclusion criteria were analysed in this study. The inclusion criteria were: (i) fertile women who were not pregnant or not in the post-partum amenorrhea period at the time of data collection and (ii) reported their blood pressure and blood glucose or their medication status to control blood pressure and/or blood glucose.
Based on women’s contraception using status (yes, no) and the type of methods they used (e.g., pill, condom, sterilization), we generated three different outcome measures: (i) no contraception use vs any contraception use (modern or traditional method), (ii) traditional method or no use vs modern method use, (iii) traditional method vs modern method use.
Explanatory variables and confounders
Women’s diabetes and hypertension status were our main explanatory variables classified as none, diabetes only, hypertension only, and both diabetes and hypertension. Confounders were considered at the individual, household, and community-level factors based on previous relevant literature [2-4, 9, 11, 17]. Individual-level factors were women's age (≤19 years, 20-34 years, ≥35 years), education (no education, primary education, secondary education, higher education) and body mass index (BMI: underweight, normal weight, overweight, obese). We followed the World Health Organization (WHO) recommendation for the Asian population body mass index to classify women's BMI in this study . Husband’s education (no formal education, primary education, secondary education, higher education), number of children ever born (no child, 1-2 children, >2 children) and wealth quintile (poorest, poorer, middle, richer, richest) were the household level factors. The BDHS created the wealth quintile variable using the principal component analysis of women’s responses on the households’ durables goods such as materials to build houses, household ownership of radio/television. Women’s place of residence (urban, rural) and administrative divisions (Barishal, Chattogram, Dhaka, Khulna, Mymensingh, Rajshahi, Rangpur, Sylhet) were community-level factors.
We reported frequency and percentage to describe the characteristics of the respondents. The prevalence of contraception in general and across diabetes and hypertension status of the women were calculated and reported with 95% confidence intervals (95% CI). We used multilevel mixed-effect Poisson regression with robust variance to determine the associations between the outcome and explanatory variables. We used Poisson regression because the odds ratio using logistic regression in cross-sectional studies may significantly overestimate the effect size when the outcomes are common (e.g., prevalence >10%) [19, 20]. Secondly, in the dataset, individuals were nested within households; and households were nested within PSUs. This nested data structure generated multiple hierarchies and dependencies. The multilevel mixed-effects Poisson regression accounts for these multiple hierarchies and dependencies in data and the problem of overestimation . We developed unadjusted and adjusted models (adjusted for individual, household, and community level confounders) for each of the three study outcomes. For instance, the first set of models examined the associations between women’s status of chronic conditions and contraception use (no use vs any methods use). The second set examined the associations between women’s status of chronic conditions and contraception use type (traditional methods/no use vs modern methods use). The third set examined the association with contraception method type (traditional methods vs modern methods use). Sampling weight and complex survey design were considered in all analyses. Results are reported as prevalence ratio (PR) with its corresponding 95% CI. Multicollinearity and interactions were checked. The study was designed and reported following the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines . All methods were performed in accordance with the relevant guidelines and regulations. All analyses were performed using the statistical software package Stata (version 15.1; Stata Corp LP, College Station, TX, USA).