Study Design & Sampling
A population-based cross-sectional telephone survey was conducted between September and November 2019 to recruit a representative sample of adult males and females in North and South Dhaka City Corporations (collectively known as DCC), in Dhaka, the capital of Bangladesh. The sampling frame was a list of mobile telephone numbers from each of Bangladesh’s four mobile telephone operators (i.e., Grameenphone, Robi Axia, Banglalink, and Teletalk). Telephone numbers were restricted to those active in DCC, or if they could not be restricted to DCC, to those active in Dhaka District. Over 75% of the population of Dhaka District resides in DCC (23). Telephone numbers were provided by each mobile telephone operator with the permission of the Bangladesh Telecommunication Regulatory Commission (BTRC).
A single stage stratified probability sampling design was used to select participants. Before selection, the telephone numbers were stratified by mobile operator and sampled in accordance with each operator’s proportionate market share in order to maximize precision of the sample and ensure a representative distribution (Supplementary Table 1) (24). Within each operator list, simple random sampling was used to select telephone numbers. At the time of contact, each selected mobile telephone respondent was screened for eligibility and an equal number of male and female respondents were recruited to allow for robust sex-specific analyses. Individuals were eligible for inclusion if they were at least 18 years of age, current DCC residents, and had been residing in DCC for the past one year.
The questionnaire was based on previous poultry exposure surveys conducted in urban China (25–28), but modified to the Bangladeshi context through discussions with an advisory panel (n = 12) consisting of local experts in survey design, mobile telephone surveys, and infectious diseases. Using a structured approach, the panel reviewed each survey question to assess the face and content validity of the items, as well as identify areas for potential adaptation or modification and item reduction or addition. Two rounds of review were conducted and any items that did not achieve group consensus (defined as 60% agreement) were modified and re-examined until consensus was reached. Key revisions from this step centered around prioritizing and selecting items that were deemed feasible and reliable to ask participants during a telephone interview. The questionnaire was translated into Bangla, and independently reviewed by two native-speakers with familiarity of the content matter to ensure comprehension and clarity.
The final survey instrument was comprised of five sections and captured information on exposure to live poultry through purchasing at live bird markets (LBMs) and food preparation, prevention practices, influenza-like illness (ILI), and socio-demographics. LBMs were defined as a collection of stalls or vendors where the general public could purchase live chickens, ducks, geese or any by-products of these in an unprocessed form (29). Specifically, questions covered the following topics: frequency of LBM visits and associated behaviours at markets, poultry processing practices during food preparation, uptake and adherence to hygiene practices and personal protective equipment (i.e., gloves, facemask, apron) during and after poultry exposure, self-reported ILI using a standard case definition (30), as well as household and individual-level socio-demographics. To minimize respondent burden while obtaining detailed information where appropriate, the survey used a significant amount of branching logic. The questionnaire underwent thorough review, and modifications were made as needed based on feedback from a pre-testing phase (n = 7) and a small-scale pilot (n = 41). The final, updated survey took approximately 10–15 minutes for respondents to complete.
Data Collection & Calling Procedure
Both English and Bangla versions of the questionnaire were programmed into a customized computer assisted telephone interview (CATI) platform developed by the Institute of Epidemiology, Disease Control and Research (IEDCR) in Dhaka, Bangladesh. This platform managed both the sampling and data collection processes, including: complex form structure, automated repeat call attempts and interview rescheduling, automated strata monitoring on key variables (i.e., mobile telephone operator, sex of respondent) across interviewers, and pairing with a mobile telephone application to facilitate automated dialling of each selected telephone number. A team of four female data collectors were recruited to conduct telephone interviews, and data were entered into the CATI platform in real-time. Data collectors received four days of training on the survey methods and questionnaire topics before the start of piloting and data collection.
The survey was conducted between September and November 2019. In advance, a Bangla-language newspaper advertisement was placed in DCC’s two most commonly circulated newspapers to inform the public that they may receive a call from IEDCR regarding a health survey, that telephone numbers were randomly selected with the permission of BTRC, and that participation was important for improving population health. Telephone calls were made every day (7 days a week) between 8am and 8pm (local time), except on Friday afternoons to account for local religious observances, to limit potential selection bias that could occur by only recruiting during weekdays and work hours.
All telephone numbers were attempted up to four times to establish contact and conduct an interview with the respondent. Each unanswered call was automatically re-scheduled for a different time of day on a different day of the week over the following seven-day period. If the respondent was not reached after the maximum number of four call attempts, with at least one daytime and one evening call attempt, the telephone number was classified as ‘no contact’ and discontinued. At first successful contact, respondents were explained the purpose of the study, survey length, that participation was voluntary, and that all information they provided would be kept confidential. Eligibility was confirmed and consent for survey participation was obtained at the time of interview. When respondents were unable to complete the interview at the time of recruitment, the telephone interview was re-scheduled for a convenient time within the next seven days. Once an interview was completed, or if a respondent declined, refused or was ineligible, the telephone number was also discontinued from the call bank. An overview of the recruitment process is displayed in Supplementary Fig. 1.
A total of 1040 complete interviews (520 males and 520 females) were required in order to detect an 8–9% difference (65% vs. 56% (26)) in live poultry exposure between strata, with 95% confidence and 80% power. The reason for explicitly stratifying by sex was to have sufficient statistical power to permit detailed exploration and identify notable differences in high-risk behaviours between males and females. This is important for ensuring appropriate and targeted risk-based implementation strategies.
Data Analysis: Response, Weighting, and Representativeness
Operational data for each telephone number dialled and the corresponding details for call outcome status were summarized. The overall and mobile operator-specific response rates were calculated according to the American Association for Public Opinion Research (AAPOR) Response Rate-3 definition, which includes those who were eligible and those estimated to be eligible in the denominator (31). The number estimated to be eligible was derived by assuming the proportion of eligible individuals amongst those contacted was the same as for those who were unable to be contacted or declined.
Demographic data for completed interviews were summarized, and the sample distributions were compared to the Dhaka City Corporation demographic profile of the 2011 census (23). To adjust for non-response and disproportionate stratified sampling by sex (i.e., oversampling of females as compared to the reference population), post-stratification weights were calculated by age, sex and education to align with the 2011 census. Participants with an invalid response to weighting variables (i.e., age, sex, education) were unable to be assigned a weight and therefore are not included in weighted analyses (n = 16). The demographic distribution of the weighted data was summarized and compared with external data sources to assess the representativeness of the sample population for other key demographic variables, including marital status and region. All analyses were conducted in Stata 16.1 (StataCorp, College Station, TX, USA).