Exploring the gap between coverage, access, and utilization of long-lasting insecticide-treated nets (LLINs) among the households of malaria endemic districts in Bangladesh

This study employed a cross-sectional design triangulated with qualitative research method. In this case, three-stage cluster random sampling of households was applied. In the first stage of sampling, 30 upazilas were randomly selected from 71 upazilas where LLINs have been distributed. In the second stage of sampling, four villages were randomly selected from each upazila for statistically reliable estimate. In the third stage, 22 households per village having at least one under 5 and/or pregnant woman in the household were randomly selected.

This study was conducted in 13 malaria-endemic districts in the north, northeast and southeast regions in Bangladesh bordering India and Myanmar (Fig. 1), which included Khagrachhari, Rangamati, Bandarban, Cox’s Bazar, Chittagong, Sylhet, Sunamganj, Moulvibazar, Hobiganj, Mymensingh, Netrakona, Sherpur, and Kurigram. Three hill districts: Khagrachhari, Rangamati and Bandarban, located in the hilly remote areas of Chittagong hill track (CHT), suffer from a geographical disadvantage with difficult communication and intense perennial transmission of malaria. The field data were collected between April and May 2017.

Information on socio-economic and demographic condition, coverage, access to and use of LLINs was collected from 2640 households. The respondents were mainly the household head. If the household head was not present at the time of the interview, the information was collected from the husband/wife of household head or a pregnant woman or mother of children under 5 years old (Fig. 2).

For the qualitative part, a purposive sampling strategy was employed to select information-rich cases associated with the field of interest [14]. Two districts from two different endemic areas (CHT and non-CHT) and two villages from each district were selected. In each village, four in-depth interviews (with pregnant women) and one focus group discussion (with community people) were carried out. Another focus group discussion was conducted in each village with high-risk population groups, e.g., Jhum (cereal crop that grows in hills) cultivators, wood cutter and coal mine workers, and migrants as they often have limited access to prevention, diagnostic testing and treatment services. Three key-informant interviews were undertaken with frontline development workers of BRAC for qualitative exploration from each district.

Data were collected by research assistants using a pre-tested, structured questionnaire during face-to-face interviews with respondents. Research assistants were trained prior to data collection, encompassing lectures on completing the questionnaires, mock interviews, and field practice. A multilayered monitoring system was in place to validate, standardize and maintain quality of data and tasks such as spot checking, back checking, and provided necessary feedback to the teams working in the field. Different types of checklists were used in different data collection tools for qualitative segment. A team of four temporarily recruited research assistants with anthropology background, along with the lead qualitative researcher to collect field data.

The background variables for respondents were: age (< 20, 20–29 and ≥ 30 years), gender, relationship to household head (head, spouse, other), and level of education (no education, completed grade 1–5, completed grade 6–9, completed grade 10 or more). The background variables for household heads were: gender, level of education (no education, completed grade 1–5, completed grade 6–9, completed grade 10 or more), occupation (farmer, service holder, businessman, self-employed, labourer, housewife, others). The household level characteristics were: area (CHT, non-CHT), ethnicity (Bengali, others), religion (Muslim, Hindu, Buddhist, Christian, others), household size (≤ 4, > 4), wealth quintile (low, middle, high), having at least one child under 5 (yes/no), having at least one pregnant woman (yes/no).

Construction of the wealth index was based on factor analysis [15, 16] of key socio-economic variables. The key socio-economic variables were: types of wall, floor and roof of the house, ownership of radio, television, computer, bicycle, mobile/telephone, refrigerator, wardrobe, table, chair, watch, bed, sewing machine, bike, motor vehicle, livestock, and access to solar system, electricity.

Household-based indicators were used for assessment of coverage and access of LLINs. LLIN use was recorded as whether the household member/s slept under it the night prior to the day of survey [17].

Descriptive analyses were conducted to show the distribution of different characteristics of the respondents and household heads with appropriate cut-offs. The Chi square test [18] was done to assess association of different background characteristics with LLIN coverage, access and use. Binary logistic regression [19] was applied to assess relationship between background characteristics and LLIN use. The final model was adjusted with area of residence, gender and education of household head, ethnicity, religion, household size, wealth quintile, status of household having at least one child under 5 years, and having at least one pregnant woman, and with respondents’ relationship to household head. The adjusted odds ratio (AOR), exponent of beta-coefficient of binary logistic regression was calculated with 95% confidence interval (95% CI). All tests were done at 5% level of significance. All analyses were performed using STATA software (Version 13.0, STATA Corp LP, TX, USA).

Information was collected from a total of 2640 households, where one-third was from CHT area and remaining from non-CHT area. The average family size was 5.5 ± 2.2, with 1.06 female to male ratio. The majority of the respondents were females (88.8%), with 20 or more years of age (91.7%). Only 13.8% respondents were household heads. Nearly one-third of respondents had no formal education and only 11.7% had completed grade 10 or more. In terms of ethnicity and religion, 79.1% were Bengali and 70.3% were Muslim. Also, 60.1% households had more than four members, 76% households had at least one child under 5 and 46.1% households had at least one pregnant woman (Table 1).

A total of 6048 LLINs were distributed among 2640 households with a mean 2.3 ± 1.1 LLINs per households. At the time of the survey, almost every household had at least one LLIN (99.8%), 93.1% households had at least one useable LLIN, 77.4% households had at least two LLINs, and 58.4% households had at least two useable LLINs (Table 2). The coverage of LLINs was high in hill track areas (P = 0.003 for at least two LLINs), households with female heads (P = 0.040 for at least two usable LLINs), and in rich households (P < 0.001 for at least two LLINs or at least two usable LLINs).

Only 43.2% households had sufficient access to LLINs at the time of survey (Table 3). Household access to LLINs was significantly higher in hill track area (54.1 vs 37.8%, P < 0.001), in households with female head (50.4 vs 42.5%, P = 0.021) and in households with 4 or fewer members (63 vs 30.8%, P < 0.001). Furthermore, LLIN access increased with an increase in household head’s level of education.

Out of a total 14,475 members in 2640 households, 91.7% slept under an LLIN the previous night. LLIN use was higher among females than males (92.9 vs 91.0%), and the use of LLINs among females than males increased if the household met with sufficient access (98.3 vs 97.7%) (Fig. 3).

The proportion of LLIN use by households was 77.9%, which means all the members of those households slept under an LLIN the previous night (Table 4). The households’ use of LLINs was higher in hill track areas than non-hill track areas (86.4 vs 73.4%, P < 0.001), and in households with 4 or fewer members than those with more than 4 members (90 vs 69.8%, P < 0.001). Although LLIN coverage was high, usage was reportedly low in rich households (74.2%). Moreover, households’ use of LLINs increased when they had sufficient access to LLINs (94%) (Table 4). This phenomenon is also reported in the qualitative part of the study. According to the qualitative analysis, the lack of access to LLINs was one of the most documented reasons for low utilization. Sleeping outdoors, migrating outside, feeling uncomfortable, lack of knowledge, incompatibility with sleeping arrangements, saving LLINs for another time, and perceived degraded efficacy of the nets were also reported as reasons for poor utilization of LLINs (Table 5).

This study identified three broad areas of gaps in utilizing LLINs: intra-household net allocation, household inaccessibility, and proper usage of LLINs. Only 0.19% (5 out of 2640) households did not have LLINs at the time of the survey. Consequently, intra-household net gap was 56.7%, and 56.8% households had insufficient access. Contrarily, 18.8% (495 out of 2640) households had excess LLINs (i.e., more than one LLIN for every two people). Moreover, the use gap was 6%, meaning members from those 6% households who had sufficient access to LLINs did not use them the previous night.

The qualitative findings of the study also discovered some gap between coverage, access and utilization. Despite having large coverage in the respective locality, some people experienced a lack of LLINs as per the needs of family members. Daily life experience of Rasheda (pseudonym) of Netrokona district can illustrate this statement:

“Having one LLIN in a family of four members is very crucial when necessity is not able to supplement. Moreover, we might face a situation in coming days as my father-in-law would possibly come. How would it be possible to manage the situation with only one net?”

Another issue came to light regarding coverage gap which signifies that the mosquito nets were provided only among families. Individuals living separately in a mess or in shared rooms might not have received LLINs from the programme. SK from Bandarban adds:

“Someone came to me one day while we were distributing mosquito net among the listed population. He asked for a net but we were not able to serve him as we do not have permission to give it to any individual who live in a mess or shared household or who was not in our distribution list. I felt pity for him but there was nothing I could do.”

The association of background factors of households with sufficient and insufficient access to LLINs is shown in Table 6. Of households with sufficient access to LLINs, LLIN use was approximately five times less for Bengali community than others (AOR 0.22, 95% CI 0.07–0.66). Besides, households with more than four members had more than double likelihood of using LLINs than households with four or fewer members (AOR 2.64, 95% CI 1.51–4.59). For households with insufficient access to LLINs, the odds of using LLINs was approximately 50% higher in hill-track areas than non-hill track areas (AOR 1.46, 95% CI 1.04–2.04). Similarly, the Bengali community had lower odds of using LLINs (AOR 0.53, 95% CI 0.30–0.94), and households with more than four members had higher odds (AOR 2.13, 95% CI 1.60–2.82). Remarkably, households with high wealth quintiles were 24% less likely to use LLINs than poor households (AOR 0.76, 95% CI 0.57–0.99). Besides, the odds of LLIN use were nearly 50% higher when there was at least one child under five in the household (AOR 1.47, 95% CI 1.03–2.08). In addition, the odds of LLIN use was 60% higher for spouses compared to household heads (AOR 1.60, 95% CI 1.12–2.27) (Table 6).

This survey was conducted following the rainy season in the study area when mosquitoes were abundant. This should be taken into account when interpreting the findings as the season might have encouraged people to use the nets more. Outcomes in coverage and access as well as LLIN utilization could show different proportions if the study was conducted in other seasons of the year. It is challenging to investigate causation from cross-sectional design, and only association can be truly estimated. A strength of this study is direct observation that has been conducted by data collectors and research staff for all applicable questions in order to minimize response bias during the interviews.