Success of health cell approach in improving knowledge, attitude and practice related to COVID-19: difference-in-differences analyses of a community-based quasi-experimental trial

Study site and participants

We conducted a community-based quasi-experimental trial among residents of Bauniabadh and the adjacent slum areas in Mirpur, Dhaka, Bangladesh. The Bauniabadh slum area has six different zones named A, B, C, D, and E blocks and Kalabagan areas. A total of 203 households from the B block and Kalabagan area received the intervention and another 203 households from the D block were selected as the comparison cohort. We used a sampling frame and a simple random sampling technique to select the participating households. A buffer zone was maintained between the intervention and control areas to avoid information spillover. The A and C blocks were kept as the buffer zone that covers approximately 400 m of distance between the intervention and control areas. Field research assistants (FRAs) visited the households and, based on the predefined inclusion (participants living within the geographical location and were willing to consent for participation) and exclusion (receiving similar interventions from other sources) criteria, they screened the families. If a family was eligible to participate, the FRAs gathered written consent from the participants (household heads (HHs) and homemakers (HMs) of the selected households) using a consent form developed according to their literacy level. Participants got the opportunity to review the consent forms thoroughly before consenting to participate in the study.

Sample size

We used the G*Power (V.3.0.10) software for calculating the sample size. We assumed that, at baseline, the interviewed participants from intervention households would be able to answer 50% of the KAP questions correctly and the interventions would enable them to answer 80% of the questions correctly at the end of the intervention. Therefore, the effect size we considered was 30%. Additionally, we considered the level of significance as 0.05 and power as 90% for the sample size calculation. Considering all these assumptions, each group’s calculated sample size (number of households to be included) was 183. Then, taking account of an attrition rate of 10% points, a minimum sample size of 203 was required in each of the groups, and a total of 420 participating households (211 intervention and 209 control) were enrolled in the study.

Design and functionality of the health cells

We set up three health cells in Bauniabadh and adjacent slum areas of Mirpur, Dhaka. Each health cell was equipped with two well-trained health service providers from International Centre for Diarrhoeal Disease Research, Bangladesh (icddr,b; one FRA and one health worker), BCC materials, sphygmomanometers, thermometers, glucometers, pulse oximeters and face masks. We used standard BCC materials developed by icddr,b and World Health Organisation. Members of the health cells visited the households to recruit the participants. These service providers were well trained in providing BCC messages and were well-known to the community.

BCC message was provided through small-scale community meetings with HMs (mother or lead female member of a household) and HHs (lead earning member of a household). These meetings were conducted in the courtyard of the participants and strict social distancing was maintained. Eight to 10 members from the selected households were present in the meeting. Each session was 20–30 min long and each household joined the sessions twice a month. Leaflets and posters on effective use of face covering, hand cleansing and social distancing were also distributed during the educational sessions. The messages included simple but scientifically proven information on how, when and where to wear face covering, maintain social distancing and clean hands. The health workers demonstrated the household members how to make soapy water—a simple, cost-effective hand cleansing solution invented by icddr,b.6 Households of comparison group received face masks and hand cleansing solutions. No educational interventions or BCC materials were provided to them. We referred the participant from the intervention and comparison households to appropriate health facilities in case of any illness they reported.

Data collection

We reviewed the KAP surveys conducted in different parts of the world and selected a set of questionnaires on COVID-19-related KAP.7–9 The questionnaire was then pretested to check the applicability in local context. The pretesting was done in the A and C blocks of the Bauniabadh slum area of Mirpur, Dhaka. A total of 25 participants answered the questions asked to them and no issues were found regarding understanding or application of the questionnaire. The pretested version of the questionnaire was used to document the before–after changes in KAP of the household members regarding different preventive measures of COVID-19. Data were collected before and after the intervention period from the HHs and HMs to document the effect of the delivered intervention on KAP related to COVID-19.

Outcome variables

The questionnaire’s KAP sections had 14, 12 and 8 questions, respectively. Answers to 34 questions (online supplemental table 1) were sought to assess the participant’s KAP before and after the delivered intervention. The total number and per cent of questions a responder could answer correctly were the outcome variables for the analyses.

Data analysis

We report the overall and cohort-specific socioeconomic, household characteristics and KAP using mean, SD, frequency and percentages. Student’s t-test was performed for group-wise comparisons of quantitative symmetric variables. Group-wise comparison of the quantitative asymmetric data was made using the Kruskal-Wallis test. The Χ² or Fisher’s exact test was done for qualitative variables. The correlation between KAP scores before and after the intervention was measured, and Spearman rank correlation coefficients with corresponding p values were reported.

The difference-in-differences (DID) analysis, a quasi-experimental data analysis technique, was used to measure the effect of the delivered intervention on KAP of the household members. In the regression models, a p value less than 0.05 was considered statistically significant and regression coefficients with 95% CIs were calculated. The multivariable DID models were statistically adjusted for WAMI index (Water, sanitation, hygiene, Asset, Maternal education and Income index), HH’s age, sex, total years of education, study group and timing of data collection. WAMI index (ranging from 0 to 1)—a socioeconomic status index that includes access to improved water and sanitation, eight selected assets (separate room for a kitchen, household bank account, mattress, TV, refrigerator, people per room, table, chair or bench), maternal education and household income—was used as a representative of socioeconomic status of the households.10 A higher WAMI index means a better socioeconomic status.

DID modelling technique measures the true effect of the intervention on the outcome and its before–after changes with the following formula: DID: [(B−A)−(D−C)], where:

A=baseline values in intervention group.

B=endline values in intervention group.

C=baseline values in comparison group.

D=endline values in comparison group.11

To assess the true effect of the intervention, we applied a regression model with a generalised estimating equation as follows: Yit=β0+β1Time+β2Group+δ (Time×Group)+β3X+ε (2), where:

Yit=outcome variable of interest for individual i at time t.

Time=(1) if endline and (0) if baseline.

Group=(1) if intervention group and (0) if comparison group.

δ=the effect of the delivered intervention, X=other covariates and ε=error term.

Patient and public involvement

Patients or the public WERE NOT involved in the design, or conduct, or reporting, or dissemination plans of our research.