Influence of engaging female caregivers in households with adolescent girls on adopting equitable family eating practices: a quasi-experimental study

Study setting, design, and population

The study was conducted in the West Hararghe zone of the Oromia regional state in Ethiopia. The study was conducted in three Woredas, where two weredas implemented SAA intervention and one Wereda served as a comparison area. The majority of the population in these areas belongs to the Oromo ethnic group and follows the Islam faith. In addition, agriculture-based living styles are dominant in the area.

This article utilized data from a large quasi-experimental study that assessed various interventions to address structural issues affecting young adolescent girls in the Western Hararghe Zone, Oromia Region, Ethiopia (Berhane et al., 2019). This study’s population was female caregivers with at least one young adolescent girl (age 13 and 14) in the household. The intervention primarily targeted female caregivers who are responsible for food preparation and allocation in the households.

Intervention (SAA)

The SAA was initiated by assembling female caregivers along with other influential adults in the community, including religious leaders and other norm holders. The group consists of about 30 influential adults to ensure adequate dialogue and greater impact by engaging influential adults in the process, which promotes open and candid discussion, seeking solutions to identified challenges, and promoting actions to address the challenges. The process starts with the training of the partcipants by trained facilitators. The SAA was facilitated by selected and trained community members from each group. Trained support staff were assigned to support the groups in person or via telephone to maintain fidelity to the intervention. The SAA is designed to change community social norms through open dialogue (Tasew et al., 2018). The groups met monthly, and each discussion session lasted an hour or more. The central discussion topics girls related to this research outcome were adolescent girls’ nutrition. The group met every month for 3 years during the intervention period.

Mechanisms/causal pathways

The intervention (SAA) influences the behavior of individuals or groups, leading to changes in their decision-making, habits, or actions related to family eating practices, focusing on adolescent girls. For instance, understanding adolescent girls’ nutritional needs and correcting cultural miss conception might lead to changes in intrahousehold family eating practices. In addition to this, Socio-cultural contexts can play a role in how individuals or groups behave in a given situation. These factors might mediate or modify treatment effects (Fig. 1).

Sampling procedure

The study population was selected by a two-stage cluster sampling method, as it was described in the previous article (Berhane et al., 2019). First, for each study arm (intervention and comparison arms), 38 clusters (locally known as development areas or ‘Gere,’ a division within the smallest administrative unit, the kebele) were randomly selected after preparing the list of all clusters in the study woredas. Then, in each cluster, the list of households with adolescent girls (aged 13–17 years) was prepared by conducting a complete household listing. In the final stage, 30 households with eligible adolescent girls were randomly selected from the list. Then, female caregivers in the sampled households were invited to participate in the study.

Data collection

A structured questionnaire was used to collect relevant data. The questionnaire contains questions related to socio-demographic variables from the Ethiopian Demographic and health survey (EDHS) (Ethiopian Public Health Institute (EPHI) [Ethiopia] and ICF, 2021) and standard questions used for the household food security assessment (Coates, Swindale & Bilinsky, 2007). The questionnaire was first developed in English and then translated into the local language, Afaan Oromo. Data were captured before the intervention (at baseline) and after a period of implementing the intervention for 3 years (at end line) using an open data kit (ODK), which is an electronic data collection platform. Data were collected by trained enumerators fluent in the written and spoken local language, Afaan Oromo.

Measurement

The effect of the intervention (social analysis and action) was measured by the number of families eating together, which was assessed at baseline before the intervention commenced and at the end-line after the intervention. The family was considered as eating together if a family mostly ate at the same time from the same dish, as reported by female caregivers. Then, the family eating practice, the outcome variable, was dichotomized into eating together or not.

Data analysis

The data analysis was done on a combined baseline and end-line dataset. A time variable was created as ‘time,’ the baseline dataset was coded as ‘0,’ and the endline was coded as ‘1’. We compare differences in eating practices between the study groups using chi-square tests. The Difference-in-Difference (DID) approach is a method to assess the program’s impact by comparing changes over time between the comparison group. For this study, the DID model is based on a mixed-effect logistic regression. The model assumes that in the absence of the intervention, the difference between the intervention and comparison group would be constant over time, which is a parallel trend assumption. Difference-in-Difference is implemented by adding an interaction term between time and program group dummy variables in a regression model and can be specified as follows:

${\mathrm{Y}}_{\mathrm{i}}={\mathrm{b}}_0+{\mathrm{b}}_1\ast [\mathrm{T}\mathrm{i}\mathrm{m}{\mathrm{e}}_{\mathrm{i}}]+{\mathrm{b}}_2\ast [\mathrm{p}\mathrm{r}\mathrm{o}\mathrm{g}\mathrm{r}\mathrm{a}{\mathrm{m}}_{\mathrm{i}}]+\mathrm{b}3\ast [\mathrm{T}\mathrm{i}\mathrm{m}{\mathrm{e}}_{\mathrm{i}}\ast \mathrm{P}\mathrm{r}\mathrm{o}\mathrm{g}\mathrm{r}\mathrm{a}{\mathrm{m}}_{\mathrm{i}}]+{\mathrm{b}}_4\ast [\mathrm{C}\mathrm{o}\mathrm{v}\mathrm{a}\mathrm{r}\mathrm{i}\mathrm{a}\mathrm{t}\mathrm{e}\mathrm{s}]+{\epsilon }_{\mathrm{i}}$

in which ${\mathrm{y}}_{\mathrm{i}}$ is the outcome variable, which is family eating practice measured at the baseline and end line in both intervention and comparison group.

${\beta }_0$ is the intercept, which is the value of the outcome variable when all of the other variables are equal to zero.

$\mathrm{T}\mathrm{i}\mathrm{m}{\mathrm{e}}_{\mathrm{i}}$ is a time variable, which is the baseline or end-line period that takes the value 0 or 1 respectively.

$\mathrm{P}\mathrm{r}\mathrm{o}\mathrm{g}\mathrm{r}\mathrm{a}{\mathrm{m}}_{\mathrm{i}}$ is a program variable, which is the control or treatment groups that coded as 0 or 1 respectively.

$(\mathrm{T}\mathrm{i}\mathrm{m}{\mathrm{e}}_{\mathrm{i}}\ast \mathrm{p}\mathrm{r}\mathrm{o}\mathrm{g}\mathrm{r}\mathrm{a}{\mathrm{m}}_{\mathrm{i}})$ is an interaction term between time and program, which used to test the effect of the intevention.

${ϵ}_{\mathrm{i}}$ is the error term, which captures the effect of all factors that affect the outcome but the model could not adequately represent.

The models were adjusted for clustering and unbalanced covariates between the intervention and control group. A mixed-effect model was used to control the clustering effect as the data is clustered. As the outcome variable was binary, we used mixed-effect logistic regression. We used the Difference-in-Difference (DID) model with the interaction terms to determine the intervention effect. The impact was examined based on a statistically significant difference of p-value < 0.05, the crude odds ratios (COR), and adjusted odds ratios (AOR) with 95% confidence intervals (CI) of the interaction term (program × time), using Stata version 14 statistical software. We checked for multicollinearity and outliers during the analysis. Mixed effect logistic regression controlled for clustering effect and other variables such as female caregiver’s educational status and income, male caregivers’ education status and income, household wealth and food security status, intervention, time, and difference in the different interaction terms.

Ethical considerations

The study protocol was reviewed and approved by the Ethical Review Board of the Addis Continental Institute of Public Health (Ref No. ACIPH/IRB/005/2016). Due to the low literacy level of the community and the minimal risk involved in the study, informed verbal consent was obtained after explaining the purpose of the study. Participants were informed about their rights to refuse participation and/or withdraw their consent at any time. All interviews took place in private settings. No personal identifiers were linked with the dataset made available for this study.

The effect of the intervention

The family eating practice, eating together, improved in both study arms but more in the intervention arm. There was a 20.57% increase in the intervention arm compared to 8.58% in the comparison group. Overall, the intervention group showed an 11.99% increase compared to the comparison group (Pr = 0.001) (Table 3).

After controlling for potential confounders and clustering effect, the DID analysis showed a significant difference in eating together between the comparison groups. The adjusted odds ratio showed that eating together was twice as likely in the intervention arm as compared to the control arm (AOR 2.08 (95% CI [1.06–4.09]), p-value of 0.033). In addition, male caregivers’ education, income, and household food security status were significantly associated with the outcome (Table 4).

Heterogeneity analyses

We also did a heterogeneity analysis by adolescent girls’ ages (13 and 14 separately) and female caregivers’ education (never attended and primary and above). After controlling for potential confounders and clustering effect, the DID interaction term showed a significant difference in eating together among the younger adolescents (AOR 2.96 (95% CI [1.09–7.99]), p-value of 0.032) and among female caregivers with better educational status (AOR 4.03 (95% CI [1.01–16.13]), p-value of 0.049) (Table 5).