This study uses cross-sectional pre-intervention baseline data from a parent study that is a longitudinal, cluster randomized control trial (Clinical Trials # NCT03665246) to evaluate the impact of a maternal mental health/ECD intervention called Integrated Mothers and Babies Course (iMBC) (16,17). The parent trial included 32 communities/clusters with 16 clusters per arm. The average cluster size was over 11 per group, leading to a final baseline sample size of 374 and >90% power for the parent study’s primary outcome (maternal mental health) (18).
The study was a collaboration between Duke University and Catholic Relief Services (CRS). CRS-Ghana implemented the intervention with approval and support from the Ghana Health Service. IMBC is part of CRS’ Rural Emergency Health Service and Transport (REST II) project that aims to scale-up community-based approaches to health services for improved maternal, newborn, child health, and nutrition practices. A priority activity of REST II is Community Surveillance and Targeted Education Sessions (C-PrES) in each community. The parent study randomized clusters such that women in the iMBC groups received all the same information as the standard C-PrES groups, but with the added iMBC intervention, which had content on stressors during pregnancy and how to better manage those stressors to decrease the risk of future depression. This manuscript adheres to the STROBE statement guidelines for reporting cross-sectional studies (19).
Study Setting & Participants
Pre-intervention baseline data was collected between August 27 and September 6, 2018 among 32 communities in the West Mamprusi Municipality and the Nabdam District of Northern Ghana. These areas are considered rural populations where a majority of the residents are in the lowest wealth quantile and have high infant mortality rates (20). Participant inclusion criteria included being pregnant at baseline, 16 years or older, planned to attend C-PrES groups at the time of the baseline survey, and planned to maintain residence in the community for at least six months.
Participants gave written informed consent via signature or thumbprint with a witness. Interviewer-administered surveys were conducted by research assistants (RAs) in the local languages (Mampruli and Nabt) on Samsung Galaxy tablets. The Mampruli and Nabt languages are not commonly written, therefore, the survey was written in English and was vocally translated at the time of each interview by the local research assistants. Before data collection began, all 24 RAs participated in a week-long training in which they reviewed the survey content, practiced administering the survey in English and, and then practiced administering the survey in their local language. In addition, mid-week during the training, all research assistants participated in pre-testing of the survey in the community in order to finalize any edits on the survey and to ensure that all research assistants were familiar with and could debrief about their common understanding of the survey items prior to formal data collection. Please see the Supplementary Materials for the full survey. Data was captured by using CommCare, a HIPAA (The United States Health Insurance Portability and Accountability Act of 1996) compliant, cloud-based, and field data collection and reporting platform, published by Dimagi, Inc. Surveys were conducted at participants’ homes or at another agreed upon private location to ensure confidentiality and participants were compensated with two bars of soap.
Study risks were minimal; however, for the mental health assessment, if participants endorsed suicidal ideation, RAs provided a referral to district social welfare and mental health officers for further support per guidelines from the Ghana Health Service who attended the RA training. Similarly, for domestic violence survey items, if participants indicated physical or sexual violence, they were offered a referral to the Domestic Violence and Victim Support Unit through the District Gender Officer, which aligned with Ghana Health Service recommendations and global best practices for gender-based violence research (21).
Ethical approvals were received from the Duke University Campus IRB (ID# 2019-0020) and the Navrongo Health Research Centre Institutional Review Board (ID # NHRCIRB314).
The main outcome variable was early stimulation behaviors performed by the pregnant women towards their pregnant belly. Throughout this paper, we will use the term ‘pregnant belly’ because that is how it is referred to programmatically and in the questionnaire. These behaviors included touching/talking to her belly, singing songs, dancing, and talking to her belly about family. These were example ESB that Catholic Relief Services used in the iMBC sessions and have used in other programmatic materials (17). As part of the original survey, two additional questions were asked regarding whether the father touches/talks to the expectant mother’s belly and if other children touch/talk to the belly (part of CRS program materials). It was decided not include these two questions beyond descriptive analysis due to confounding factors such as living situation, parity and relationship status that could interfere with the results. Women were asked to self-report how often they performed each behavior, with scoring as never (0 points), rarely (1 point), sometimes (2 points), frequently (3 points).
Previous research has shown that touching and talking to the belly is widely recognized and measured as an indicator of maternal-fetal attachment (2,22). To better align with the literature, we created a binary outcome variable for analyses, that combined communication-related ESB performed by the expectant mother. These behaviors were touching/talking, singing, and telling about family. Dancing was not included because it does not necessarily involve directed sound or touch. Women who reported sometimes or frequently to any of the three communication behaviors were one category and women who reported only rarely or never were in the other.
Intimate partner communication was measured using the communication subscale of the Relationship Quality Index (RQI) (23). The Constructive Communication subscale is divided into a three-question constructive and a four-question destructive scale. Responses were on a 5-point Likert scale, ranging from very unlikely (1) to very likely (5). The sum of the four ‘destructive’ items was subtracted from the sum of the three ‘constructive’ items to create the total intimate partner communication score (24). Constructive communication items included couples discuss problems, express feelings, and suggest solutions and compromises. Deconstructive communication items included couples blame each other, threaten with negative consequences, male partner calls the woman names and attacks her character, and female partner calls the man names and attacks his character.
Mental health was assessed using the Patient Health Questionnaire (PHQ-9), a common depression screener that has been validated for use among pregnant women in Ghana (25). The PHQ-9 has nine items that are summed for a score between 0 and 27 and standard categorization: minimal or no depression (score of 0 to 4), mild depression (5-9), moderate depression (10 to 14), moderately severe depression (15-19), and severe depression (20-27) (26). For analysis, we dichotomized the PHQ-9 into none to mild depression (score of less than 10) and moderate to extremely severe (score of 10 or greater), based on those who would have screened for no treatment or treatment in a clinical setting (25,27). Cronbach’s alpha for the PHQ-9 was 0.815, indicating high internal consistency.
Food insecurity was assessed using the Household Hunger Scale (HHS), a six-question scale with three main questions [went 24 hours with no food, went to sleep hungry, and no food in the house due to lack of resources], each of which are followed by a question asking how often this event occurred. Questions are scored 0-6, with higher scores indicating higher food insecurity. For analysis, we used the standard categorical variables indicating little to no hunger (0-1 points), moderate to severe hunger (2-3 points), and severe hunger (4-6 points) (28).
Intimate partner violence (IPV) during the past 12 months was assessed using items from the 2008 Ghana Demographic and Health Survey (DHS). The four domains were controlling behaviors, physical IPV, sexual IPV, and emotional IPV (28). Binary variables were created for each domain, indicating if the respondent endorsed at least one item within that domain.
Participants’ hopefulness was measured using the 12-item Herth Hope Index and analyzed as a continuous variable, with higher scores indicating higher hopefulness (30,31). The Cronbach’s alpha was 0.793, indicating high internal consistency.
Four questions were asked to determine participants’ perceived level of social support. These questions asked how much assistance you received in the past month from your husband/partner, female relatives, male friends, or female friends. Response options were sufficient, insufficient, or never received social support, and each question were recoded as a binary variable (never/insufficient support or sufficient support).
Additional questions included relationship status, woman’s age, parity, whether they had ever attended formal education, and self-reported physical health.
Data Management and Analysis
Data collected via CommCare was uploaded and synced with the main database and analyzed using Stata version 16.1. A generalized estimating equations (GEE) modified Poisson model was used for bivariate and multivariable analyses, taking into account clustering by using an exchangeable working correlation (32). The Kauermann-Carroll bias correction was used to account for potential small-sample bias in the standard errors, since the trial had fewer than 40 clusters (33,34). For the bivariate model, the correlation between the early stimulation factor score and each covariate was evaluated. The multivariable model was determined by including both a priori variables, education and parity, and excluding non-significant variables at the p-value level of 0.10.