Evaluation of a community-based mobile video breastfeeding intervention in Khayelitsha, South Africa: The Philani MOVIE cluster-randomized controlled trial

Background In South Africa, breastfeeding promotion is a national health priority. Regular perinatal home visits by community health workers (CHWs) have helped promote exclusive breastfeeding (EBF) in underresourced settings. Innovative, digital approaches including mobile video content have also shown promise, especially as access to mobile technology increases among CHWs. We measured the effects of an animated, mobile video series, the Philani MObile Video Intervention for Exclusive breastfeeding (MOVIE), delivered by a cadre of CHWs (“mentor mothers”). Methods and findings We conducted a stratified, cluster-randomized controlled trial from November 2018 to March 2020 in Khayelitsha, South Africa. The trial was conducted in collaboration with the Philani Maternal Child Health and Nutrition Trust, a nongovernmental community health organization. We quantified the effect of the MOVIE intervention on EBF at 1 and 5 months (primary outcomes), and on other infant feeding practices and maternal knowledge (secondary outcomes). We randomized 1,502 pregnant women in 84 clusters 1:1 to 2 study arms. Participants’ median age was 26 years, 36.9% had completed secondary school, and 18.3% were employed. Mentor mothers in the video intervention arm provided standard-of-care counseling plus the MOVIE intervention; mentor mothers in the control arm provided standard of care only. Within the causal impact evaluation, we nested a mixed-methods performance evaluation measuring mentor mothers’ time use and eliciting their subjective experiences through in-depth interviews. At both points of follow-up, we observed no statistically significant differences between the video intervention and the control arm with regard to EBF rates and other infant feeding practices [EBF in the last 24 hours at 1 month: RR 0.93 (95% CI 0.86 to 1.01, P = 0.091); EBF in the last 24 hours at 5 months: RR 0.90 (95% CI 0.77 to 1.04, P = 0.152)]. We observed a small, but significant improvement in maternal knowledge at the 1-month followup, but not at the 5-month follow-up. The interpretation of the results from this causal impact evaluation changes when we consider the results of the nested mixed-methods performance evaluation. The mean time spent per home visit was similar across study arms, but the intervention group spent approximately 40% of their visit time viewing videos. The absence of difference in effects on primary and secondary endpoints implies that, for the same time investment, the video intervention was as effective as face-to-face counseling with a mentor mother. The videos were also highly valued by mentor mothers and participants. Study limitations include a high loss to follow-up at 5 months after premature termination of the trial due to the COVID-19 pandemic and changes in mentor mother service demarcations. Conclusions This trial measured the effect of a video-based, mobile health (mHealth) intervention, delivered by CHWs during home visits in an underresourced setting. The videos replaced about two-fifths of CHWs’ direct engagement time with participants in the intervention arm. The similar outcomes in the 2 study arms thus suggest that the videos were as effective as face-to-face counselling, when CHWs used them to replace a portion of that counselling. Where CHWs are scarce, mHealth video interventions could be a feasible and practical solution, supporting the delivery and scaling of community health promotion services.

This study was conducted in an underresourced region of the Western Cape Province [13,35]. This part of the province is characterized by high infant mortality rates and a low prevalence of EBF compared with country-wide statistics [36]. Within this region, Philani mentor mothers serve more than 100 neighborhoods [11], providing in-home health promotion counseling services and social support. Settlements in this region are characterized as either “formal” (containing government-constructed permanent dwellings made of brick and mortar), “informal” (containing community-constructed shacks made of wood, corrugated iron, and other available materials), or “mixed” (containing both types of dwellings). Mentor mothers are positive role models as well as frontline healthcare providers in their communities. Each mentor mother is trained in performing growth monitoring, counseling pregnant women and mothers on perinatal health, infant feeding, HIV and tuberculosis prevention and management, basic nutrition, early child development, and alcohol/drug avoidance [11,12,37]. Fig 1 shows the study setting. Left panel: map of the Western Cape Province, South Africa, (illustration by Sufian Ahmed). Right panel A: a Philani mentor mother walking between home visits. Right panel B: collage of homes and shops in Khayelitsha, South Africa. Right panel C: Bird’s eye view of Khayelitsha, a sprawing informal settlement (photos by Maya Adam). MOVIE, MObile Video Intervention for Exclusive breastfeeding. We designed a stratified, cluster-randomized controlled trial, with mentor mothers as the unit of randomization [36]. Mentor mothers recruited pregnant women as participants in this trial during the enrollment period. Analyses of existing program data from Philani suggested that the mentor mother’s settlement type (formal, informal, or mixed) was the only significant predictor of EBF, among a range of covariates. We thus chose to stratify our randomization by settlement type, thereby attempting to balance the video intervention and control groups. Based on prior research [9,38–45] we also expected several covariates to influence our primary outcome measure: (participant’s number of previous children, participant’s age, running water in the home, electricity in the home, participant’s employment status, and participant’s education level). We measured these covariates at baseline and adjusted for them in sensitivity analyses to increase the statistical efficiency of our estimation. At the outset of the trial, each mentor mother was living and counseling participants within her neighborhood. In October 2019, 5 months before data collection was terminated prematurely due to the Coronavirus Disease 2019 (COVID-19) pandemic, the provincial government altered Philani’s service demarcation. Roughly half of the mentor mothers at Philani were affected by the change and instructed to cease counseling the mothers who lived outside of their new demarcations. In our study, 12 mentor mothers (6 in the video intervention arm and 6 in the control arm) needed to be replaced by other mentor mothers, some of whom lived in settlement types that were different from those of the originally assigned mentor mother. The new demarcations led to a slight change in the distribution of settlement types in which the mentor mothers lived at endline (Intervention:Control = Formal 3:5, Informal 14:9, Mixed 25:27); however, the participants originally recruited remained in the study. For those participants whose mentor mother was replaced, data were collected by the replacement mentor mother. The COVID-19 pandemic and the changes in demarcation resulted in an unanticipated number of participants who were lost to follow-up, especially at the 5-month data collection point. Our sample size calculation was based on the primary outcomes: EBF at ages 1 and 5 months. We used standard methods for cluster-randomized controlled trials (with stratification, as well as with and without baseline covariate adjustment) to calculate our sample size [46]. The mentor mothers served as our unit of randomization. We assumed an intracluster correlation for each of our 2 primary outcomes of 0.1. Our power calculation was informed by routine program data describing the performance of the mentor mothers in our study and other sources of data on breastfeeding in South Africa [3–5,8,43,47,48]. Based on these data, we assumed that 40% of mothers exclusively breastfeed their infants at age 1 month and 10% of mothers exclusively breastfeed their infants at age 5 months. We further assumed that each mentor mother would enroll an average of 12 pregnant participants over the course of our trial. For the sample size calculation with baseline covariate adjustment, we assumed a correlation between the baseline measurements and the primary outcome of 0.30. We estimated that the trial would have 80% power to detect, at the 5% significance level, a 13-percentage point increase in the primary outcome at age 1 month and 9-percentage point increase in the primary outcome at age 5 months. These minimal detectable differences satisfied our condition for policy relevance (an improvement of more than 15 percentage points). As a result, we initially set the total sample size for outcome assessment at 840 pregnant women plus 20% (to allow for loss to follow-up), i.e., 1,008 pregnant women [36]. With the approval of our data safety and monitoring board (DSMB), we allowed enrollment to continue to 1,502 pregnant women to offset unanticipated data missingness that occurred over the December 2018 holiday period. Faculty based at Heidelberg University in Germany performed the stratified randomization of the 84 mentor mothers, eligible for participation, using a computer-generated random allocation sequence. Randomization was stratified by settlement type. Email was used to transfer the allocations to Philani, where they were implemented by senior Philani staff overseeing the mentor mothers. We chose cluster randomization over individual randomization in this trial due to the organically occurring clusters formed by each mentor mother counseling the pregnant women within her neighborhood. This organization of mentor mothers’ work made individual participant randomization logistically challenging, while cluster randomization was both easier and aligned with community practice. Mentor mothers enrolled individual participants on a rolling basis, including checking eligibility criteria and eliciting informed consent. A total of 1,502 women (age 18+ years) participated in our trial. Eighty-four mentor mothers recruited the participants. (Out of 100 mentor mothers working for Philani in the Western Cape at the beginning of this study, 16 were not eligible for study participation because they had been employed by Philani for less than 6 months.) Written, informed consent was collected from all participants by their mentor mother, prior to data collection. Participants were advised that they could exit the trial at any time. We originally intended that both forms of data collection, telephone and face-to-face, would be concluded when the last child of the enrolled participants reached age 5 months. However, we had to terminate face-to-face data collection 3 months prematurely due to the COVID-19 pandemic. Our DSMB allowed us to continue the telephone survey for outcomes data collection until the originally scheduled time point. Eligible pregnant participants were recruited between the 20th and 35th weeks of pregnancy. Fig 2 shows the participant flow diagram for this study. LTFU, loss to follow-up; MM, mentor mother; MOVIE, MObile Video Intervention for Exclusive breastfeeding. A total of 13 short (2 to 5 minutes) teaching videos comprised the Philani MOVIE intervention. We developed this content over a 10-month period, in collaboration with local government health advisors, Philani, and other local maternal–child health advisors. We used a human-centered design approach to tailor the intervention to address many of the specific needs and challenges that were identified by our target community. [36] The videos present learning objectives that are aligned with World Health Organization (WHO) recommendations for infant feeding. Videos were narrated in English and isiXhosa, the languages most commonly spoken among study participants. The primary health and motivational messages were illustrated by a local South African artist and interspersed with narratives from 3 South African celebrities and 4 community mothers. The videos avoided medical jargon, using simple language to convey each health message [49]. Fig 3 provides an overview of the videos, including the titles, durations, and illustrative thumbnails showing one scene from each video. Links to the videos used in the intervention can be found in the supporting information section (S1 File) at the end of this manuscript. MOVIE, MObile Video Intervention for Exclusive breastfeeding. In the intervention arm, mentor mothers delivered the Philani MOVIE videos. The videos were modular rather than sequential, allowing each mentor mother to tailor the order, frequency, and combination of videos used, to meet the individual needs of each pregnant woman and/or mother they supported. All video sequencing decisions were made by the mentor mothers. Philani ensures that the mentor mothers are trained to align their home-based health counseling with the individual needs and circumstances of the mothers they counsel. The intervention mentor mothers were asked to administer each video at least once per participating woman during the trial period. The tablets containing the videos were equipped to track video views. The video intervention mentor mothers were tasked with delivering the video intervention during their regular perinatal home visits, which typically include counseling on infant feeding methods. Our outcome measures were based upon the WHO indicators for the study of infant feeding practices [50] and the most recent available, country-wide infant feeding data for South Africa [4]. All outcome measures pertain to the individual participant. Our primary outcomes were short-term EBF (at 1 month) and long-term EBF (at 5 months), measured using both point-in-time (24-hour recall) and life-long (since birth) data. Prior research has recommended a dual approach to measuring EBF [51]. Primary outcomes data were collected via both face-to-face surveys and independent telephone surveys, as registered at clinicaltrials.gov (#{“type”:”clinical-trial”,”attrs”:{“text”:”NCT03688217″,”term_id”:”NCT03688217″}}NCT03688217). Where data points were missing from telephone surveys, these were replaced with data points from face-to-face surveys. EBF at 1 month is widely used in the literature as an indicator of breastfeeding status [52,53]. We chose 5 months over 6 months as our second data collection point because we anticipated that real-world challenges in our study setting might result in some home visits occurring slightly after the desired date collection time point. Since complementary feeding is recommended from 6 months onward, we feared that the wording of our outcomes survey could yield misleading data from mothers whose babies had recently turned 6 months at the time of data collection and were appropriately receiving complementary foods. Our secondary outcomes included the following: Table 1 summarizes our primary and secondary outcomes for this trial. DHS, Demographic and Health Survey; EBF, exclusive breastfeeding; MOVIE, MObile Video Intervention for Exclusive breastfeeding; WHO, World Health Organization. To ensure that throughout the study period the 2 study arms did not receive differential treatment other than the randomly assigned intervention, all investigators and field team staff were blinded to the impact of intervention assignment on study outcomes, apart from one Stanford-based investigator (JJ). This investigator monitored preliminary intervention impact to report any concerns to the DSMB and to check on data collection integrity between the 2 study arms. The primary analysis was based on intention to treat (ITT) at the level of the individual participant. We used Poisson regression and adjusted standard errors for clustering at the level of the mentor mother. We chose modified Poisson models, because they generate estimates of risk ratios. This approach avoids the interpretational difficulties often associated with odds ratios and converges more easily than alternative approaches, such as negative binomial models, which also generate risk ratios [59]. We used generalized linear models with Gaussian distribution and identity link function for the continuous secondary outcomes (11 and 12—see Table 1). Single variable analyses were carried out for all prespecified and registered outcomes. In the first set of sensitivity analyses, we adjusted our estimates for the following 6 baseline covariates: participant’s number of previous children, participant’s age, running water in the home, electricity in the home, participant’s employment status, and participant’s education level. In the second set of sensitivity analyses, we used multiple imputation to account for missing data. Multiple imputation is commonly used to account for differential loss to follow-up by drawing from a distribution of likely values to replace missing data while adequately accounting for the uncertainty associated with such replacement. In multiple imputation, multiple datasets are created, then analyzed and combined to yield final results [60]. Mentor mother data, telephone surveys, and baseline variables were included in the multiple imputation model to generate a full dataset. We computed the number of multiple imputations required for this sensitivity analysis using the “howmanyimputations” package in Stata [61]. At the conclusion of the trial, we conducted a mixed-methods performance evaluation including quantitative measurements of time usage and in-depth interviews with a subset of 26 mentor mothers (15 from the video intervention group and 11 from the control group) to gain a more nuanced interpretation of the trial results [62]. We asked mentor mothers about their personal experiences integrating tablets into their home visits as well as how they used the video intervention with participants. Maximum-variation purposive sampling was used to select the interview mentor mothers [63], and the interviews were continued until saturation and redundancy were reached [64,65]. Cape Town–based investigator, NJ, who gathered informed consent and conducted the interviews, speaks isiXhosa fluently and is trained in qualitative methods. We conducted weekly debriefings between the investigator conducting the interviews and 2 additional members of the research team, including the study lead [66]. These debriefings allowed the research team to glean early insights into the qualitative data. They also served to support and enhance the interview process for the research associate conducting the interviews, thereby enhancing the overall quality of the data [66]. All interviews were recorded, transcribed, translated into English, and quality controlled for consistency and accuracy. A qualitative analysis of the interview data has been completed using a grounded theory approach. We summarize key findings in this manuscript and report further details in an upcoming publication focused on the qualitative performance evaluation that accompanied this trial. All participating mentor mothers were trained in obtaining written, informed consent, recording baseline variables, and collecting data about participants’ infant feeding practices using their tablets. The mentor mothers in the intervention arm were trained in accessing and showing the video interventions contained within the Digital Medic App. The app could be used offline while collecting usage data on frequency of video views that could then be downloaded upon subsequent connection to the internet. Infant feeding data, collected through the face-to-face surveys, was similarly stored, then downloaded when tablets were reconnected to the internet at Philani. After reconnection, all survey data were automatically transferred to the local research team for cleaning and analysis. Data were collected after the 1 month and 5 months postdelivery time points, first through face-to-face surveys conducted by the mentor mothers, with the majority of surveys collected within 4 weeks after the 1-month and 5-month birthdates. The surveys were translated into isiXhosa, and audio recorded versions of each question were available for mothers who preferred listening to the questions rather than reading them. Mentor mothers recorded survey responses on the tablets using the software, Survey CTO. All 84 mentor mothers carried tablets with the infant feeding surveys throughout the study period. Only tablets provided to the mentor mothers in the intervention arm were preloaded with videos. All mentor mothers received training on the use and care of their tablets, which were Android 8 devices with 16 GB of storage. The face-to-face surveys, completed by participants on tablets, were used to triangulate data points collected through 30-minute telephone surveys, conducted after the 1-month and 5-month mentor mother visits. These telephone surveys were conducted by Social Surveys, a professional telephone research firm in South Africa. The face-to-face surveys were used to impute missing telephone survey data, and both data collection modalities were compared in a subset of participants to verify the data collected. We used only the telephone surveys to administer the knowledge assessment, asking participants to respond to 15 true/false questions related to the learning objectives in the video series. Finally, telephone surveys were used to detect potential contamination (i.e., to confirm directly, in a subset of participants, that the videos had been seen by participants enrolled in the video intervention arm and not seen by participants enrolled in the control arm). Computer-assisted telephone interviewing was used to enhance the accuracy of telephone survey data collection and reporting. All participants’ data were deidentified by local research staff in South Africa and stored in password-protected online storage drives. The theoretical underpinnings of the elaboration likelihood model (ELM) [67] served as a foundation for the intervention tested in this study. The ELM describes 2 main pathways leading to attitude shifts that predict a desired behavioral outcome, like EBF. The first “central route” relies on an individual’s intrinsic motivation and cognitive decision-making, activated by the successful delivery of information. The second “peripheral route” relies on an enhancing transient motivation, influenced by peripheral cues that elicit emotion or identification within the learner. Peripheral cues can elicit temporary attitudinal shifts that support an individual’s intrinsic motivation and likelihood of processing informational messages via the central route [67]. Prior studies have leaned upon ELM, as well as the related extended elaboration likelihood model (eELM), to explain the impact of E–E on health-related attitudes and behaviors [18,21,68]. Fig 4 illustrates the intersection of the ELM and the eELM theoretical models with desired long-term health outcomes. EBF, exclusive breastfeeding; eELM, extended ELM; ELM, Elaboration Likelihood Model. This study was overseen by a DSMB, consisting of a senior health systems researcher at the Norwegian Institute of Public Health and the Medical Research Council of South Africa, a senior biostatistician at the South African Medical Research Council, and a professor of pediatrics and health policy at Stanford University. All members of the DSMB have extensive expertise in health outcomes research. The DSMB met every 6 months throughout the duration of the study. The members of the DSMB helped us to evaluate the study progress, oversaw the study conduct, and were notified when face-to-face data collection needed to be terminated prematurely due to the COVID-19 pandemic. Ethical approval was granted by the Stanford University IRB (Protocol #46667), the University of Stellenbosch IRB (Project ID #6318 HREC/UREC Reference #: N18/02/013), and the Ethics Committee of the Medical Faculty of Heidelberg University (Project #S-706/2018). All 3 committees are recognized Ethical Review Committees. Throughout the study, the investigators respected the principles of ethical research on human participants. Informed consent was obtained from all eligible participants in writing, before data collection began. The protocol [36] for this study was published in April 2019.

• adam, maya
• johnston, jamie
• job, nophiwe
• dronavalli, mithilesh
• le roux, ingrid m.
• mbewu, nokwanele
• mkunqwana, neliswa
• tomlinson, mark
• mcmahon, shannon a.
• lefevre, amnesty e.
• vandormael, alain
• kuhnert, kira leigh
• suri, pooja
• gates, jennifer
• mabaso, bongekile p.
• porwal, aarti
• prober, charles
• bärnighausen, till