Evaluation of conditional cash transfers and mHealth audio messaging in reduction of risk factors for childhood malnutrition in internally displaced persons camps in Somalia: A 2 × 2 factorial cluster-randomised controlled trial

Background AU Cash: Pleaseconfirmthatallheadinglevelsarerepresentedcorrectly transfer programmes are increasingly used in humanitarian : contexts to help address people’s needs across multiple sectors. However, their impact on the key objectives of reducing malnutrition and excess mortality remains unclear. mHealth interventions show great promise in many areas of public health, but evidence for their impact on reducing the risk factors for malnutrition is uncertain. We therefore implemented a trial to determine the impacts of 2 interventions in a protracted humanitarian context, a cash transfer conditionality and mHealth audio messages. Methods and findings A 2 × 2 factorial cluster-randomised trial was implemented in camps for internally displaced people (IDP) near Mogadishu, Somalia, starting in January 2019. The main study outcomes were assessed at midline and endline and included coverage of measles vaccination and the pentavalent immunisation series, timely vaccination, caregiver’s health knowledge, and child diet diversity. Twenty-three clusters (camps) were randomised to receive or not receive conditional cash transfers (CCTs) and an mHealth intervention, and 1,430 households were followed up over 9 months. All camps received cash transfers made at emergency humanitarian level (US$70/household/month) for 3 months followed by a further 6 months at a safety net level (US$35). To be eligible to receive cash, households in camps receiving CCT were required to take their children <5 years age to attend a single health screening at a local clinic and were issued with a home-based child health record card. Participants in camps receiving the mHealth intervention were asked (but not required) to listen to a series of audio messages about health and nutrition that were broadcast to their mobile phone twice a week for 9 months. Participants and investigators were not blinded. Adherence to both interventions was monitored monthly and found to be high (>85%). We conducted intention-to-treat analysis. During the humanitarian intervention phase, the CCT improved coverage of measles vaccination (MCV1) from 39.2% to 77.5% (aOR 11.7, 95% CI [5.2, 26.1]; p < 0.001) and completion of the pentavalent series from 44.2% to 77.5% (aOR 8.9, 95% CI [2.6, 29.8]; p = < 0.001). By the end of the safety net phase, coverage remained elevated from baseline at 82.2% and 86.8%, respectively (aOR 28.2, 95% CI [13.9, 57.0]; p < 0.001 and aOR 33.8, 95% CI [11.0, 103.4]; p < 0.001). However, adherence to timely vaccination did not improve. There was no change in the incidence of mortality, acute malnutrition, diarrhoea, or measles infection over the 9 months of follow-up. compiledforthoseusedthroughoutthetext Although there was no evidence :Pleaseverifythatallentriesarecorrect that mHealth increased : Mother’s knowledge score (aOR 1.32, 95% CI [0.25, 7.11]; p = 0.746) household dietary diversity increased from a mean of 7.0 to 9.4 (aOR 3.75, 95% CI [2.04, 6.88]; p < 0.001). However, this was not reflected by a significant increase in child diet diversity score, which changed from 3.19 to 3.63 (aOR 2.1, 95% CI [1.0, 4.6]; p = 0.05). The intervention did not improve measles vaccination, pentavalent series completion, or timely vaccination, and there was no change in the incidence of acute malnutrition, diarrhoea, measles infection, exclusive breastfeeding, or child mortality. No significant interactions between the interventions were found. Study limitations included the limited time available to develop and test the mHealth audio messages and the necessity to conduct multiple statistical tests due to the complexity of the study design. Conclusions A carefully designed conditionality can help achieve important public health benefits in humanitarian cash transfer programmes by substantially increasing the uptake of child vaccination services and, potentially, other life-saving interventions. While mHealth audio messages increased household diet diversity, they failed to achieve any reductions in child morbidity, malnutrition, or mortality. The Ministry of Health & Human Services of the Federal Republic of Somalia (reference: MOH&HS/DGO/0993/Jun/2018) and the Research Ethics Committee of UCL (project ID: 4684/001) granted ethical approval. The study was registered on November 5, 2018, with ISRCTN (ISRCTN24757827). This study is reported as per the Consolidated Standards of Reporting Trials (CONSORT) guideline (S1 CONSORT Checklist). A consort checklist for this paper is included in the Supporting information (S1 CONSORT Checklist). The trial took place in the Afgooye Corridor, in peri-urban Mogadishu, Somalia. The Afgooye Corridor hosts the largest IDP camps in Mogadishu, the cumulative result of multiple protracted crises in the country over the years. The IDP camps are spontaneous and privately run settlements, often managed by informal settlement managers. The camps are often overcrowded, lack basic sanitation and health services, and residents face recurrent evictions. The IDP living in these camps subsist mostly on humanitarian assistance from relief agencies and from casual labour when available. All study participants were enrolled in a cash transfer programme and received the same level of transfers. We tested 2 separate interventions that were delivered as part of this cash transfer programme: (1) a cash transfer conditionality, where household registration for the cash transfer programme, was conditional on a one-time health screening of their children aged 0 to 59 months; and (2) an mHealth BCC intervention, where we transmitted prerecorded audio messages to the mobile phones of people registered for the cash transfer programme. The cash transfer programme consisted of 3 monthly electronic transfers of US$70.00 (henceforth humanitarian cash transfer), followed by 6 monthly electronic transfers of US$35.00 (henceforth safety net cash transfer). The humanitarian cash was aligned to the recommended transfer value from the Somali Cash Working Group, which reflected 80% of the multipurpose (i.e., full) minimum expenditure basket (MEB) for the given location. The safety net cash transfer value followed standard practice of the implementing agency at the time of intervention design, i.e., half the full humanitarian value. Whereas the higher value was distributed at times of peak hunger, the safety net value was lower so as to reach as many households in need as possible while supporting household income, promote positive coping strategies, and reduce risk of dependence. Upon registration, a household female representative received a mobile phone SIM card with a unique number, which was recorded and through which they received the cash transfer via a mobile money transfer made by the telecommunications company we contracted (Hormuud Telecom Somalia). The cash transfer programme targeted women as the household cash recipient on the assumption that their spending was more likely to benefit their children. Cash transfers were made on March 24, April 27, June 3, July 4, July 27, August 25, September 29, October 27, and December 8, 2019. The conditionality intervention, henceforth the conditional cash transfer (CCT) intervention, was a one-time requirement for all children aged 0 to 59 months to attend the local clinic at the beginning of the study, where health staff screened their health and issued them with a CINS health record card. The child was offered any necessary vaccinations, but agreement to receive vaccination was not necessary to meet the conditionality. Children born after registration for the cash transfer had taken place were not required to attend. Households in the CCT intervention arm were actively encouraged to meet the conditionality, which all did. If households had failed to meet the conditionality, they would have been denied registration for the cash transfer programme. During registration for the cash transfer programme, we verified whether households in the CCT arm met the conditionality by asking the mothers/caregiver to show the child CINS health card. Additional checks of the child’s CINS health card were carried out monthly by study staff and community health workers (CHWs) to encourage any newly arrived children, in the households of the CCT arm, to be health screened and receive their CINS health card. The health record card was designed for the study based on recommendations from the World Health Organisation (WHO), the existing MoH Child Health booklet, and experience in Malawi, where passport-sized, home-based health record cards had been developed [21]. The card was designed to be able to fit in a pocket and had a water-resistant cover. It received some limited pretesting with volunteer Concern staff before being printed in Mogadishu. Study health record cards were only issued to children living in the CCT intervention arm camps. All households in the CCT and unconditional cash transfer (UCT) arms had similar access to the local health clinic. In addition, CHWs visited all households in all the study arms each month to review the vaccination status of their children, aged 0 to 59 months, and to remind their caregivers of any due child vaccinations. The mHealth intervention consisted of a series of 30 pairs of health-promoting audio messages. Each pair of messages comprised a 2-minute long drama and a 1-minute long reenforcing message that were transmitted on separate days, usually on Friday for the drama and Sunday for the reenforcement, to the recipient’s mobile phone. The message series comprised 5 seasons, each of 6 pairs of messages and transmission of the whole series took place over 8 months (May to December 2019). The drama component was a serialised fictional story, following a family similar to those in the study camps. The content of the message series covered 6 topics: (1) vaccination; (2) water, sanitation, and hygiene (WASH); (3) infant and young child feeding (IYCF) practices; (4) identifying signs of serious illness and seeking care; (5) recognition, treatment, and prevention of acute malnutrition; and (6) maximizing health and nutrition for all household members. We identified these 6 topics as important barriers to optimal health and nutrition during the implementation of the REFANI-S study. [7] We designed the message content based on information gathered during a formative human-centred design workshop held in Mogadishu with Concern staff, CHWs, and mothers and fathers from the IDP community. The design process was also informed by the theoretical behaviour change approach used in a previous mHealth project in Bangladesh [22]. We developed the messages in collaboration with 2 Somali companies that specialised in interactive voice response (IVR) technology (Shaqodoon Organisation, Hargeisa, Somaliland) and creative message development (Media INK, Mogadishu, Somalia). Messages were recorded in Mahrati and Maay, the 2 Somali languages widely spoken in the IDP camps, and recipients received broadcasts in their preferred language. We implemented a 2 × 2 factorial cluster-randomised controlled trial, where IDP camps were the units of randomisation. We surveyed 23 clusters between January 17 and February 12, 2019. We used 3 criteria to select these clusters: first, the clusters should not be receiving any cash assistance from another organisation; second, the clusters needed to be located between 0.3 and 1 km straight-line distance from the Concern Worldwide–operated health centre, where we planned to undertake the health screening of children; and third, clusters needed to be of comparable size. There were no exclusion criteria for individuals participating in the study except that the household was required to have a child aged 0 to 59 months. Informed verbal consent was obtained from camp leaders in all IDP camps before starting data collection. In addition, following a detailed explanation of the study objectives and data collection process, informed verbal and written consent was obtained from caregivers at the household. Study participants were informed about their right to withdraw from the study and that participation or withdrawal from the study would not affect their entitlement to humanitarian assistance. Confidentiality and data security of the respondents was ensured by the use of password-protected encrypted devices and the removal of any information from datasets that could be used to identify respondents or their location. We planned to include 20 clusters in the study with an average of 65 children, aged 0 to 59 months, per cluster. However, the surveyed clusters had fewer children than expected (average 54.3 children, range: 11 to 102), so we included all 23 surveyed clusters in the study, and we used the Round 1 (R1) data that we collected as our baseline. We then randomly allocated half the 23 clusters to the CCT intervention resulting in 12 clusters assigned to this intervention and 11 to the control (see Fig 1). The allocation was done by first listing the clusters in order of size (number of households) and then randomly assigning the CCT intervention or control to alternate members of the list, using a random number, generated using the Excel RANDBETWEEN() function, to determine the allocation for the first cluster. The process was then repeated in an identical way to randomly allocate the mHealth intervention, separately, to the CCT control and intervention groups. The randomisation of interventions was performed in London by the study PI. As with most interventions of this nature, participants could not be “blinded” to allocation status. Within each cluster, all households with children aged <5 years (n = 774) were eligible to be included in the study, together with all the children aged <5 years and their mothers/caregivers that were members of these households. CCT, conditional cash transfer; HH, household; R, data collection round; UCT, unconditional cash transfer. Following the baseline survey and cluster randomisation, the cash distribution and mHealth interventions started. We then implemented 8 additional rounds of data collection from the study households, beginning on April 29, 2019 (see Fig 1). The “midline” data collection took place at R3 in June (3 months after the start of the intervention), and the “endline” took place at R9 in December (9 months after the start of the intervention). In all 9 rounds of data collection, the sample of households and mothers/caregivers remained the same (a closed household cohort). However, we included in the study all the children aged <5 years that joined the household after baseline. We collected information at the household and individual level, i.e., mother/caregiver and child (see S1 Table). We gathered data from household heads and mothers/primary carers regarding household demographics, food security, expenditure, morbidity, WASH, breastfeeding practices, mother/caregiver’s knowledge on health, and nutrition and vaccination programme coverage. We also collected mid-upper arm circumference (MUAC) measurements and assessed bipedal pitting oedema (henceforth oedema). We identified children as acutely malnourished if they presented with a MUAC <12.5 cm and/or oedema and referred them to the health centre for treatment. We identified all children <5 years old who died between R1 and R9 data collections and arranged a verbal autopsy (VA) interview with their mothers/caregivers. Data were collected by the research field teams using a structured questionnaire, translated into Somali, on mobile Android devices running Open Data Kit (ODK). Data collection at baseline (month 0), midline (after 3 months of intervention), and endline (after 9 months of intervention) was conducted by study enumerators, who were hired specifically for the data collection exercises. Data were collected at the monthly time points in between 0, 3, and 9 months by CHWs. Prior to starting data collection, we implemented 1-week training sessions for enumerators and supervisors at R1, R3, and R9 data collection rounds. During these training sessions, we piloted and revised questionnaires. The R1, R2, and R3 data collection was undertaken by 10 teams. Each team had an enumerator and 2 CHWs. The rest of data collection rounds were conducted by 10 teams of CHWs, with each team having 2 members. Teams in all rounds were supervised by 4 nutrition supervisors, under the overall supervision of a field study coordinator. During each round of data collection, if we could not reach a household after repeated attempts or any of its members were absent, we enquired about the reasons for this absence from family or neighbours. To understand household composition, we collected data on the numbers of members and their ages. We then estimated total, child, and aged dependency ratios as the ratio between the number of members aged <15 years or aged ≥65 years, members aged <15 years, and members aged ≥65 years, each divided by the number of members aged 15 to 64 years, respectively. Additionally, we obtained information on the sex of the head of household, and whether they lived in a monogamous or polygamous arrangement. Using a questionnaire and through observation, we obtained data on whether households had access to piped water, whether they pay for this access, and the time that it took them to collect water. We also obtained data on household access to handwashing facilities, soap availability, and the type of toilet. We classified households as engaging in open defecation if their members reported to relieve themselves in open fields or if their household showed no evidence, reported or observed, of having access to latrines. We used household and child dietary diversity score (DDS) and the reduced coping strategies index (rCSI) to assess food security. We applied a 24-hour food recall to a checklist of 12 food groups for estimating the household DDS, as recommended by the Food and Agriculture Organisation of the United Nations [23]. The household DDS food groups were as follows: (1) cereals; (2) white tubers and roots; (3) vegetables; (4) fruits; (5) meat; (6) eggs; (7) fish and seafood; (8) pulses, nuts, and seeds; (9) dairy products; (10) oils and fats; (11) sweets; and (12) spices, condiments, and beverages. The household DDS has a range of 0 to 12. Similarly, we applied a 24-hour food recall to a checklist of 7 food groups to estimate child DDS, as recommended by WHO [24]. The child DDS food groups were as follows: (1) grains, roots and tubers; (2) legumes and nuts; (3) dairy products; (4) flesh foods; (5) eggs; (6) vitamin A-rich fruits and vegetables; and (7) other fruits and vegetables. The child DDS has a range of 0 to 7. The rCSI is a simple tool applied in different contexts that assesses the frequency, in days within a 7-day period, and the severity of 5 coping strategies commonly used by households, when they cannot access enough food [25]. The 5 coping strategies are consuming less preferred foods, borrowing food, reducing meals, reducing portion sizes, and restricting adult’s food consumption to preserve children’s food consumption. As per recommendations, we weighted and summed the frequency responses to these strategies to create an index where higher scores indicate greater food insecurity. The rCSI range is 0 to 56. Rations of prepared food were distributed in this setting, so we asked how many days, within a 7-day period, the household relied on these wet rations. We asked how many meals were consumed in the household in the past 24-hour period. We collected 30-day and 4-month household expenditure from a list of 10 food groups and 27 nonfood items. We then standardised all expenditure figures to a 30-day period and added them to calculate household total, food, and nonfood expenditure. We asked mothers/caregivers their age in completed years, if they could read or not, if they had engaged in paid labour in the past year, and if they slept under a mosquito net in the past night. We assessed knowledge on health and nutrition topics using a list of 20 questions/statements covering 5 topics, for which respondents needed to answer yes or no to indicate whether they agreed with the statement (see S4 Table). Knowledge on health topics and nutrition was defined as the number of correctly answered questions. We collected data on vaccination coverage, based on the Somalia Expanded Programme on Immunisation (EPI) schedule (see S2 Table), by checking the child’s health card at baseline, midline, and endline, or by mothers/caregivers recall if no health card was available or there was no entry for a particular antigen [26]. Bacillus Calmette–Guérin (BCG) vaccination status was assessed by the presence of a scar. Adverse events were not specifically recorded for trial purposes as there was no plausible reason to suspect that an increase in these would be associated with the intervention. During the trial, there was no indication of an increase in adverse vaccination events from routine health facility reports. We ascertained the children’s month and year of birth using a calendar of local events. We then randomly selected a day of the month to generate the date of birth. We estimated age in months as the difference between the day of data collection and the date of birth divided by 30.4. We measured MUAC on the left arm to 1 mm precision using a standard MUAC insertion tape. We measured children’s MUAC in duplicate and entered both measures into the ODK questionnaire, which we designed to automatically estimate difference between these measurements and their mean. If the estimated difference was greater than 5 mm, we obtained a third measurement and estimated the mean using the pair of values with the lowest difference. We recorded presence of bilateral pitting oedema if an imprint remained in both feet after pressing them with the thumbs for 3 seconds. We defined acute malnutrition as the presence of a MUAC <12.5 cm and/or bilateral pitting oedema. We assessed exclusive breastfeeding in children aged <6 months using the WHO tool for IYCF practices [24] at R1, R3, and R9 visits. The WHO tool collects a 24-hour breastfeeding recall, as described by the child’s mother/caregiver, and information of liquids and food intake. We defined exclusive breastfeeding as reporting consumption of breastmilk in the past 24 hour while reporting no consumption of any other food or liquid. We asked mother/caregivers whether the child had been unwell in the last 4 weeks or since the last round of data collection. We defined all-cause morbidity in children as having been reported unwell. Starting at the second round of data collection for children identified as having been unwell, we asked whether was due to diarrhoea or measles. We defined diarrhoea if the mother/caregiver reported the child had 3 or more loose or watery stools per day, and measles if the child had fever and maculopapular rash, with cough or coryza or conjunctivitis. For all the children, <5 years old, that died in the period between R1 and R9, we arranged for VA interviews, by trained staff, with their mothers or caregivers that reported having been with them in the period preceding death. VA interviews are designed to collect cause-of-death information—signs and symptoms—from deaths that occurred outside of health facilities and where death certification is weak or absent [27]. We adapted and shortened the 2016 WHO VA tool by removing nonrelevant questions, correcting skip patterns, and simplifying language to facilitate use by interviewers with no medical training. We calculated the probability of different causes of death using InterVA-5 software, which uses Bayesian reasoning to calculate the probability of cause of death categories based on the reported presence of signs, symptoms, and the circumstances of death [20]. For each death, InterVA-5 software reports the probability of up to 3 of the most likely causes. If the sum of these probabilities was less than 100, we allocated the remnant to an “indeterminate” category. We created a category called “lost to follow-up,” defined as the sum of the deaths lost to follow-up with VA interviews, each given a value of 100. We compiled and added all reported likelihoods for these causes of death to estimate the population level burden and estimated the population cause-specific mortality fractions (CSMFs) by dividing the population-level burden by the total number of deaths. The primary outcomes of the trial were assessed at the midline and endline time points. For the CCT intervention, these were measles vaccination coverage, defined as the proportion of children aged 9 to 59 months who had received MCV1, and “EPI vaccination coverage,” defined as the proportion of children aged <60 months who had received all their vaccinations according to the EPI schedule and their age (see Table 3). For the mHealth intervention, 2 additional primary outcomes were assessed: mother/caregiver’s knowledge of health and nutrition topics, defined as the mean number of correctly answered knowledge’s questions (see S1 Table), and dietary diversity of children aged 6 to 23 months, defined as the mean child DDS. 1CCT, conditional cash transfer. Following trial registration with ISRCTN, the name of the “EPI vaccination coverage” outcome was changed to “Timely Vaccination” to align better with terminology used in the literature, and “Pentavalent series completion,” defined as the proportion of children aged 12 to 23 months that had received all 3 does of Pentavalent vaccine, was added as an additional indicator of vaccination coverage. The secondary outcomes of the trial were as follows: acute malnutrition incidence, assessed as a single failure; mortality incidence; all-cause morbidity incidence, assessed as a single failure; exclusive breastfeeding in children, defined as the proportion of exclusively breastfed children aged <6 months; and the CSMF obtained from VA interviews. Other outcomes of the trial included the following: all-cause morbidity, assessed as multiple failure; diarrhoea incidence, assessed as a single and multiple failure; measles incidence, assessed as a single failure; total, food, and nonfood household expenditure; and the food security indicators household diet diversity (HDDS) and the rCSI. We estimated the sample size to observe differences in measles vaccination coverage and child DDS using the user-written clustersampsi function in Stata (StataCorp. 2015. Release 14. College Station, TX: StataCorp LP) [28]. As this function is designed for use with large numbers of clusters and because we did not include the Holm–Bonferroni adjustment for multiplicity, the calculated power was overestimated. We hypothesised that a greater measles vaccination coverage would be the results of the CCT intervention and that a greater child DDS would be the behavioural result of the mHealth intervention. Based on prior data [7], we expected a baseline measles vaccination coverage of around 64%. With 10 clusters per arm, with an average population of 65 children (aged <5 years) per cluster, and allowing for an alpha risk of 0.05 and an ICC of 0.06, we calculated we would have 80% power to detect a minimum increase of 18 percentage points in measles vaccination coverage. This minimum detectable difference laid below our hypothesised effect size of 20 percentage points. In addition, based on prior data [7], we expected a baseline child DDS of 3 with a standard deviation of 2. With 10 clusters per arm with an average population of 65 children (aged <5 years) per cluster, and allowing for an alpha risk of 0.05 and an ICC of 0.01, we calculated we would have 80% power to detect a minimum increase of 0.42 in child DDS. We undertook data analysis using Stata v17. Prevalence estimates at baseline were computed accounting for the trial cluster design using the Stata complex survey svy commands and differences in means and proportions were tested for using the lincom command. For variables collected at R1, R3, and R9 (baseline, midline, and endline), we assessed impact using mixed effects, multilevel, regression models: linear, logistic, or ordered logistic, depending on whether the outcome variables were continuous, categorical, or ordinal, respectively. Models included cluster and household variables as random-effects and dummy variables denoting the time point (baseline, midline, or endline) and intervention (CCT and mHealth) as fixed-effects in the model. All models included baseline data to account for any observed differences at baseline and were adjusted for age and sex. We conducted an available case analysis using mixed models, and the analysis is valid under the missing-at-random assumption. We used robust variance for calculation of standard errors. We tested for interactions between the 2 interventions and found no evidence of a significant effect for any outcome. We therefore report only the main effects in the results and provide a table of effects for the combined interventions in the annex. The absolute difference in risk for the vaccination coverage outcomes were calculated as the difference in regression adjusted contrasts, using the margins post-estimation command. The absolute difference in child diet diversity and maternal/caregiver knowledge scores was calculated using mixed effects linear regression. To provide a description of how the interventions effected the vaccination for each antigen, we also calculated the arithmetic difference-in-difference (DiD), by subtracting the baseline prevalence from the midline and endline values, and then subtracting the control prevalence from the intervention prevalence at each time point. For assessing the impact on incidence, we estimated hazard ratios using Cox proportional hazards models, adjusting for child’s age and sex as covariates in the models, and using svy to calculate cluster-robust standard errors. For comparing probable causes of death, we used chi-squared tests, after we stratified the CSMF by arm. All analysis of intervention effects is “intention to treat,” meaning we compare clusters as randomised irrespective of whether participants received the intended intervention. P values were adjusted for multiple comparisons using the Holm–Bonferroni step-down procedure [29], but all confidence intervals are calculated at the 95% level. The adjustments were made based on all 72 tests that were conducted to test the impacts of the 2 interventions on the primary and secondary outcomes.