Ethics

The Ministry of Health and Human Services of the Federal Republic of Somalia (reference: MOH & HS /DGO/0469/August/2015) and the Research Ethics Committee of University College London (project ID: 1822/003) granted ethical approval. The study was registered on 19 January 2016 with ISRCTN (ISRCTN29521514).

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 in the participating households. Study participants were informed about their right to withdraw from the study and that participation in or withdrawal from the study would not affect their entitlement to humanitarian assistance. Confidentiality and the data security of the respondents were ensured by the removal of any information from datasets that could be used to identify respondents or their location.

Study setting

The study was conducted in IDP camps located in Weydow area, Deyniile district, Mogadishu. Deyniile and Dharkenley, 2 of the 17 districts of Banaadir region, host the majority of IDPs in Mogadishu, who are primarily from marginalised tribes or minority groups [18]. Concern Worldwide has been implementing multi-sector development and humanitarian assistance programmes in Somalia since 1986.

The IDP camps in Weydow area are spontaneous and privately run settlements that are often overcrowded and lack basic sanitation and health services, and residents face recurrent evictions. Morbidity estimates (diarrhoea, pneumonia, and fever) in these IDP camps are high and are considered a major driver of high estimates of acute malnutrition [6]. During the wet seasons (Gu and Deyr, April–June and October–December, respectively), morbidity estimates, primarily from diarrhoea, increase [19]. Most of the IDPs were previously agro-pastoralists and riverine farmers who lived in the Bay, Bakool, and Shabelle regions [18]. Now, their primary livelihood sources are casual labour, petty trading, and humanitarian assistance received from local and international humanitarian organisations [19].

Study design

Details of the study design have been published elsewhere [20]. The study was a non-randomised cluster trial, where IDP camps were the cluster units. Ten clusters were first selected for the CBI based on vulnerability criteria, following a routine needs assessment exercise conducted by Concern Worldwide. The routine needs assessment ranked vulnerability based on whether (1) households were headed by elderly people, children, women, or people with disabilities, (2) there were orphans in the household, (3) all household members had no income sources, (4) households had children that had recently recovered from acute malnutrition, (5) households had high dependency ratios, (6) households had pregnant and/or lactating women, and (7) the camp had been relocated recently following eviction from a previous location. In our study, a recent eviction was the primary criterion for camp selection. The intervention was not randomised in our study because the available funds were allocated based on need using the vulnerability criteria listed above. The number of clusters selected for intervention was determined by the availability of funds. All the households in the selected intervention clusters were registered for the CBI.

Afterwards, the control clusters were selected, which included 10 IDP camps located adjacent to the intervention clusters. These control clusters were not targeted to receive the intervention and were known to not have received CBIs recently. The control clusters contained an average of 236 households (range 70–433 households) and the intervention clusters had an average of 113 households (range 48–215 households).

Intervention

The CBI took place in 2016 and comprised 3 elements: (1) a monthly unconditional cash transfer of US$84.00/month for 5 months, (2) a once-only distribution of a non-food-items kit, and (3) the provision of piped water free of charge, through tap stands.

The monthly cash amount was based on the cost of the Minimum Expenditure Basket developed by the Food and Agriculture Organization’s Food Security and Nutrition Analysis Unit–Somalia. The Minimum Expenditure Basket represented a minimum set of basic food items such as sorghum, vegetable oil, and sugar meeting the 2,100 kilocalories/person/day basic energy requirement for a household of 6–7 members, and non-food items such as water, kerosene, firewood, soap, and cereal grinding costs. Upon registration, a household female representative received a mobile phone SIM card with a unique number through which they received the cash transfer via mobile money transfer made by the contracted telecommunications company (Hormuud Telecom Somalia). The CBI targeted women as the household cash recipients on the assumption that their spending was more likely to benefit their children [21]. Cash transfers were done on May 30, June 25, July 25, August 25, and September 29, 2016.

The non-food-items kit comprised 1 plastic sheet, 2 mosquito nets, 1 blanket, 1 sleeping mat, 1 kitchen set, 1 bar of soap, 2 collapsible jerry cans, and 1 set of sanitary pads. The non-food-items kits were distributed in January and February 2016.

In addition to providing the CBI, Concern Worldwide provided support and equipment to the local maternal and child health and nutrition centres, where malnourished children and pregnant and lactating women in the study area could access treatment. In addition, Concern Worldwide and other non-governmental organisations supported building of improved pit latrines in most of the IDP camps included in both arms of the study.

Study components and participants

The study consisted of 3 components for data collection: a household cohort, a child cohort, and a process evaluation.

Sample size

We performed sample size calculations using Stata (Release 14; StataCorp). To assess a change in malnutrition risk factors in the household cohort, we estimated a sample size able to detect a difference of 0.6 units in the mean Child Dietary Diversity Score (Child DDS). A positive change of about 2 units in Household Dietary Diversity Score (Household DDS) was observed in response to a CBI in Malawi [23], but other studies observed smaller changes [24,25]. Scarce data exist on Child DDS change following cash transfers, so we decided to power the study to detect a small change, assuming a baseline Child DDS of 3.0 with a SD of 1.5, a power of 80%, an alpha risk at 0.05, and an intra-cluster correlation coefficient of 0.01. The resulting sample size estimate was 120 children per arm. Based on previous nutrition survey data, we assumed an average of 1.2 children aged 6–59 months per household. Based on previous survey experience, we added 20% to allow for non-response. This resulted in a required total sample of 240 households.

To assess acute malnutrition incidence in the child cohort and detect a hazard ratio of 0.5, we assumed, based on operational guidance for caseload estimations [26], that the proportion of followed-up children who would develop a MUAC below 12.5 cm and/or oedema during a 6-month period was 7% and that there would be 20% loss to follow-up. Power was set at 80% and the alpha risk at 0.05. This resulted in a total sample size needed of 1,167 children, which equated to a rounded sample of 600 participants per study arm.

Data collection and data handling

For the household cohort, we collected data using a structured questionnaire, translated into the local Somali language, on mobile devices running a proprietary version of Android Open Data Kit (PSI Mobile). The collected data comprised information at the household and individual level.

For the child cohort, we collected data using paper forms. These paper forms were later digitally captured, and data entry errors identified and corrected.

To ensure collection of high-quality data in the household and child cohorts, a 2-week training was implemented for enumerators and supervisors prior to the start of data collection. During this training, we piloted questionnaires. Details regarding key validation studies of the tools used, as well as information on their adaptation to the local context, are provided in S1 Table. All teams were supervised during field data collection. During home visits, if a child could not be found after repeated attempts, the reasons for absence from the household were ascertained from family or neighbours.

Primary and secondary outcomes

The primary outcome measures were (1) mean Child DDS values of children aged 6–59 months, assessed using data from the household cohort, and (2) incidence of acute malnutrition in children aged 6–59 months, as defined by a low MUAC and/or oedema, assessed using data from the child cohort.

The secondary outcome measures, assessed using data from the household cohort, were (1) prevalence of acute malnutrition in children aged 6–59 months, as defined by a low WHZ and/or oedema; (2) mean WHZ value in children aged 6–59 months; (3) mean 30-day household expenditure; (4) mean Household DDS; (5) mean FCS; (6) mean HFIAS score; (7) mean rCSI score; and (8) mean Women DDS.

Data analysis

Data analysis was done using Stata. For the household cohort, prevalence and mean estimates at baseline and endline were computed using the survey (svy) command, which accounts for the clustering of values. We estimated the difference-in-differences (DiD) between the study arms as the arithmetic difference that resulted from subtracting the difference (endline minus baseline) in the control arm from the difference in the intervention arm. We tested, using a Student t test, whether the DiD was significantly different from 0 using the lincom command. To calculate confidence intervals for 0% or 100% proportions, we used the Newcombe–Wilson method without continuity correction [36]. We did not perform adjustment for multiple comparisons, but we opted to describe in full the statistical tests performed, as recommended [37].

As requested by PLOS Medicine reviewers, we also obtained the Child DDS DiD between the study arms using linear regression and ordered logistic regression, before and after adjustment for other variables that were found to be significantly different at baseline.

For the child cohort, we performed analysis of the incidence of acute malnutrition in the trial arms using Cox proportional hazards analysis. Survival time before the first episode of acute malnutrition was measured in days and calculated for everyone using the dates of household visits. When children were not found during a household visit, and were reported to be absent or dead, it was assumed that this reported status had existed for the entire period since the previous visit, and this time was not included in their calculated survival time. Measurements from children who died were treated as right censored data. The incidence of acute malnutrition was calculated using the ir command. The stset and stcox commands were used to define survival time and run the hazard model, and assumptions of proportionality were tested using the estat phtest command. For calculation of standard errors and statistical testing, the svy linearized: stcox command was used to account for clustering.

Intervention implementation

Records obtained from the telecommunications company showed that cash transfers took place as scheduled, with 1,313 beneficiary households receiving around US$420.00 in total each. The amount received each month varied between US$83.00 and US$85.00 due to fluctuating market prices.

Participant flow

Fig 1 presents the participant flow of the household cohort. Overall, loss to follow-up at endline was 5%, 8%, and 16% for households, women, and children, respectively. Loss to follow-up was greater in the intervention arm. Fig 2 presents the participant flow of the child cohort. At baseline, 199/2,337 (8.5%) children presented with acute malnutrition and were therefore excluded from the analysis of incidence (111 [7.4%] and 88 [10.4%] from the control and intervention arms, respectively); oedema was not observed.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Household cohort participant flow diagram.

https://doi.org/10.1371/journal.pmed.1002684.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Child cohort participant flow diagram.

https://doi.org/10.1371/journal.pmed.1002684.g002

Household cohort baseline characteristics

Table 1 shows the baseline characteristics. Overall, the households were reported to be mainly monogamous and female headed. Most households reported access to piped water that took them about an hour to collect, and few had hand-washing facilities in the household. Food utilisation indicators showed that households had a mean dietary diversity score above the mid-range value, and most had an acceptable FCS. However, food security indicators assessing the experience of food insecurity also showed that most households experienced severe food insecurity as most households engaged in coping strategies (S2 Table). Reported household expenditure was, on average, greater than reported income, with food accounting for a large proportion of that expenditure. The largest food expenditure item was cereals (S3 Table), whilst the largest non-food expenditures were on firewood and cooking fuel, drinking water, and transportation.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Baseline characteristics of the household cohort.

https://doi.org/10.1371/journal.pmed.1002684.t001

At baseline, we found 3 main differences between the study arms (Table 1). First, the intervention arm appeared to be less poor and more food secure than the control arm. Households in the intervention arm reported greater income; spent proportionally more on food; presented greater dietary diversity scores for the household, women, and children; presented lower rCSI scores, and reported relying on less preferred and less expensive foods for a lower number of days (S2 Table). These households also reported spending more on meat, fish and eggs, dairy products, fats and oils, sugary products, and transportation (S3 Table). However, households and children in the intervention arm also reported consumption of fewer meals in the past 24 hours. Second, the intervention arm appeared to have more adults caring for dependants, as households in the intervention arm reported having fewer members away and presented lower household dependency ratios. Finally, the intervention arm reported a greater prevalence of measles vaccination. We did not observe other significant differences at baseline. S4 Table describes some of the baseline differences between arms associated to the vulnerability criteria used for targeting of the CBI. Unfortunately, the original data used for the humanitarian vulnerability assessment were not available for assessing differences at baseline.

Household cohort Child Dietary Diversity Scores

Table 2 presents the differences observed between baseline and endline for each study arm and the DiD between these arms. We observed a significant increase in Child DDS of 0.53 (95% CI 0.01; 1.05). S5 Table presents the results of the linear regression and ordered logistic regression, requested by PLOS Medicine reviewers. A positive increase in Child DDS at endline was significant for the intervention in the unadjusted analysis in both linear and ordered logistic regression. This positive association continued after adjustment for other variables found to be significantly different at baseline along with child’s age.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Difference between endline and baseline for each study arm and the difference-in-differences.

https://doi.org/10.1371/journal.pmed.1002684.t002

Child cohort acute malnutrition incidence

Details of the child cohort at each round of measurement are presented in S6 Table. Data on the incidence of acute malnutrition and mortality are shown in Table 3. Overall, the incidence of acute malnutrition and mortality appeared lower in the intervention arm, but the unadjusted (0.83, 95% CI 0.48; 1.42) and adjusted hazard ratios (0.94, 95% CI 0.51; 1.74) were not significant, indicating no protective effect of the intervention. Forty-six children (2.2%) received hospital treatment, with no difference between study arms.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Effect of the intervention on time to first episode of acute malnutrition or death.

https://doi.org/10.1371/journal.pmed.1002684.t003

In addition to the intention-to-treat analysis, we analysed the incidence data including only the 61.3% of children in the intervention clusters whose mother or primary carer had been confirmed as a recipient of the cash transfer using the transfer list from the telecommunications company. The analysis yielded similar results (see S7 Table).

Household cohort difference-in-differences

In Table 2 we observed 3 patterns of change associated with the intervention, as indicated by a significant DiD. First, water, sanitation, and hygiene indicators improved, with an elimination of payment for water and an increase in soap presence in households. Second, household wealth improved, with significant increases in household income, total expenditure, and non-food expenditure, but without an increase in food expenditure. Lastly, food security improved, with increases in most household food security indicators, dietary diversity for women, and meal frequency for children. No significant DiD values were observed for children’s anthropometric data, although both arms showed significant improvements. In addition, we observed a reduction of reported health problems in children that was significant only in the control arm, but with no significant DiD between arms. Interestingly, the improvements in dietary diversity appeared to be higher in women than in children, although this difference was not tested. The DiD values for dietary diversity for households, women, and children represented a relative increase from baseline values in the intervention arm of 14%, 38% and 17%, respectively.

Household cohort coping strategies

Table 4 presents details on the 5 coping strategies making up the rCSI. The proportion of households that employed each of the 5 coping strategies was significantly reduced in the intervention arm compared to the control arm. Similarly, the average number of days each coping strategy was employed was significantly decreased among those receiving the intervention, compared to the control arm, for most coping strategies, except for reliance on less preferred and less expensive foods. Of the 5 coping strategies investigated, adults’ restricting food consumption in order for small children to eat and reducing the number of meals eaten in a day showed the largest decreases in the intervention arm.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Difference between endline and baseline for each study arm and the difference-in-differences for coping strategies.

https://doi.org/10.1371/journal.pmed.1002684.t004

Household cohort household expenditure

Table 5 presents the DiD values between the trial arms for 30-day household expenditures for food and non-food items. Overall, the results show that households receiving the intervention significantly increased their expenditure, primarily on non-food items. Regarding food items, households receiving the intervention significantly increased their expenditure only on dairy products. Households increased their non-food expenditure on cooking fuel, health, clothing, debt repayments, and housing, whilst significantly decreasing their expenditure on drinking water, presumably due to the inclusion of piped water free of charge in the intervention. The greatest increases in expenditure were for clothing and debt repayments.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. Difference between endline and baseline for each study arm and the difference-in-differences for 30-day household expenditure (in US dollars).

https://doi.org/10.1371/journal.pmed.1002684.t005

Summary of results

To our knowledge, this is the first study evaluating the nutritional impact associated with a humanitarian CBI on IDP children in Somalia. We found that after households received this CBI, household wealth and food security and children’s dietary diversity appeared to improve significantly. However, despite these improvements, we did not find evidence that this intervention was associated with an improvement in children’s nutritional status or with a reduced risk of developing acute malnutrition.

The impact of humanitarian CBIs on the nutritional status of children has also been studied, with mixed results, in other contexts, including Niger [17,38,39], Pakistan [40], Burkina Faso [41,42], and Congo [43]. However, direct comparisons are difficult because of contextual differences in the humanitarian emergency (e.g., seasonal hunger versus forced displacement); the modality, timing, and duration of the interventions; and the public health, livelihoods, and market environment in each study setting.

Food security improvements

Improvement of household wealth and food security has often been observed following CBIs in different settings [44], as observed in our study and other humanitarian settings. Compared to control households that did not receive cash transfers, households receiving cash in other studies significantly increased their individual dietary diversity and reported increased consumption of nutritious food, animal protein, and iron-rich foods [40,41]. Similarly, in our study we observed apparent increases in household and individual dietary diversity (in mothers/primary carers and children) and increases in the reported consumption of more nutritious food groups such as dairy products. Furthermore, we observed changes across different food security indicators that reflected improvements in food utilisation and the experience of greater food security in the intervention arm. Interestingly, the increases in dietary diversity observed suggested that cash might have increased dietary diversity more among mothers/primary carers than among children. Although we lacked a large enough sample size to be able to assess further the relationship between the dietary diversity changes of women and children, this observed difference agrees with other changes in the reported household coping strategies. That is, the greatest change in coping strategies observed in households receiving the CBI was reduction in the restriction of food consumption by adults in order for children to eat, suggesting the possibility that children’s access to food was already prioritised in households prior to receiving the CBI.

The CBI did not appear to reduce the risk of acute malnutrition

Our finding that a humanitarian CBI appeared to increase food security but was not associated with a reduced risk of developing acute malnutrition in children was also observed in a study in Burkina Faso and in one of the intervention arms in a study in Pakistan [40,42]. Conversely, improvements in the nutritional status of children were observed in a study in Congo and in another intervention arm in Pakistan study [40,43]. Among the possible reasons for the lack of nutritional impact in our study are that the cash transfer might not have been sufficiently large, that the cash transfer may not have been used in the most nutritionally optimal way because of potentially limited market choices, and that a positive nutritional impact might be observable only among those who are nutritionally vulnerable.

In the abovementioned study in Pakistan, a reduced prevalence of acute malnutrition was observed only in the arm receiving double the amount of money that the standard CBI recipient received (US$28.00 versus US$14.00 per month); the amounts were designed to cover 20% and 10% of the energy requirements of a typical household, respectively [16]. However, in our study the cash transfer was designed to cover 100% of the energy requirements of the typical household [20], but despite this relatively larger transfer, we observed no associated impact on nutritional status.

Targeting of cash transfers to individuals who are nutritionally vulnerable or providing specific nutritious foods together with cash transfers has been shown to reduce the risk of acute malnutrition in children. In Niger, children of families that received specific nutritious foods plus cash transfers showed a lower risk of acute malnutrition compared to those receiving cash transfers only [39]. Authors of that study suggested that this combination ensured that vulnerable children would consume foods that are more nutritious, as the risk of sharing the specific nutritious foods with other household members would be reduced compared to if the nutritious foods were provided without the cash transfer. In Congo, children affected by severe acute malnutrition showed a lower incidence of relapse when receiving cash transfers during outpatient therapeutic programme care and after discharge, for a total of up to 6 months [43]. In our study, no additional specific nutritious food was provided, and the additional components of the CBI were designed to improve water access and shelter infrastructure (plastic sheets) and to reduce the likelihood of malaria transmission (mosquito nets). Whether the addition of specific nutritious foods to the CBI, or providing social and behaviour change communication messages to influence expenditure decisions, could result in a lower acute malnutrition risk for children in this setting will need to be explored in future studies. Similarly, whether changing the way vulnerable households are selected for a CBI intervention in this context, such as using severe acute malnutrition cases as a primary household vulnerability criterion, might result in greater nutritional impact, although the perverse incentive potential of this criterion requires further assessment.

How CBIs in humanitarian settings might affect disease burden remains to be understood. In Pakistan, households in the intervention arms receiving cash transfers (the standard and the double cash) saw a significant reduction in reported fever/malaria compared to the control group (no cash), but only the arm receiving double the amount of money saw a significant reduction of reported respiratory infections. In contrast, in our study we observed a significant reduction in reported health problems in both arms, but no differences between these arms, despite clear evidence of a significant improvement of soap availability and the removal of the economic costs of accessing piped water in households receiving the CBI.

It should be noted that data collection for this study ended a few months before the onset of the 2017 Somalia drought and health emergency. Although evidence of this humanitarian emergency was evident as early as November 2016, large increases in acute malnutrition and mortality did not materialise until early 2017 [5,8]. Therefore, our findings represent the apparent impact of a humanitarian CBI on a chronically vulnerable, relatively stable population, rather than on a population undergoing an acute shock. The generalisability of our findings to other humanitarian contexts therefore needs to be carefully considered, as it is likely that a CBI would yield different results during periods of higher nutritional stress, such as during a drought.

Strengths and limitations

There are limitations to our study, associated with the conflict and insecurity in this context, which made reducing risks to the study team and the beneficiary population paramount. First, the intervention was not randomised because it was allocated based on need, reducing our ability to guard against allocation bias and confounding. In addition, the data from the humanitarian vulnerability assessment used to allocate the intervention was not available for analysis. However, our results indicate that although the intervention group appeared to be wealthier and more food secure, most indicators suggested good matching between intervention and control groups. Second, to minimise the field presence of data collection teams, we recruited and followed a small cohort of households to assess food security indicators, but the sample was not adequate to assess nutritional impact or to study the relationship between changes in nutrition status and food security. This small sample also prevented us from assessing the relationship between morbidity and nutritional status and from collecting data on other important factors such as food preparation hygiene. Nonetheless, the large improvement in household food security indicators observed in the intervention group during this brief period provides a strong indication that, in this context, improving food security alone might be insufficient to improve the nutritional status of children. Third, due to logistics and security considerations, for the child cohort we did not collect other anthropometric measurements apart from MUAC. This prevented us from assessing the impact of the intervention on acute malnutrition, as defined by WHZ or stunting. Fourth, we did not collect information on the types of nutritious foods that were in the local markets, which could have affected the likelihood of the nutritional status of children being improved via cash transfers. Lastly, the observed acute malnutrition incidence was lower than expected (3.6% observed versus 7% expected), reducing our ability to detect significant differences.

There are also strengths to this study. To our knowledge, this is the first study to assess the associated nutritional impact of this type of humanitarian intervention in a population exposed to conflict and recent displacement. Other studies assessing the impact of humanitarian CBIs on nutritional outcomes have been undertaken on populations exposed to a seasonal hunger gap [17,38,39,41,42]. Our study provides evidence on what changes could be observed in populations exposed to conflict and recent displacement. In addition, this study used a varied selection of food security indicators, which allowed for a wider understanding of possible changes and aid interpretation. That most indicators in the intervention arm changed in a similar fashion provided us with greater confidence of the veracity of our findings.