Stigma as a barrier to treatment for child acute malnutrition in Marsabit County, Kenya

Acute malnutrition affects millions of children each year, yet global coverage of life-saving treatment through the community-based management of acute malnutrition (CMAM) is estimated to be below 15%. We investigated the potential role of stigma as a barrier to accessing CMAM. We surveyed caregivers bringing children to rural health facilities in Marsabit County, Kenya, divided into three strata based on the mid-upper arm circumference of the child: normal status (n=327), moderate acute malnutrition (MAM, n=241) and severe acute malnutrition (SAM, n=143). We used multilevel mixed effects logistic regression to estimate the odds of reporting shame as a barrier to accessing health care. We found that the most common barriers to accessing child health care were those known to be universally problematic: women’s time and labour constraints. These constituted the top five most frequently reported barriers regardless of child acute malnutrition status. In contrast, the odds of reporting shame as a barrier were 3.64 (confidence interval: 1.66-8.03, P<0.05) times higher in caregivers of MAM and SAM children relative to those of normal children. We conclude that stigma is an under-recognized barrier to accessing CMAM and may constrain programme coverage. In light of the large gap in coverage of CMAM, there is an urgent need to understand the sources of acute malnutrition-associated stigma and adopt effective means of de-stigmatization. Marsabit County, Kenya, covers approximately 26 000 square miles and is divided into six administrative divisions. It is a geographically remote and arid region in North Kenya, home to approximately 290 000 people and subject to drought and food insecurity. Nomadic pastoralism is the dominant livelihood and is practiced by the dominant ethnic groups in the region: the Borana, the Gabra, the Rendille, and the Burji. The Borana and Gabra share a common language, the Borana language, which is widely spoken across the region. Political and ethnic tension, primarily between the Borana and Gabra groups, dates back for decades, originating in disputes over land and livestock ownership and water use. Violent clashes over grazing land and civil disputes in the town of Moyale, on the border of Marsabit County and Ethiopia, were ongoing throughout 2013 and during the time of this study (Kenya Humanitarian Response 2013). Rates of wasting (>12%) and stunting (>27%) among children under the age of 5 in Marsabit County are among the highest in the country (Wambua 2013). This study was conducted in three administrative districts of Marsabit County. Districts were selected if Concern Worldwide, an international non‐profit humanitarian organization based in Dublin, Ireland, supported their child health and nutrition programming and if they had active CMAM participants in June of 2013. Eighteen health facilities were randomly selected for participation in this study as sites for data collection, six from each study district. We randomly selected three from above and below the 50th percentile of CMAM admissions2 according to June 2013 records for each district. We used this stratification approach to ensure even representation of population densities and to include facilities with a range of potential accessibility issues. Eligible participants were adult women (≥18 years) who accompanied a child aged 6–59 months at a study facility between August and September 2013 and gave informed consent to participate in the study. Women accompanying children <6 months were not included, as CMAM protocol for children of this age was not established at the time of the study. From the pool of eligible women, study participants were systematically and purposively selected into one of three study subgroups based on their child's MUAC. Women whose children had SAM (MUAC <115 mm or oedema) were eligible for the SAM group. Those whose children had MAM (MUAC <125 mm, ≥115 mm) were eligible for the MAM group and those with children with a normal MUAC (MUAC ≥125 mm) were eligible for the Normal group. Group selection was based on MUAC and oedema only and did not discriminate between other indicators of child health or nutrition status or the purpose of the clinic visit (e.g. women in any of the three groups could have been present for a well child visit or treatment for illness). The minimum target sample size for this study was 144 individuals per group or 432 individuals in total. This target was calculated to detect a 20% difference in the frequency of a single access barrier between any two subgroups with 80% power, alpha 0.05 and 95% confidence. The sample size was adjusted to account for correlation among individuals from a single facility on a single day using a design effect factor of 0.05. We did not have prior data with which to estimate the design effect; our chosen value of 0.05 value reflects our assumption that participants at the same facility were only 5% more likely to have correlating values than participants selected at random. We designed our sampling strategy knowing that it might exceed the target sample size. Due to uncertainty around CMAM admissions during the study time period, we needed to design a strategy that would reach the target sample size but not over‐sample the early weeks of the study. We also expected admissions at some facilities to be lower than in others, in which case low‐admissions facilities would need a longer period of time to reach their target sample size. To attain sampling breadth across time as well as ensure that the target sample would be reached at each clinic, we chose to operate the study for a pre‐specified period of time (5 weeks) regardless of whether the overall target sample size had already been reached. Four women were recruited per group (SAM, MAM, Normal) per facility per day for four ‘study days’ spread across 6 weeks (August–September 2013). Study days were pre‐selected by health facility staff to coincide with CMAM activities. On each of the four study days at each study facility, every second eligible woman was systematically selected for the MAM and the Normal groups, beginning with the first eligible woman for each group and continuing with every other woman until four women in each group had been selected. This approach was used to limit oversampling of participants early in the day and to reduce the burden on enumerators. As a result of the limited number of women eligible for the SAM group, every eligible woman for that group was selected on each study day. Enumerators were community health workers working under the supervision of one designated study nurse at each study clinic. Enumerators and study nurses were selected based on the their prior experience with data collection, a positive work record, their designation as the primary individuals working at the health facilities selected for the study and written and spoken proficiency in English and Borana. The survey took approximately 40 min and was administered at study facilities after participants had completed their clinic visits. The survey consisted of 40 closed or open‐ended questions, presented in the English or Borana languages, as preferred by the mother. All questions were originally written in English, translated to Borana by a professional translator and then back translated to English during enumerator training before the survey was finalized. The survey was pilot tested by enumerators among patients and families at the Marsabit County Hospital, allowing for corrections and adjustments to question wording and order prior to finalization. Child MUAC, weight and height were obtained from the child's health card using the measurements recorded by the study nurse during their clinic visit that day. MUAC was assessed using a standard MUAC tape on the left arm. Length of children less than 87 cm long was measured using a baby mat. Height of children greater than 87 cm was measured using a wooden or plastic stadiometer. Child age was recorded from the child's health card (n = 595, 84%) or estimated by the 116 mothers (16%) whose child did not have a health card. Mothers reported their own age and marital status (married monogamous, married polygamous, single, separated, divorced, widowed), their formal education level (none, primary, secondary, higher), ethnicity, the residential status of their household (resident, refugee, visitor), the number of people living in their household and the number under the age of 5 years and the primary source of income in their household. Food security was assessed using a condensed version of the Household Food Insecurity Access Scale (Swindale & Bilinsky 2006) that queried whether any of the following five events occurred within the previous 4 weeks: inability to eat healthy and nutritious foods, worry about running out of food, not having any food to eat in the household, going to sleep at night hungry due to lack of food and going a day and night without eating anything due to lack of food. We asked participants about the presence or absence of 17 locally relevant household items to assess non‐agricultural assets (lantern, cart, hoe or axe, clock or watch, radio, television, car battery, electricity, generator, tape player, bicycle, cell phone, tin roof, latrine, running water and mosquito net) in addition to recording the number and type of livestock owned by the household. Participants reported their primary reason for being at the clinic on that day, whether their child was ill, and any symptoms they noticed in their child. They also answered questions about care‐seeking behavior including choice of provider and typical time to care. Barriers to care were assessed through open‐ended questions. Participants were asked to freely list any barriers, challenges or obstacles that they had faced, either in the past or on the day of the study, in accessing health care for a child. They were then asked to pick what they considered the three most important issues, which were recorded by enumerators. A pre‐specified list of responses was available to enumerators to aid in data collection, but this list was not made available or indicated to participants. It included a pre‐coded response for shame using the Borana term fokifade amale cherfad; this phrasing also means shyness or embarrassment and was the preferred term for identifying the concept of stigma. Responses that were not pre‐specified were written separately and coded by JRB later. All participants were also asked directly about the concepts of stigma, regardless of whether they had freely listed shame as a barrier to accessing care. Consensus among survey staff was that using the terms ‘stigma’ or ‘shame’ in direct questions could alienate or insult participants and should be changed to ‘discomfort’. Thus all participants were asked about their level of comfort with their purpose at the clinic that day. This question was worded to elicit emotional discomfort rather than physical discomfort; enumerators were given additional prompts such as ‘worried’ and ‘unhappy’ to assist if the question was not received as intended. The Borana term used for discomfort was dansa indagau. To get a sense of social norms and perceptions of acute malnutrition, we also asked whether participants would expect women whose children were wasted to feel uncomfortable coming to the clinic and reasons why this might be the case. Answers were pre‐coded; unique responses were noted by enumerators and coded post hoc by JRB later. The Cornell University Institutional Review Board for Human Participants granted ethical approval for this study (protocol #1306003948). In Kenya, approval was granted by the Nairobi Nutrition Information Working Group and the Marsabit County Ministry of Health. All participants gave verbal informed consent and were free to exit the survey at any point. Data were assembled and analysed using STATA 12 (StataCorp 2013). Descriptive statistics (means, medians, frequencies) were calculated for the caregiver, child and household characteristics as appropriate, and for responses regarding perceptions of acute malnutrition. We used Fisher's exact tests to identify differences in frequencies across the three study groups. We created a continuous count index of household food insecurity, ranging from zero (no measures of food insecurity indicated) to five (all five measures of food insecurity indicated). We also created a continuous count index of household assets. Of the 17 asset variables measured, eight were reported by at least 10% of the sample. We used these eight variables to create a scale ranging from zero (none of the assets indicated) to eight (all eight assets indicated) to assess household wealth. Access barriers were categorized using an adapted version of a framework proposed by Barton (Streatfield et al. 2008; Barton 2010). We used multilevel mixed effects logistic regression to estimate the odds of reporting a barrier given the acute malnutrition status of the child using the responses to our open‐ended questions about access barriers. The 10 most commonly reported barriers were used as binary outcome variables in 10 separate models, each adjusted for the same set of covariates. We chose 11 covariates that we expected to be potentially confounding because of their association with exposure (acute malnutrition) and outcome (shame), as well as covariates that we expected to be significant predictors of at least one of the 10 outcome barriers. Covariates comprised child age (months) and sex (male = 0, female = 1), maternal age (years), marital status (single, separated, divorced or widowed = 0, married = 1), education (0 = no formal education, 1 = any formal education), household distance from the clinic (minutes), household size, food insecurity (0 to 5), physical household assets (0 to 8), district, and facility caseload (low = 0, high = 1). Health facility was treated as a random effect. We used a dummy variable to test for the association between barriers and study group. The threshold for statistical significance for all variables was P < 0.05.