The software after considering the population size, estimated

The nutritional status data used for
this study were obtained from food security and nutrition analysis unit
(FSNAU). Food Security
and Nutrition Unit (FSNAU) is a unit in World Food Programme of United Nations
(UNFAO), which was set up in 1994 to provide evidence-based analysis of Somali
food, nutrition and livelihood security to enable both short-term emergency
responses and long- term strategic planning. Therefore, in partnership with
UNICEF, FSNAU has been conducting bi-annual seasonal nutrition assessment
surveys since 2001. Our study focuses on survey data ranging from 2007 –
2010Figure 1. Within this period, FSNAU, in partnership with United Nations
Children’s Emergency Fund (UNICEF), conducted bi-annual seasonal nutrition
assessment surveys using standard methods, indicators and tools for data
In each survey, a stratified multi-stage cluster sampling design was
adopted where the sampling frame of a selected district was based on the four
livelihood definitions (pastoral, agro-pastoral, riverine and fishing) within
which 30 communities and 30 households within each village were selected using
systematic  random sampling method
and the urban population were
clearly defined and considered separately. Vulnerable groups that could not be
classified in any of the livelihood such as IDPs were surveyed separately. Respective
samples sizes (number of households and number of children) were calculated
using the Epiinfo/Ena 2008 software after considering the population size,
estimated prevalence and desired precision. A list of all villages and
population within each of the assessed livelihoods served as a sampling frame
and was used to construct cumulative population for the assessment area. Villages
were then randomly selected from these villages with the chance of any village
being selected being proportional to the size of its population. This is called
sampling with “probability proportional to population size (PPS)”. Selection
of households within the village was done using systematic random sampling,
preferably from a list of eligible names or a map of households. Where these
were not available, the number of households in the village was estimated from
the population figures (the total population divided by the mean household
size). Starting from a random household, every nth household was selected and
all eligible children (aged 6-59) in that household measured (Figure SI 1).
Retrospective mortality data was collected from all the households in each village
from each livelihood, including even those that did not have children aged 6-59
months. At the individual child level, age, gender, weight, height, mid-upper
arm circumference (MUAC), vitamin A supplementation in the last six months,
diarrhoea, acute respiratory infections (ARI) and febrile illness in the two
weeks before the survey, and Polio and Measles vaccination history were
collected. At the household level, information recorded included the household
size and age structure, gender of the household head, and access to different
types of foods in the last 24 hours. Data on falciparum malaria infection in children aged 5-59 months were
collected in sub-sets of villages at the request of UNICEF29–31. The data used in this study were
therefore a subset of the whole survey dataset with information on both the
childhood malnutrition and malaria.


We considered two outcome measurements
to describe the anthropometric indicators of malnutrition, low
weight-for-height (wasting) and low-MUAC, which detect different sets of children
as malnourished. Wasting is traditionally the main indicator in community
surveys. Although MUAC is a better predictor of mortality32, few studies have examined associations
between MUAC and specific pathogens.


A child was defined as wasted when
s/he was below -2 Z scores for weight-for-height, according to World Health
Organization (WHO) 2006 standards33. A child with MUAC below 125mm was
classified as having low-MUAC. Malaria parasitaemia was determined using
Paracheck Pf™ (Orchid Biomedical Systems, Goa, India) in a subset of the sample
in every FSNAU surveys during this period33. A child was regarded as malaria infected
when s/he had a positive Paracheck Pf™ test result, regardless of any clinical


A detailed search were undertaken to
establish a set of spatial coordinates for each village in Somalia using the village
names in the data. The location of village was verified by using Google Earth
(Google, Seattle, USA) and other online databases to visually inspect whether
the coordinates matched evidence of human settlement. Those settlements for
which no reliable source of the coordinates was obtained were excluded from the


Environmental data

A set of four plausible environmental
covariates, together with wasting, low-MUAC and malaria in children were
included in the analysis18,34. These were rainfall, enhanced
vegetation index (EVI), mean temperature, and urbanization. Rainfall and mean
temperature were derived from the monthly average grid surfaces obtained from
WorldClim database35. The EVI values were derived from the
MODerate-resolution Imaging Spectroradiometer (MODIS) sensor imagery for period
2007-2010 while the urbanization information was obtained from Global Rural
Urban Mapping Project (GRUMP)36,37. All the environmental covariates
were extracted from 1 x 1 km spatial resolution grids. Rainfall, temperature
and EVI were summarized to compute seasonal averages corresponding to the time
of survey.


Statistical methods

overall aim of this study was to model the ecological comorbidity of wasting
and low-MUAC with malaria parasitaemia among children aged 6-59 month in
Somalia from 2007 – 2010. To achieve this, we implemented the Bayesian
geostatistical shared component model through stochastic partial differential
equation (SPDE) approach in integrated nested Laplace approximations (INLA)
using R-INLA library in R project version 3.2.3. 22–24,38,39. Therefore, we modelled two
underlying spatial risks common to: (1) wasting and malaria and (2) low-MUAC and
malaria at child level. The relative risk of each condition depends on a latent
spatial component shared by each pair and a condition-specific component after
controlling for environmental covariates40,41. The household survey and
environmental predictors of malnutrition and malaria were controlled at
individual, household and village level.


Finally, to determine
if the risks were correlated, we performed a significant test by looking at the
2.5% and 97.5% quantiles of each element of the random effect using the
quintile correction (QC) method as implemented by Bolin and Lindgren 201242.  Further, the empirical correlation between
the conditions were explored using correlation plots.