Objectives: Patterns of infection among children with varying levels of iron status in a malaria endemic area may vary spatially in ways requiring integrated infection and iron deficiency control programmes. The objective of this secondary analysis was to determine the geospatial factors associated with malaria and non-malaria infection status among young Ghanaian children at the end of a 5-month iron intervention trial. Design: Cluster-randomised controlled trial. Setting: Rural Ghana Participants: 1943 children (6-35 months of age) with geocoded compounds. Interventions: Point-of-use fortification with micronutrient powders containing vitamins and minerals with or without iron. Primary and secondary outcome measures: Generalised linear geostatistical models with a Matern spatial correlation function were used to analyse four infection response variables, defined using different combinations of inflammation (C-reactive protein, CRP >5 mg/L) and malaria parasitaemia. Analyses were also stratified by treatment group to assess the independent effects of the iron intervention. Results: The by-group and combined-group analyses both showed that baseline infection status was the most consistent predictor of endline infection risk, particularly when infection was defined using parasitaemia. In the No-iron group, age above 24 months and weight-for-length z-score at baseline were associated with high CRP at endline. Higher asset score was associated with a 12% decreased odds of endline infection, defined as CRP >5 mg/L and/or parasitaemia (OR 0.88, 95% credible interval 0.78 to 0.98), regardless of group. Maps of the predicted risk and spatial random effects showed a defined low-risk area around the District centre, regardless of how infection was defined. Conclusion: In a clinical trial setting of iron fortification, where all children receive treated bed nets and access to malaria treatment, there may be geographical variation in the risk of infection with distinct high-risk and low-risk areas, particularly around municipal centres.
The data used in these analyses were generated from a study population of young children (6–35 months of age) who participated in a community-based cluster-randomised trial conducted in the Brong-Ahafo Region of Ghana in 2010.15 The trial consisted of a 5-month intervention period, where participants received micronutrient powders with or without iron, followed by a 1-month postintervention follow-up period (6 months in total). At the time of the trial, the estimated prevalence of malaria in Ghana was 7.2 million cases per year, and the prevalence of anaemia among preschool aged children was 76.1% (95% CI 73.9% to 78.2%).16 17 Details of the clinical trial15 and geographical layout of the study area have been described elsewhere (Aimone, Brown, Zlotkin, Cole, Owusu-Agyei 2016). Biological samples collected at the beginning and end of the 5-month intervention period were analysed for plasma ferritin (Spectro Ferritin S-22, Ramco Laboratories, Stafford, USA), CRP (QuickRead CRP, Orion Diagnostica, Espoo, Finland) and malaria parasite density (microscopy).15 Malaria screening was also performed on a weekly basis throughout the intervention period. If a child had a history of fever (within 48 hours) or an axillary temperature >37.5°C, a blood sample was drawn and analysed in the field via antigen test (Paracheck Pf), and in the lab using microscopy (thin and thick smears). Parasite density was combined with fever information to calculate clinical malaria incidence (episode counts). Demographic and nutrition-related information was also collected at baseline and included household assets, maternal education and child body weight and length. Weight-for-length and length-for-age z-scores were calculated using the WHO Child Growth Standards.18 Geographical coordinates for the compounds of 1943 trial participants (representing 1539 clusters), surrounding health facilities and major road networks were collected using handheld global positioning system (GPS) units (WGS 1984 coordinate system, universal transverse Mercator zone 30N projection, EPSG code: 32 630). Satellite-derived data were downloaded as global datasets19–21 and cropped according to the geographical boundaries of the trial area. Elevation had a range of 116–530 m, and values were centred by subtracting 250. Land cover (LC) type consisted of three categorical values, representing woody savannah (LC=8, n=21/1943 observations), urban and built up land (LC=13, n=243/1943 observations), and cropland/natural vegetation mosaic (LC=14, n=1679/1943 observations). Normalized difference vegetation index (NDVI), a vegetation index included as a proxy for moisture,22 was averaged over the year that the trial was conducted (2010), and ranged in value from 0.22 to 0.62. NDVI has also been used to create malaria risk distribution maps through the characterisation of vector habitat potential (eg, closed forest versus open forest).23 Two NDVI-LC interaction terms were created (NDVI*LC8 and NDVI*LC13) by overlaying the final NDVI and LC rasters, and masking the LC cells except where they had a value of 8 (or 13). These unmasked cells were given a value of zero. Baseline age, in months, was calculated using the reported date of birth and trial enrolment date with a change point at 24 months. Household asset score was generated using a principal component analysis of six economic indicators (farm ownership, size and type of crops grown, type of toilet facility, and house ownership). Maternal education was included as a binary variable representing ‘no’ (0) versus ‘any’ (1) level of education. Baseline iron status was defined as serum ferritin concentration corrected for inflammation (baseline CRP) using a regression-based method (Namaste, Rohner, Huang, Bhushan, Flores-Ayala, Kupka, Mei, Rawat, Williams, Raiten, Northrop-Clewes, Suchdev, Adjusting ferrith concentration for inflammation, 2016, unpublished), and re-scaled by multiplying the corrected values by the inter-quartile range. Straight-line (Euclidean) distance to the nearest health facility was measured using the Near Table tool in ArcMap (ArcGIS V.10.2, Environmental Systems Resource Institute, Redlands, California, USA). The endline data were analysed using generalized linear geostatistical models (GLGM)24 25 with Bayesian inference via an Integrated Nested Laplace Approximation algorithm.26 Weak or uninformative priors were used for all model parameters with the exception of the Matern shape parameter (fixed at two). Four different endline infection outcomes were modelled separately: 1) inflammation (CRP >5 mg/L) with/without malaria parasitaemia; 2) inflammation (CRP >5 mg/L) without parasitaemia; 3) parasitaemia with measured concurrent fever (axillary temperature >37.5°C) or reported history of fever within 48 hours (ie, clinical malaria)) and 4) parasitaemia with or without concurrent fever or history of fever. All variables were binary-valued (coded as ‘1’ for positive infection status at endline) and analysed using logistic regression, with the exception of the third definition (parasitaemia with fever), which was a count variable (number of new clinical malaria episodes during the intervention period) analysed using Poisson regression. Median infection probabilities (and 95% credible intervals, CrIs) were modelled as the sum of the contributions of the independent variables, residual spatial variation, and a compound-level random effect term. The glgm function from the ‘geostatsp’ package in R was used for all spatial modelling.27 28 The spatial analyses were conducted in three modelling steps: 1) No-iron group only; 2) Iron group only; 3) both intervention groups combined. The interventions groups were analysed separately in order to differentiate the effects of time and the iron treatment. Infection probabilities from the combined-group models were plotted on a base map of the trial area, with study compounds, and major road networks. The maps depicted a spatial risk surface of predicted infection probabilities, which were computed as the posterior means of the odds or risk of infection. The posterior means were estimated assuming baseline values for individual-level covariates and location-specific values for spatial covariates. The posterior mean of the spatial random effect was also plotted, representing the residual spatial variation that corresponded to the difference between the predicted and expected odds or risk of infection at each location. The original clinical trial was approved by the Kintampo Health Research Centre (KHRC) Institutional Ethics Committee, the Ghana Health Service (GHS) Ethical Review Committee, the Hospital for Sick Children Research Ethics Board and the Food and Drugs Authority of Ghana. Approval for conducting the secondary analyses was obtained from the Hospital for Sick Children and University of Toronto Health Sciences Research Ethics Boards.