Study area, design, period and population
The study was conducted in Bussi Islands which are located in Wakiso district of Uganda. There are five Islands that make up Bussi Islands. The Islands have an approximate catchment population of 10,000 people, they are completely surrounded by Lake Victoria and are located approximately 41 kilometres (24.6 miles) south of Kampala City (8, 9).
A cross-sectional survey that used quantitative methods of data collection from the community was conducted in Bussi Islands from the 13th to the 18th of April 2019. The study respondents were child caretakers in households which had at least one child aged 6 to 59 months as the study population were children aged 6 to 59 months in Bussi Islands.
An adjusted sample size of 409 households with at least a child aged 6 to 59 was estimated using the Kish Leslie formula (10). The prevalence of stunting in rural areas of Uganda which was 30% (5), a 95% confidence interval and a maximum acceptable error of 5% were used to estimate the sample size. A design effect of 1.2 and a 5% potential non response are the adjustments that were performed.
Inclusion and exclusion criteria
Households in Bussi Islands that had at least a child aged 6 to 59 months were eligible for inclusion. Households in which child caretakers were unavailable after two follow up visits and those households that had children who were eligible for inclusion but had physical body deformities that would have interfered with anthropometric assessment results.
The sample was got from three out of the five Islands that make up Bussi Islands. With the help of the local leaders, all households with at least one child aged 6 to 59 months were identified from which the sample households were randomly selected using simple random sampling method. Sample selection was done proportionate to the number of households with children aged 6 to 59 months in each Island. Where there was more than one child aged 6 to 59 months in the household, balloting method was used to select the child to take part in the study.
Outcome variables: This was malnutrition inform of stunting and underweight. The basic information and anthropometric measurements that were used to assess for stunting and underweight in children aged 6 to 59 months were: age, sex, weight and height/length. Following the WHO recommendations, the decision to measure height or length depended on age and physical condition of the child. Height was measured for children who were aged ≥24 months and length for children who were aged <24 month. In cases where the exact age of the child was not known, height was measured for children ≥87 cm and length was measured for children <87 cm. In cases where a child ≥24 months of age was too sick to stand, length of the child was measured and 0.7 cm were subtracted from it. Weight for Height/Length (WFH/L) Z-scores were used to identify stunting and a child was stunted when the WFH/LZ-scores were < -2 Standard Deviations (SD). Weight for Age (WFA) Z-scores were used to identify underweight and a child was underweight when the WFAZ-scores were < - 2 SD (11, 12).
Exposure variables: These were the independent variables and they included:
(a) Dietary intake and child characteristics: Indicated by introduction of solid, semi-solid or soft foods, feeding method, breast feeding status, child meal frequency, age, sex, delivery place, delivery attendant, birth type and birth order.
(b) Child health status characteristics: Indicated by immunisation; sickness (diarrhoea, fever, respiratory diseases and measles); Vitamin A supplementation and deworming statuses.
Household food security status characteristics
a) Food access; indicated by the modified HFIAS levels, household dietary diversity scores (HDDS), food consumption scores (FCS), household members earning an income and household food source.
b) Food availability: indicated by agricultural land access; crop growth and purpose; fishing and fish purpose; food stocks; livestock ownership and livestock purpose.
Health system characteristics; indicated by transport mode to health facility, ownership of transport means, transport cost to health facility.
WASH characteristics; indicated by drinking water source, amount of water used in household per day, waste disposal method, child's hand washing before feeding and drinking water treatment method.
Socio-demographic and economic characteristics: indicated by caretaker age, sex, type, marital status and education; maternal parity, pregnancy and breastfeeding status (where the caretaker was the biological mother); income source of household head; household members; children aged < 60 months in households and household lighting type.
A face to face interview method was used to collect data on child dietary intake factors, child health status factors, child characteristics, household food security status (food access and food availability) factors, health system factors and WASH factors as well as socio-demographic/economic factors. The Anthropometric assessment data was obtained by measuring the child’s height/length, weight, MUAC, assessing for bilateral pitting oedema and recording of the date of birth.
A structured questionnaire written in English was developed into an ODK software form on mobile phones and was used to collect data from the households. A translated version of the questionnaire in Luganda (local language) was also availed to the research assistants.
The questionnaire was pre-coded at the time of proposal development and the codes were used in the development of the ODK software questionnaire forms. A fully compiled dataset in form of an excel sheet from an online server where the ODK forms were uploaded at the time of data collection was sent to the principal investigator. The anthropometric data was first entered into ENA SMART version 2011 where nutrition status classification according to the 2006 WHO growth standards was done. It was then joined to the rest of the data in an excel sheet, after which the excel sheet was exported to STATA version 14 for cleaning and subsequent descriptive and statistical analyses.
Univariate, bivariate and multivariate analyses of the results was done and the results were presented using summary tables and narrative texts.
Objective I: To assess the prevalence of stunting and underweight among children aged 6 to 59 months in Bussi Islands.
The child anthropometric indices were analysed by classifying stunting and underweight according to 2006 WHO child growth standards using the ENA SMART software. The prevalence of stunting and underweight was then got by dividing the number of children who were stunted and underweight respectively with the total number of children in the study expressed as a percentage.
Objective II: To assess the determinants of stunting and underweight among children aged 6 to 59 months in Bussi Islands.
All characteristics in this study that are known from literature to be associated with stunting and underweight were analysed as categorical variables but means and their corresponding standard deviations were also provided in the text for those characteristics that were collected as continuous variables. Frequencies and percentages were also used.
The measure of association between the outcomes of interest (stunting and underweight) with the independent variables were the Prevalence Ratios (PRs) at their 95% confidence intervals and p-values of <0.05 showed statistically significant associations between the outcomes and the independent variables. The PRs were used instead of Odds Ratios (ORs) because the prevalence rates of stunting and underweight were more than 10%. The ORs tend to overestimate the strength of association in such scenarios (13, 14).
PRs at both the bivariate and multivariate analysis level were estimated using the Modified Poisson regression analysis model, with robust standard errors, via generalized linear models with family (Poisson) and link (log) (15). After testing for co-linearity, covariates with p-value ≤0.2 at the bivariate analysis were considered for the multivariable model.
At multivariable level, modified Poisson regression was done to identify adjusted estimates for the determinants of stunting and underweight. Variables, which when included resulted in loss of significance were removed. The final model selection was based on Akaike Information Criteria (AIC), with smaller AIC value suggesting a better model. Covariates with p-values < 0.05 after multivariable analysis were considered as determinants of stunting and underweight.