Study design and settings
This was a cross-sectional study conducted in the districts of Iganga, Luuka and Buyende in eastern Uganda. This region is predominantly rural, cover an area of 3,549.8 km², and have an estimated population of 1,065,284 inhabitants living in 208,030 households [13]. These districts are served by at least 75 government-run health facilities and several private not for profit (PNFP) health centres [14]. Malaria, which is mostly attributable to Plasmodium falciparum, is endemic in this area. The main economic activity in these districts is subsistence farming, but other occupations include small-scale businesses, such as fishing, grain milling, market vending, motorcycle transport and formal employment. The Basoga, a Bantu-speaking group, are the predominant ethnic group, which make up to 9% of Uganda’s population [14].
Study domain, eligibility and sampling
The study units were households, and the study domain included women who had delivered in the last 12 months prior to the start of the study and were resident in the area. Mothers were included in the study whether the child was delivered preterm or full-term, and irrespective of the birth outcome (whether the baby was alive or dead). Those who had not lived in the community for at least 1 year were excluded from the study.
Data were collected from 2,062 mothers in three health sub-districts (HSDs): Buyende, Luuka, and Iganga. Sixteen (16) sub-counties (6 in Buyende, 6 in Luuka, and 4 in Iganga) were proportionately selected from the HSDs. The sub-counties in each HSD were randomly selected and within each sub-county, one parish was randomly selected. Two villages were randomly selected from each parish, and a list of households with mothers who met the criteria were listed. Participants were sampled at the village level using simple random sampling from the village listing made with the aid of local council 1 (village) leader. From each selected village, at least 50 households were visited by the enumerators from which one eligible respondent was selected per household.
Data collection and study variables
This study utilized secondary data from a broader study entitled “Innovations for increasing access to integrated safe delivery, PMTCT and newborn care in Rural Uganda”. An interviewer-administered structured questionnaire developed based on the literature on the uptake of IPTp-SP and ITNs among pregnant women was used to collect quantitative data. The original English questionnaire was translated to Lusoga, the local language spoken by the study participants. Data were collected on socio-demographic characteristics, uptake of IPTp-SP, ITNs, and frequency of ANC visits. Research assistants were trained on appropriate methods of data collection, and the tool appropriately piloted. The primary outcome variables of the study were consistent ITN use and optimal uptake of IPTp-SP which were self-reported. Consistent ITN use was defined as sleeping under an ITN every night for the full duration of the last pregnancy, while optimal uptake of IPTp-SP was defined as 3 or more doses received during pregnancy. The covariates (independent variables) included the timing of first ANC, number of ANC visits, sociodemographic characteristics (such as maternal age, marital status, level of education of women, occupation, household size, parity, and wealth (measured using a wealth asset index). The wealth quintiles were generated using principal component analysis based on the information collected on assets owned and household structure. The covariates used for this study were selected from critical review of related published literature [11, 12].
Statistical analysis
Data were analysed using Stata Version 14.0 (StataCorp, Texas, US). Descriptive statistics such as frequencies and percentages were used for categorical data, while means and standard deviations were used where data were continuous. The associations between the outcome variables (consistent ITN use and uptake of 3 or more IPTp-SP doses) and explanatory variables were explored using modified Poisson regression. Initially, unadjusted prevalence ratios (PRs) were obtained for the association between each outcome and each predictor variable. Prevalence ratios were preferred over odds ratios since odds ratios would overestimate the effect size when outcomes are common (prevalence > 10%) [15, 16], as was the case in the current study. All epidemiologically meaningful independent variables were considered for a fully saturated model. Interactions between predictor variables and the primary outcomes were as well examined. A stepwise backward elimination method was then applied, removing variables with the largest non-significant p values, systematically until only significant variables and those that improved the fit of the model were retained. The prevalence ratios (PR) and 95% confidence intervals are presented. A p-value of less than 0.05 was considered statistically significant.