Study design and study setting
This was a cross sectional study conducted within the catchment communities of two major private not for profit (PNFP) facilities; Kisiizi hospital in Rukungiri District, and Bwindi Community Hospital in Kanungu District in Western Uganda. In Uganda, the health care system is organised into a four-tier system (i.e., hospitals, health centres of levels IV, III and II). Hospitals are considered higher level facilities having consultant physicians and offering more specialised services, while health centres II, III and IV, found at parish, sub-county, and county levels, respectfully, are lower level (also primary health care) facilities. These primary health care facilities offer basic services. For example, health centre IIs offer out-patients consultations, and health centre IIIs offer some inpatient care and some laboratory services. In addition to all the services provided at health centres II and III, health centres at level IV also offer caesarean deliveries and blood transfusion services.
In this study, the hospital catchments included parishes that were located within 30 kilometres of the hospital. Kanungu District has an area of 1,291.1 Km2 and an estimated population of 252,144 people (28) and a population growth rate of 2.1% per annum (29). Kanungu has two hospitals: one public hospital (Kambuga Hospital) in Kambuga Town council and one private hospital (Bwindi Community Hospital) on the outskirts of Bwindi Impenetrable Forest. Additionally, there are two health centre (HC) IVs, 10 HC IIIs, and 17 HC IIs. Rukungiri District covers an area of 1566.8 Km2 and has an estimated population of 314,694 people in 2016 (28) with an annual growth rate of 3.0% (29). Some of the major hospitals in Rukungiri District include Nyakibale Hospital (public) and Kisiizi Hospital (private). Rukungiri has three HC IVs, nine HC IIIs, and 43 HC IIs.
Sample size determination and sampling procedures
Sample size estimation
This was a cross sectional cluster survey where a sample was selected from Kisiizi and Bwindi health facility catchment areas. For this study, a cluster was defined as a village, which is the local administrative unit in Uganda.
A total of 60 clusters, each consisting of 15 households, were selected proportionately from the two catchment areas leading to a minimum target sample size of 900 households that was estimated taking into consideration a design effect of 2 and a non-response rate of 10%. The expected proportion of women who selected places of delivery based on WASH in the previous 12 months was conservatively assumed to be 50% due to lack of literature from Uganda.
Selection of clusters and households
All villages per catchment area were listed before assigning random numbers. The random numbers were sorted, and the first villages based on sample size of clusters were then selected.
From each village, with the help of the village health teams (VHTs), the survey team listed all households with a child 0-12 months old during the survey period. If the number of eligible households in a village was more than 15, random sampling was applied to recruit the target number of households. If the number of eligible households was less than 15, a nearby cluster was then connected and blocking applied. If a household had more than one eligible woman, the one with the youngest child was recruited and interviewed in order to reduce recall bias.
Data collection and management
A semi-structured questionnaire was developed in English and translated to Lunyakitara, the local language of the study area. Trained research assistants interviewed mothers using the local language paper questionnaire but recorded the responses on the English questionnaires. The questionnaire captured de-identifiable data on the mothers’ socio-demographics, awareness of the services at the health facility, antenatal visits, distance to the nearest health facility, place of delivery of the most recent child, reasons for their choice of delivery place, as well as challenges experienced while at the health facility. The questionnaire was translated into Lunyakitara and back translated to English, and both English copies were compared for consistency. The English questionnaire, which is provided as supplementary file 1, was developed after reviewing relevant literature (11, 16, 17, 30). For quality control, all questionable data were reviewed and resolved by a member of the supervisory team every day.
Data entry and analysis
Descriptive statistics, such as frequencies and proportions, were used to summarise categorical data, while continuous data was expressed as means and standard deviations. To assess the determinants of delivery place, the outcome (delivery place) was dichotomized into health facility delivery coded as 1 or non-facility delivery coded as 0. Any delivery that occurred outside the health facility including at home was considered as a non-facility delivery. We ran a generalized linear model of the Poisson family with logarithm as the link function with robust standard error variances to obtain the prevalence ratios and their 95% confidence intervals for the determinants of mothers’ choice of place of delivery. Prevalence ratios were used instead of odds ratios since odds ratios tend to overestimate effect of predictor variables when probability of obtaining the outcome of interest is >10% (31). To assess the factors associated with choice of public versus private facility for child delivery, a similar regression technique was used. In both cases, covariates that had a p value less than 0.1 in bivariate analysis were considered in the final model building. All variables were added in the model and removed stepwise from the model starting with those with highest p values until only those variables with p values less than 0.05 or that significantly improved model fit were retained in the model. The adjusted prevalence ratios (PRs), their 95% confidence intervals and p values are been presented. All analyses were performed using STATA 14.0 (StataCorp, Texas,USA).