Study design and setting
This prospective cohort study was conducted between August 2018 and January 2020 at the Prevention of mother-to-child transmission of HIV (PMTCT) clinic in the Lira Regional Referral Hospital (LRRH). LRRH has an annual antenatal care attendance of about 5,000 patients and conducts approximately 6 – 7,000 deliveries annually. Maternity services are offered freely at LRRH. The PMTCT clinic is an initiative of the Ugandan Ministry of Health where free HIV care and treatment is offered to HIV-infected pregnant women. At the PMTCT clinic, the women receive their antenatal and routine HIV care till they deliver their baby. The mother has a choice of delivering her baby at any health facility or clinic of her choice. The Ugandan healthcare system is subdivided into the national and district levels. These health facilities by nature are either public health facilities, private-for-profit (PFPs) or private-not-for-profit (PNFPs). PFPs are mainly clinics while PNFPs are religious based. Public health facilities (PHF) offer free services and are ranked as national, district and community based levels. The lowest level for the PHFs is the Village Health Teams (VHTs). VHTs is considered level one and is comprised of community health workers who deliver preventive and curative services as well as health education in the communities. Health centre two (HCII) offers outpatient services and is run by a nurse. Health centre three (HCIII), that is run by a clinical officer offers outpatient, inpatient, simple diagnostic and maternal health services. The Health centre four (HCIV) offers all services offered by HCIII as well as surgical, blood transfusion and emergence obstetric care services. HCIVs are run by medical doctors. National and regional referral hospitals are ranked at national level and offer all services of HCIVs as well as more specialised services (19).
Participants and procedures
HIV infected women with a gestational age of 20 weeks or more and receiving antenatal care at LRRH were consented, consecutively enrolled and interviewed on socio demographic characteristics as well as HIV-related information like antiretroviral regimen, duration and a viral load test done during pregnancy. They were then followed up with a telephone interview around the time of delivery. At this point, women were interviewed on circumstances surrounding labour and delivery like time of onset of labour, type of delivery, place of delivery, person who supervised the delivery, maternal ART adherence. Five hundred and five (505) HIV infected pregnant women were included in the final analysis because they had the completed data required (Figure 1).
Data was collected by trained research assistants that were fluent in Lango and English. Participants were requested to avail their telephone contacts or that of a trusted person to minimize loss to follow-up. The research team also documented detailed mapping for each participant’s physical address. In case all the participant’s telephone contacts were unavailable, a home visit would be done only if the participant had consented to it at enrolment.
Sample size estimation
A total of 505 HIV infected pregnant women were enrolled in the study. This sample size for detecting a difference between two independent proportions was calculated using STATA version 14.0 (StataCorp; College Station, TX, USA) assuming 80% power, 95% confidence interval (CI) and 5% precision. We also assumed that 51% of HIV infected women delivered in a health facility (14) and that 24.6% of HIV infected women delivered at home (15). The total sample size was then 455 women. After accounting for a non-response of 10% our final sample size was 505 HIV infected women.
Measurement of variables
The interviews were conducted in Lango (the language predominantly spoken in the study setting) and English by trained study staff using a structured questionnaire (this has been provided as a supplementary file). The questionnaires were translated into Lango and back translated into English to ensure consistency in interpretation of information. Marital status was categorised into married and single. Those who were married or cohabiting were combined into one group and labelled “married”. Those who were separated, divorced, widowed or not married were combined into one group and labelled “single”. We created a composite index of wealth (socio-economic status) using principle component analysis (PCA) (20). We used PCA on house ownership, availability of electricity in the house, source of drinking water and fuel used for cooking. Scores were obtained and categorized into five groups (quintiles) ranging from the poorest to the least poor.
Women whose labour started between 0600 hours to 1859 hours (Ugandan time) were all categorised and labelled as “day-time onset of labour” and for those whose labour started from 1900 hours to 0559 hours were categorised and labelled “night-time onset of labour”. During the follow-up at the time of delivery, for the measurement of maternal ART adherence, we asked the mother, “In the past week, did you miss taking any dose of your medication?” This was a “yes” or “no” response. If the mother answered “yes” she was considered “non-adherent”.
The outcome of this study was “home delivery”. Women who delivered in any type of health care setting like national referral hospital, regional referral hospital, public health centre or private clinic were all categorised and considered to have delivered in a “hospital setting”. Those that had delivered at the traditional birth attendant, home or on the road side were all categorised and considered to have delivered in a “non-hospital setting” which we refer to as “home delivery” in the rest of the text for comparability purposes.
Data analysis and management
We collected data using pretested, structured questionnaires. Two independent people entered the data into Epi data (www.epidata.dk, version 4.4.3.1). Data was then exported to Stata version 14.0 (StataCorp, College Station, Texas, U.S.A.) for analysis. Continuous data, if normally distributed, was summarised into means and standard deviations and if skewed, was summarised into medians with their corresponding interquartile ranges. For categorical variables, frequencies and proportions we calculated. The incidence of home delivery was estimated and its confidence limits calculated using the exact method. Poisson regression models / analysis were used for bivariate and multivariate analyses (21). All variables that had a p value < 0.25 at bivariate level and those of biological plausibility were collectively put into a multivariable model while controlling for confounding. We estimated unadjusted and adjusted risk ratios with their corresponding 95% confidence intervals.