We conducted an analysis of inequality in utilization of hospital-based maternal delivery services in the Bushbuckridge and Hlabisa health sub-districts of Mpumalanga and Kwa-Zulu Natal Provinces, respectively. These two sub-districts were chosen because they both have HDSS providing population-level data on SES, births, and location of delivery. In the analysis, the SES of households with a birth in a woman 18 years or older in the previous year, obtained from HDSS data, was compared with the household SES, obtained from a representative sample of women, 18 years or older, who had delivered in hospitals in the two sub-districts. In 2009, 90.4% and 79.4% of maternal deliveries in Bushbuckridge and Hlabisa, respectively, took place in the formal health system (i.e. with skilled attendance). Of these, the vast majority (95% in Bushbuckridge and 92% in Hlabisa) occurred in hospital facilities []; hence the decision to conduct interviews at hospitals. Pregnancies that terminated in abortions or where the outcome was unknown were not included in the analysis.
Population level data
The Agincourt HDSS (AHDSS) consists of an annual census of approximately 107500 people (as of May 2013) in an area of Bushbuckridge []. The following data were extracted from the AHDSS for the year 2007 from the 10,511 households with complete socio-economic data: number of pregnancies and their outcomes, maternal age and education, household characteristics, namely type of material used to construct the house walls and roof, access to water, toilet type, fuel used to cook, and ownership of assets such as a TV, fridge, stove, radio, landline telephone, vehicle, bicycle, and livestock. The 1,527 households with a woman over 18 years of age who had delivered in the preceding year were defined as the households needing maternal health services. The Africa Centre Demographic Information System (ACDIS) collects similar data on approximately 85,000 people in an area of Hlabisa []. Data from this database were extracted for 2009 on 8,448 households with complete socio-economic data and the subset of 1,491 households with a woman over 18 years of age who had delivered in the preceding year. Additional data on ownership of the following assets were available from this census: bed-nets, bed, block-maker, car battery, hot plate, kettle, gas cooker, kombi (vehicle), sink, motorcycle, primus stove, sofa, sewing machine, table and chairs, DVD player and wheelbarrow. Household characteristics and assets from both datasets were used to estimate an SES measure for each household in the two populations.
We use Multiple Correspondence Analysis (MCA) to create an SES index. MCA, an extension of Correspondence Analysis, is used to measure the relationships between several categorical variables. MCA aims to decrease high dimensional data space through finding dimensions that capture the largest amount of information common to all the variables []. The SES index was computed separately for each sub-district using the HDSS population data on access to basic services (water, electricity, sanitation), type of house, and the household assets listed above. We only use the index formed by the first dimension identified in MCA, as this index already captures a very large proportion of the common information between the socioeconomic variables (79% and 74% in Bushbuckridge and Hlabisa, respectively). Once the continuous SES index was constructed for each sub-district, households were ranked by SES and grouped into quintiles ranging from lowest to highest SES. We use this relative measure of socioeconomic status, SES quintiles, to allow comparison of SES gradients across the two sub-districts included in our analyses. The absolute values of the continuous indices cannot be directly compared because their meanings differ in the communities.
Sub-district level user data
We conducted patient exit interview surveys in women over 18 years of age delivering in one of the three hospitals in the two sub-districts (two in Bushbuckridge and one in Hlabisa) during the study period. Based on a Chi-squared Goodness-of-Fit test, we estimated that a sample of 300 women per sub-district would be required to detect SES differences with 80% power. In Bushbuckridge the sample was distributed proportional to the number of deliveries in each of the two hospitals. Respondents were recruited systematically at the time of discharge from the post-natal ward until the required sample size was achieved in each facility. Trained interviewers carried out the exit interviews in the local language of the respondent, collecting socio-economic data, as well as additional access variables related to the geographic accessibility, financial affordability, and cultural acceptability of hospital delivery services. During the course of the survey, a structured quality inventory on health systems inputs, processes and outputs was also completed in each hospital to measure the hospital capacity for comprehensive emergency obstetric care.
Data were collected over a period of 15 months, from June 2008 through September 2009. Ethical clearance for this study was obtained from the Universities of Cape Town, Witwatersrand and KwaZulu Natal, and provincial and local Departments of Health authorized the study. Written, informed consent was obtained from each participant in the exit interviews.
Comparing the SES distributions of people needing and people using maternal delivery services
The SES distribution of the population in each sub-district, categorised into quintiles, was used to compare the SES distributions of the women needing and using maternal delivery services. We determined first the proportion of women in each quintile in need of maternal delivery services and then the proportion of women in each quintile who actually used these services. In this way these SES distributions are directly comparable. In order to test for trends and associations between those needing maternal delivery services and those who used these services, the Partitions of Pearson’s Chi-squared test for ordered columns, a contingency table analysis of ordered categorical variables (such as quintiles), was conducted [].