Study sample
The analysis is based on a nationally-representative survey of ever-married women aged 15–49 from the 2008 Egypt DHS. To examine health-seeking behaviours related to maternal care, we assessed behaviours surrounding the most recent birth among women who reported having given birth in the five years preceding the survey. We analysed costs of care among the subsample of women whose most recent birth occurred in the twelve months prior to survey to limit the need for women to recall costs over longer periods of time. The average annual inflation rate in consumer prices in the period between 2003 and 2007 was 7.5% [19].
Ethics
The collection of the DHS data was approved by local authorities in Egypt; respondents’ informed consent was sought. This secondary analysis of anonymised data was approved by the Research Ethics Committee of the London School of Hygiene and Tropical Medicine, UK.
Measures of SEP
Socio-cultural capital
Education and literacy capture knowledge, ability to access new information, cognitive skills, previous exposure to authority, ability to interact with modern institutions such as healthcare providers, and have been linked to effective negotiation within familial power structures [20-23]. The education level of other decision-making members of the household influence health-seeking decisions through awareness of the benefits of medical assistance during pregnancy and support in seeking care [24]. Employment status captures the utilisation of attained education and exposure to wider social networks through workplace interactions. The latent measure of socio-cultural capital was based on woman’s and her husband’s education (continuous variable reflecting number of years of education) and woman’s literacy (illiterate, reads/writes with difficulty or reads/writes easily). Husband’s occupational category (not employed, unskilled manual, skilled manual, services, agriculturally employed, agriculturally self-employed, sales, clerical and professional) was used. A binary variable captured the working status of the female respondents, as the large majority (87.1%) reported not to be working. High scores on the latent variable represented higher socio-cultural capital.
Economic capital
Household-level material resources available to meet the direct and indirect costs of care were captured by the economic capital latent variable [25]. This construct would ideally be captured by measures such as income, consumption or expenditure. However, the collection and post-processing of such measures is resource-intensive and requires sophisticated econometric techniques. The DHS wealth index provides a more stable measurement of household-level resources than consumption expenditure [26], although the underlying constructs may not coincide [27]. A household wealth index score based on principal component analysis of 79 separate household-level variables was constructed in the DHS. In order to be able to replicate the current analysis on other datasets collected in Egypt with fewer available variables, we constructed a simpler variable to reflect the relative distribution of accumulated resources among households in which women who have had a birth in the five-year recall period resided. Its ten variables consisted of binary descriptive characteristics of the current living residence: utilities (water piped into dwelling, flush toilet), household ownership of assets (fridge, car, mobile, colour TV, water heater, automatic washing machine), ownership of a bank account, and level of crowding. Crowding was calculated as the number of household members per bedroom, and dichotomized as being above or below the median level (1.5 members per bedroom) within the sample of women. High scores on the latent variable index of economic capital represented wealthier households.
Health-seeking behaviour outcomes
Antenatal care
Seven dimensions of ANC utilisation for the most recent pregnancy were assessed (Figure 1). A binary variable indicated whether the woman received any facility-based ANC during the pregnancy. If ANC was utilised, binary variables described its timeliness (whether first ANC visit occurred in first trimester of pregnancy), intensity (four or more ANC visits were received during pregnancy), and the type of provider used (public or private). The definition of private provider included any facility-based non-public providers, such as private hospitals, clinics, doctors, the Egyptian Family Planning Association, the Clinical Services Improvement project, and other non-governmental organisation/private providers. Only 2.1% of women who used ANC reported receiving care from a combination of public and private providers; we grouped women who used both public and private providers with those who solely used private providers.
Delivery care
We used five health-seeking behaviours to describe women’s utilisation of delivery care (Figure 1). Firstly, a binary variable captured whether the most recent delivery in the five-year recall period occurred in a health facility. Among the subset of women with facility deliveries, we examined the use of private providers. A binary categorisation of private providers was constructed, combining all non-public sector providers (private hospitals/clinics, private doctor’s offices and other private medical facilities, including non-governmental organisations).
Cost of care
The analysis of price of antenatal and delivery care was limited to births occurring in the 12 month period before survey. Among women who used public providers, we analysed the binary outcome capturing whether this care was obtained free of charge. Among paying users of public services and women who used private providers, we analysed the amount paid for care (Figure 1). Specifically, women were asked whether they paid for ANC services (excluding laboratory or medication costs) separately during each visit, on a one-time basis, or received ANC for free. Among paying ANC users, we created a variable capturing the per-visit cost of ANC. In order to arrive at the per-visit cost among women who incurred one-time payments, the total ANC expenditure was divided by the number of ANC visits during pregnancy. For women who reported paying for each ANC visit separately, the amount reported paid for the last ANC visit during pregnancy was used. The cost of delivery service (excluding laboratory and medication expenses), reported by women with a facility-based birth was analysed. We constructed a binary variable capturing whether delivery care was received for free or not. Among women who reported paying for delivery care, a continuous variable captured the amount paid. The resulting continuous variables reflecting price of ANC and delivery care in Egyptian pounds (EGP), 1USD = 5.5 EGP in 2008), which were estimated separately by provider type.
Complete maternal care package
For the purposes of analysing the receipt of the basic elements of maternal care, we defined a complete maternal care package as the receipt of timely (first visit in the first trimester of pregnancy) and regular (four or more ANC visits during pregnancy) facility-based ANC and facility delivery. Women who did not receive any or all of these three care elements were considered not to have received the complete package. This binary classification was made regardless of whether such care was obtained from public or private providers and irrespective of the cost incurred for this care.
Confounders
We identified a priori confounders of the association between socio-cultural capital, economic capital and maternal health-seeking behaviours [14], including woman’s age group at the time of the most recent birth, parity group and whether pregnancy under analysis was intended or not [28]. Elements of availability of health services were captured in the residence variable (urban or rural) and whether respondent had unmet need for contraception at the time of the survey [29]. We created a binary variable for female head of household to capture the extent of the respondent’s autonomous decision-making. Additional variables related to maternal care were also used in the analysis of subsequent health-seeking outcomes, including the use of any ANC, use of regular ANC, use of private ANC, receipt of information about delivery complications during pregnancy, and delivery by caesarean section.
Statistical analysis
Latent variable modelling is an approach to quantify unobservable constructs by utilising common variance among observed indicators. Variance that is not common, including random error, is disregarded from the latent summary. The aim is to reduce the dimensionality of the observed data, but to retain a good representation within the latent variable identified [30,31]. Latent variables capturing socio-cultural and economic capital were constructed in Mplus/v.7.11 using the Weighted Least Squares, Mean and Variance adjusted (WLSMV) estimator. Factor loadings of each observed variable represent the association between this indicator and the underlying construct. Proportion of missing data in the observed variables in both latent constructs was minimal, and all observations were included. Model fit was assessed with the Comparative Fit Index (CFI), the Tucker Lewis Index (TLI) and the Root Mean Square Error of Approximation (RMSEA). The latent scores were standardised to a mean of zero and standard deviation of one.
Figure 2 shows the conceptual framework of the analysis in which socio-cultural capital can be directly or indirectly (through economic capital) associated with the outcomes. Continuous latent scores for both variables were entered in the mediation model, in order to jointly estimate their associations. The direct effects of both measures of SEP on binary outcomes was modelled in logistic regression and odds ratio was the main effect estimate. The total effect of socio-cultural capital (sum of its direct and indirect effects) on binary outcomes is expressed as the sum of changes in the probability of outcome (ΣΔp).
We estimated the mean price of ANC and delivery care by provider type and assessed the effectiveness of free public care targeting by comparing the mean socio-cultural and economic capital scores between women who received ANC or delivery care free of charge at public facilities with those who used these public services but paid for care. The extent of SEP inequalities accumulated throughout the health-seeking process for the most recent birth was estimated by comparing the mean SEP scores of women who received the complete maternal care package with those who received no maternal care. Inequalities in this multidimensional outcome and in samples used for assessment of targeting were examined in the subsample of women who delivered in the twelve month period preceding the survey, using the t-test.
We accounted for the complex survey sampling (clustering, stratification and weights) by using the svyset command in Stata in the descriptive overview of the sample and in analysis of targeting and multidimensional outcomes. The Stata medeff command was used for mediation analysis, incorporating robust standard errors adjusting for clustering and sampling weights [32]. The proportion of missing data in the majority of the outcome variables was minimal and we utilised complete case analysis in the mediation analysis.