Menu
Menu
Menu
Menu
Introduction In the era of Sustainable Development Goals, reducing maternal and neonatal mortality is a priority. With one of the highest maternal mortality ratios in the world, Malawi has a significant opportunity for improvement. One effort to improve maternal outcomes involves increasing access to high-quality health facilities for delivery. This study aimed to determine the role that quality plays in women’s choice of delivery facility. Methods A revealed-preference latent class analysis was performed with data from 6625 facility births among women in Malawi from 2013 to 2014. Responses were weighted for national representativeness, and model structure and class number were selected using the Bayesian information criterion. Results Two classes of preferences exist for pregnant women in Malawi. Most of the population 65.85% (95% CI 65.847% to 65.853%) prefer closer facilities that do not charge fees. The remaining third (34.15%, 95% CI 34.147% to 34.153%) prefers central hospitals, facilities with higher basic obstetric readiness scores and locations further from home. Women in this class are more likely to be older, literate, educated and wealthier than the majority of women. Conclusion For only one-third of pregnant Malawian women, structural quality of care, as measured by basic obstetric readiness score, factored into their choice of facility for delivery. Most women instead prioritise closer care and care without fees. Interventions designed to increase access to high-quality care in Malawi will need to take education, distance, fees and facility type into account, as structural quality alone is not predictive of facility type selection in this population.
Primary data about individual women and their deliveries were obtained from the 2013–2014 Millennium Development Goal Endline Survey (MES),18 19 a nationally-representative household survey that used a multistage stratified sampling strategy to include households within enumeration areas (EAs) identified by the 2008 census. Locations of EAs in the MES were obtained from the Malawi National Statistical Office, 2008 Malawi Population and Housing Census, 2013 update.18 Responses were weighted for national representativeness. More detail on the MES survey has been published elsewhere.8 A total of 7750 deliveries was captured by the MES; women were surveyed about their most recent pregnancy (if more than one) in the past 2 years (2013 and 2014). Exclusion criteria for this study included the first entry of any duplicated record (n=50), women with no documented delivery location (n=240) or a reported delivery location that could not be matched to the Malawi Service Provision Assessment (SPA)19 facility types (eg, ‘Other’; see next section on delivery facilities) (n=102), an EA with a location that could not be matched to the census (n=107) and delivery more than 100 km away (n=51) (figure 1). Because of the legal barriers to delivering with a TBA and because fewer than 10% of women in the survey reported delivering at home, home delivery was also excluded (n=575), leaving an analytic sample of 6625 women. Application of exclusion criteria to create analytic sample (n=6625). Health facility data, including geographic location, were obtained from the 2013 Malawi SPA,19 a census of the health system20 that includes a detailed audit of facility resources and clinical practices, including whether fees are charged for labour and delivery services. The MES asks women about the type of facility where they delivered (eg, government health centre, mission hospital and private maternity home), but not the name of the specific facility where they delivered. The SPA includes information on facility tier (eg, central hospital, district hospital and clinic) and management type (eg, government and Christian Health Association of Malawi). We aligned facility type responses between the MES and SPA surveys. We identified up to eight facilities of each facility type located near the woman’s EA centroid using Euclidean distance and calculated road distance to each of these in order to select the nearest facility by road. If road distance could not be calculated, we selected the closest facility of the appropriate type using Euclidean distance. Women were assigned to the closest facility matching the facility type they reported on the survey. A choice set of six facilities was created for each woman in order to analyse her facility preferences. This included the five nearest facilities to a woman’s EA in linear distance that were providing delivery services. The sixth choice included the facility the woman matched to if it was outside the five closest facilities in the choice set. A description of the choice set and associated characteristics can be found in table 1. Description of facilities in choice set The service readiness score for basic obstetric care was used as the marker for structural quality of delivery services in each health facility. The score is based on the recommended essential items needed to provide quality facility-based delivery services from the WHO Service Availability and Readiness Assessment Manual.21 The tracer items that compose the score include the availability of management guidelines, staff up-to-date with training and essential equipment, medicines and commodities for delivery care. The basic obstetric care service readiness score for each health facility was derived from the 2013–2014 SPA data.19 Data from the SPA and MES in Malawi were used to directly link characteristics of facilities to the delivery choices made by a nationally-representative sample of women who gave birth in 2013 or 2014. We hypothesised that facility characteristics could predict choice22 but that different preferences for these characteristics might exist across this cohort of women. To identify this unobserved, or latent, heterogeneity within this population, we chose to conduct a latent class analysis. Latent class analysis assumes a discrete number of segments (or ‘classes’) in the population, each with its own preference structures.23 24 In the context of this study, this analysis allows us to identify the different utilities for facility characteristics (as revealed in women’s facility type selection), determine the number of latent classes, calculate the probability of each woman belonging to each group or class and, finally, summarise the sociodemographic characteristics of the women likely to belong in each class. Following random utility theory, we assign the utility for woman i choosing alternative f to be24: where βi is the vector of preference coefficients for a woman for each facility-level characteristic, and x is the vector of facility-level characteristics (eg, obstetric readiness and fees). The error term, εif, is assumed to follow a Gumbel Type 1 distribution. With this assumption, the probability of a woman choosing an individual facility is: where F is the total number of facilities in a woman’s choice set. The value of each β in the βi vector is identical for each woman within a class but can take different values across classes. The probability of each woman belonging to a class is: where γ is the vector of logistic regression coefficients on sociodemographic variables, δ is the vector of examined sociodemographic variables and Q is the total number of classes in the latent class analysis.23 The variables included in the analysis are defined in the data dictionary online supplementary appendix table 1. Four facility-specific variables were selected a priori based on literature suggesting that accessibility, quality and out-of-pocket payment factors into facility selection, as cited above. Twenty-four individual-specific variables were chosen based on prior literature and author consensus and were tested stepwise. One hundred and ninety-four combinations of individual-specific variables were tested, with the Bayesian information criterion (BIC) informing the selection of the best-fitting formula. Sensitivity analyses were performed. After the model was selected, it was tested with 2–6 latent classes to determine the likely number of underlying preference structures; BIC informed. bmjgh-2018-000930supp001.pdf It should be noted that this is a revealed-preference latent class analysis: women reported characteristics of their deliveries retrospectively. Therefore, this is an analysis of the facility types that women chose, which reveal preferences, but this is not assumed to be the same as each woman’s stated-preference. Entropy, an indicator of quality of the model, was calculated to determine the separateness of the classes: where Q is the number of classes, N is the sample size, R is the latent class indicator variable for each woman i, Wi is the vector of latent class indicator variables for each woman i and the probability Extra close brace or missing open brace is generated from the final model.24 Lastly, to evaluate the potential bias introduced by random effects at the facility level, given that data are clustered by facility, a multilevel model was examined.25 The analytic dataset was created using Stata V.14.1. Geographic distances were calculated based on Google Maps using Python 3.6.1. Statistical analyses were performed using R V.3.4.0 (R Foundation for Statistical Computing, Vienna, Austria) gmnl package.15 26
N/A
