Existing frameworks to understand use of skilled attendants

The DiSBA framework draws on prior research on maternal mortality and the determinants of use of maternal health services. In particular we draw on three prior models [21,15,5]: McCarthy and Maine’s (1992) framework for analyzing the determinants of maternal mortality; the three delays model by Thaddeus and Maine (1994), which posits three delays—the delays to seek, reach, and receive care—that lead to maternal mortality from the onset of an obstetric complication; and the recent expansion of the three delays model by Gabrysch and Campbell (2009) to include preventive obstetric care.

McCarthy and Maine’s comprehensive framework highlights the influence of contextual factors yet it does not illustrate the complex interactions amongst the intermediate determinants. Thaddeus and Maine’s three delays model focuses on pregnant women and their care-seeking behavior when they experience a complication. This model has been widely used, yet it implies a single pathway for the effect of SES—a single arrow from socioeconomic/cultural factors to the delay to seek care. But socioeconomic/cultural factors can also affect the delay to reach and receive care through a number of pathways. Gabrysch and Campbell’s model includes care seeking for uncomplicated pregnancies and partially addresses the multiple pathways for socioeconomic/cultural factors by separating out economic factors from sociocultural factors, with sociocultural factors affecting to the delay to seek care, and economic and physical accessibility affecting the delays to seek, identify, and reach care. However, certain sociocultural factors may still lead to other delays besides delays in deciding to seek care. For example, sociocultural factors related women’s autonomy can potentially affect the time to reach care, given situations where women cannot travel freely or are not allowed to use certain types of transportation, which may be the most readily available form of transportation [15,22].

None of these models address the effect of socioeconomic factors on the delay in receiving adequate and appropriate treatment. SES can impact the timeliness and quality of care received in several ways. First, in settings where patients are expected to buy supplies and medication upon reaching the health facility, higher SES women who can afford such supplies are much more likely to receive prompt attention. It has also been demonstrated that higher SES women are treated differently than lower SES women, suggesting that quality of care may be better for women with means [23–25]. In addition, higher SES women have the means to choose facilities—such as private, non-government facilities—that may offer better quality of care [26–29]. These factors will in turn affect their perceptions of the quality of care, which will affect future decisions of whether or not to return for care [5,24,30].

These examples suggest a complex interaction between socioeconomic factors and the phases of delays that are not explicitly illustrated in the three delays models. Our framework builds on the strengths of these earlier models and attempts to address their limitations to provide an integrated, yet simple, conceptual framework for understanding the sources of disparities in the use of SBAs. In addition, we draw on the health belief model, which posits that people are more likely to act if they perceive a need (based on their perceived susceptibility, perceived etiology, and severity of the condition), perceive that benefits of the action outweigh the barriers; and have cues to action and reinforcements [31].

The Disparities in Skilled Birth Attendance (DiSBA) framework

The main premise of the DiSBA framework is that the decision to use maternal health (MH) services, including use of SBAs, is based on three factors: perceived need for care, perceived accessibility (physical and financial) of the service; and perceived quality of the care. We refer to these three factors as the proximal determinants of use of maternal health services. SES and place of residence, the most common determinants of service use, are distal factors that affect use through these three proximal factors.

Perceived need is influenced by women’s current health status (e.g., having a pregnancy complication and the type and severity of the complication), reproductive factors (including age and parity), prior health status or pregnancy complications, health knowledge (general and specific to pregnancy), as well as by unknown factors that influence the development of pregnancy complications [21]. In addition, perceived need is shaped by socioeconomic and sociocultural factors.

Perceived accessibility is influenced by actual accessibility (physical access and cost of services) [15], as well as by socioeconomic factors, and may be modified by illness factors. As noted in the three delays model, accessibility has a dual role: it indirectly influences the decision to use the service through perceived accessibility and directly influence reaching the service once the decision to use the service has been made [15]. Perceived quality is based on an assessment of quality of care from a personal previous experience, or the experience of others [15,32]. The assessment of quality can be based on the structure, process (technical or interpersonal) or outcome of care as described by Donabedian (1988), though it may be dominated by the interpersonal aspects of care; and can be based on any type of prior encounter with the health system [15,33,34]. In addition, perceived quality of care is influenced by socioeconomic and sociocultural factors [30]. All the factors are influenced by larger contextual factors in which the health system and sociocultural factors are embedded.

This paper provides a framework to guide questionnaire development, data collection, and quantitative analysis of factors that mediate SES disparities in use of SBAs. The simplicity of the model is not to undermine the numerous factors associated with use of SBAs in the literature, but to provide a way of thinking about the key predictors. It will help maternal health researchers to move beyond simply including covariates in regression models to thinking about what role the covariates are playing.

The DiSBA framework is illustrated in Fig 1. The key assumption of the DiSBA framework is that SES differences in perception of need, access, and quality of care account for the SES disparities in use of SBAs.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. The Disparities in Skilled Birth Attendance (DiSBA) Framework, showing mediated pathways for socioeconomic status.

https://doi.org/10.1371/journal.pone.0154110.g001

Depending on the purpose of the research, the context, and type of data available, we may pay greater attention to one or more proximal factors. It is our hope, however, that the framework will guide work from the conceptualization phase of projects that involve primary data collection and ensure that we collect data on all the proximal factors, in addition to data on the factors that influence them, to ensure more complete analysis. For the case of Ghana, where ANC attendance is much higher than use of SBAs, with large SES disparities in use of SBAs, but not in ANC attendance; and where quality of care has been mentioned as a reason for non-use of SBAs [35–37] we ask three questions:

1. Does quality of ANC predict use of SBAs among women who attend ANC at least once during pregnancy?
2. Does quality of ANC explain some of the SES disparities in the use of SBAs?
3. Compared to the other proximal determinants, what is the relative contribution of ANC quality to the SES disparities in the use of SBAs?

We focus on ANC quality because the only measures of quality in our data are from ANC. Also, since the effect of quality of care on the decision to use a SBA must be from a prior encounter (personal or vicarious) with the health system, looking at the effect of ANC quality during pregnancy on the delivery provider for that pregnancy captures this temporal ordering.

Data

The data for this analysis are from the 2007 Ghana Maternal Health Survey (GMHS). The survey was conducted by the Ghana Statistical Service and the Ghana Health Service with technical assistance from Macro International, and has been described in detail elsewhere [38]. Based on a multistage cluster design, households were randomly selected from all regions of Ghana and household and women’s questionnaires were administered face-to-face by trained interviewers. Verbal consent was obtained from respondents. The response rate was 99% at the household level and 98% for the individual women, with 10,858 completed household interviews and 10,370 individual interviews with women aged 15–49 years [38].

Unlike the Ghana Demographic and Health Survey (GDHS) [8], the GMHS collected health service utilization data for all women who had a birth (live or still birth, not only live births) in the five years preceding the survey (N = 5,088, or 49% of all women interviewed); this is the base sample for the analysis. The analytic sample consists of 5,042 women (99.1% of the base sample) because 46 observations are missing on key study variables. The main analysis is further restricted to women who had at least one ANC visit during their last pregnancy, since quality of ANC cannot be measured for women who did not have any ANC. Ninety seven percent (N = 4,868) of women in analytic sample had at least one antenatal visit. The full analytic sample is examined in sensitivity analysis.

This study was granted an exemption under the University of California, Los Angeles Institutional Review Board exemption category 4 for research involving the study of existing data. Ethical approval for the GMHS was however obtained from the Ghana Health Service Ethical Review Committee.

Variables

Statistical methods

Initial analyses involved descriptive statistics and examining the bivariate associations between the independent variables and the dependent variable using chi-squared tests. We then examined associations using multivariate binary logistic regressions to account for the effects of other factors. The models are built starting with the null model, then adding the key independent variables, the control variables, and finally the mediating variables. Mediating variables are those variables that explain the relationship between the key independent variables and the dependent variable. In this analysis the mediators are the proxies for the proximal determinants.

When the dependent variable is continuous, the mediated effect is the change in the coefficients of the key independent variables when the mediators are added to a linear regression model. However, because the addition of variables to a logistic model changes its scale, it is not accurate to consider the difference in the coefficients in nested logistic models as the magnitude of the mediated effect [43,44]. We therefore used the ‘khb’ rescaling method to calculate the magnitude of the mediated effects. In this method, the residual of the mediators are applied to the reduced model to fix the scale of the reduced model to that of the full model, so that the coefficients for the key independent variables can be compared across the nested models. The coefficients of the key independent variables in the models with the residuals of the mediators (the rescaled reduced model) are their total effects (c); and the coefficients of the key independent variables in the models with the actual mediators (the full model) are their direct effects (c’). The difference in the coefficients (c-c’) in the two models are the mediated effect, and the proportion of the total effect mediated is (c-c’)/c [43,45]. All the analyses are weighted using the sample weights provided with the data to account for the complex sampling design.

Sample distribution

Table 1 shows the distributions of key study variables for the 97% of women who received ANC at least once during their last pregnancy (N = 4,868). These are similar to the distributions for the full sample, which have been described elsewhere [26,48]. The average woman in the sample is about thirty years old, has had about four pregnancies, and is married (72%). About a third of the women have no formal education and only about 8% have attended senior secondary school or higher. As expected, the analytic sample is almost evenly distributed between the five wealth quintiles. About two-thirds of the women live in rural areas. Of the women who attended ANC at least once, about 80% had four or more visits as recommended by WHO. About 61% received higher quality ANC (8–9 of the 9 services), and 58% have high need (2–6 of the need variables). A SBA assisted a little over half (57%) of the women at delivery—10% by doctors and 47% by nurses or midwives.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Weighted distributions of key study variables for women who attended ANC at least once, GMHS, N = 4,868.

Notes: ANC = Antenatal care; GMHS = Ghana Maternal Health Survey; JSS = Junior Secondary School; SSS = Senior Secondary School; SBA = Skilled Birth Attendant.

https://doi.org/10.1371/journal.pone.0154110.t001

Bivariate results

Table 2 shows the bivariate distributions for the key study variables by the delivery assistant. Use of SBAs increases with both education and wealth: 35% of those with no education used SBAs, compared to about 90% of those with secondary education or more; and 30% of the poorest women used SBAs, compared to 92% of the richest. Forty-one percent of women in rural areas were assisted by a SBA, compared to 87% of women in urban areas. About 64% and 65% of women with high need and those who received higher quality ANC respectively were assisted by a SBA, compared to 48% of those with low need and 45% of those who received lower quality ANC.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Crosstabs of key variables by delivery assistant, GMHS, N = 4,868.

Notes: See Table 1 notes for abbreviations.

https://doi.org/10.1371/journal.pone.0154110.t002

Multivariate logistic regression results

The results from the weighted multivariate logistic regression for use of SBAs are shown in Table 3. We present three sets of models, model 1 contains the key independent variables controlling for age, parity, marital status, age at first union, religion, and ethnicity; model 2 adds on the mediators; and model 3 adds on ANC frequency, timing, facility, and provider.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Weighted binary logistic regression of use of SBAs on relevant predictors, GMHS, N = 4,868.

Notes: See Table 1 notes for abbreviations; *p<0.05, ** p<0.01, *** p<0.001.

https://doi.org/10.1371/journal.pone.0154110.t003

All three models show that use of SBAs still increases with education and wealth net of other factors. Women with a middle school education or more and those in the highest wealth quintiles have over two times higher odds of using a SBA than those with no education and those in the lowest wealth quintiles respectively. The effects of the mediators are also significant net of other factors. Women who live in urban areas have about four times higher odds of using a SBA than those living in rural areas. Women who received higher quality ANC have about 47% higher odds of using a SBA than those who received lower quality ANC; and those with higher perceived need have about 28% higher odds of using a SBA than those with lower perceived need. We also find that women ages 20–24 are less likely to use a SBA than older women, while primiparous women are more likely to use a SBA than women with higher order births. Women who are cohabiting, those who belong to the Traditional/other religion group, and those of Ewe, Mole-Dagbani or Hausa ethnicity are less likely to use SBAs than those who are married, Christian, and Akan respectively.

The other factors that are positively associated with using a SBA in model 3 are attending ANC four or more times and receiving ANC from a doctor (compared to a nurse). Also, women who received ANC in lower-level facilities like health centers and health posts are less likely to use a SBA at delivery than those who did so in higher-level government facilities like hospitals or polyclinics. The mediation analysis is based on models 1 and 2. We exclude the ANC frequency, timing, facility, and provider from the mediation models because these variables are all associated with SES, rural/urban residence, and quality of ANC, as in previous studies [26,49,50]. In addition, they are associated with our perceived need measure—women with high need are more likely to start ANC early and be seen by doctors in higher-level facilities. Because these variables lie in the mediation pathway from SES to use of SBAs and are theoretically antecedent to ANC quality and perceived need, including them significantly reduces the magnitude the effects mediated by ANC quality and perceived need (although they are still significant).

Mediation results

The results for the mediation analysis are shown in Table 4. Our measures of perceived access, perceived need, and perceived quality of care account for about 23% of the difference between women with no education and those with only primary school education and about 55% of the difference between women in the lowest wealth quintile and those in the middle wealth quintiles. In general, urban residence has the largest contribution to the mediated effect, which is not surprising because of its contextual nature, followed by quality, and then need—except for the difference between primary and no education where need has a slightly bigger effect.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Mediation analysis using weighted binary logistic regression of use of SBAs and the ‘khb’ rescaling method.

https://doi.org/10.1371/journal.pone.0154110.t004

Sensitivity results

For the most part, the findings from the unweighted multilevel logistic models are essentially the same in direction and significance of associations, and comparable in magnitude of the associations to the results from the weighted single level models presented. One exception is that the difference between those with no education and those with a primary education is not significant when the mediators are added to the multilevel model, where as there is still a significant difference in the single level model—potentially because standard errors tend to be smaller in single level models when data are clustered. The models using the full sample (including the 3% who did not attend any ANC) are essentially the same as that for the sample restricted to only women who had at least one ANC visit. In this model, quality of care is scored zero for those who did not attend ANC and an indicator variable is included in the model for whether or not the person received any ANC.

Strengths and limitations

A major strength of this paper is that it uses a theory-based approach to data analysis. We propose a conceptual model to help understand the factors underlying SES disparities and illustrate how it can be used. The DiSBA framework can be adapted to examine the sources of other disparities including place of residence and age, as well as for other maternal health services. Such theory-based analysis will help address an important gap in the MCH literature, which is the dearth of quantitative studies exploring the mechanisms underlying various associations with use of maternal health services.

The main limitation is that we have had to use proxy measures for the proximal determinants, which may not adequately capture all dimensions. It would also have been preferable to have data on women’s existing comorbid medical conditions or her psychological state to determine if they had a role in her care seeking. However, such data were not available. GMHS data are cross-sectional, which limits causal inference; and recall and social desirability bias are potential problems, since the data are based on self-report. The age of the data is also a limitation. Nonetheless, this dataset has an advantage over the DHS of including a nationally representative sample of women who had a birth (live or stillbirth) in the five years preceding the survey, which reduces the chances of excluding women who received the worst care, as these women may be more likely to have a stillbirth.