Decomposing Kenyan socio-economic inequalities in skilled birth attendance and measles immunization

Kenya, located in the eastern part of Africa, is divided in 8 provinces. There are various ethnic groups and two main religions: Christianity and Islam. The economy is predominantly agricultural with a strong industrial base [15]. In 2008, the Gross National Income per capita (PPP) was $1560 [16] and the Gini index was 42.5 [17]. Forty% of the labour force was unemployed [17]. In 2009, the population was estimated at 39 million [16, 18], 77% living in rural areas [16]. Adult literacy rate was 87% [16]. The number of women aged 15–49 years old was 9.6 million and the number of live births was 1.5 million. Fertility rate was estimated at 4.6 births per woman [19]. In 2008/09, 47% of women 20–24 gave birth before age 20 [15].

Data from the Demographic and Health Survey (DHS) conducted in Kenya in 2008/09 were used for the study. Data from preceding Kenyan DHS: 1993, 1998 and 2003 were used for a trends analysis. Data collection and processing are described in [15, 20‐22]. In brief, a two-stage sampling design stratified on region and place of residence (urban/rural) was used. Information on SBA (self-reported) was available for 4249, 2403, 5929 and 6059 children under 5 in 1993, 1998, 2003 and 2008/09 respectively. Information on measles immunization (recorded on the vaccination card or self-reported if not recorded) was available for 806, 780, 1096 and 1116 children aged 12–23 months in 1993, 1998, 2003 and 2008/09, respectively. The recall period was 5 years for both indicators but the measles analysis was restricted to children aged 12–23 months because this is the youngest cohort of children who have reached the age by which they should be fully vaccinated (<12 months) [15].

Data were transferred to RGui (R version 2.14.2., The R foundation for Statistical Computing) for analysis. Both health care indicators were binary variables. SBA takes a value of 1 if the delivery has been attended by skilled health personnel (doctor, nurse or midwife). Measles immunization takes a value of 1 if the child aged >12-23 months has been immunized against measles. The independent variables used in the analysis were: antenatal care attendance, sex of the child, child’s age, birth order, mother’s age at birth, type of residence (urban/rural), province, ethnic group, religion, marital status, parents’ level of education, parents’ occupation, insurance coverage and household wealth. The wealth index, computed by DHS, comprises household assets (type of flooring, water supply, sanitation facilities, electricity, persons per sleeping room, ownership of agricultural land, domestic servant, and other assets).

The concentration index (C) [23] was used as a measure of socioeconomic inequality. The method is described in details elsewhere [11‐13]. Briefly, a relative concentration curve plots on the x-axis the cumulative percentage of children ranked by household wealth and on the y-axis the cumulative percentage of the variable of interest (SBA, measles immunization or a determinant). The relative C is defined as twice the area between the concentration curve and the diagonal (line of equality). If the curve is below the diagonal, C is positive and the variable of interest is more prevalent among the wealthier households. Given that the bounds of the C of a binary health indicator depend on the mean of this indicator, a normalized C (C*) proposed by Erreygers was computed when comparing time periods or geographical areas [24].

In order to satisfy the linearity assumption of the decomposition method used [25], inequality in predicted SBA (ln odds SBA) and measles immunization was computed and decomposed into its contributing determinants [14]. For this, a multivariate logistic regression model was used to estimate the regression coefficients of each determinant on the two indicators, and compute the resulting predicted health care outcome. The C of the predicted health outcome can be expressed as:

the sum of contributions of the k determinants [13, 25]. The contribution of each determinant is a function of its regression coefficient β _k , its mean x ¯ k, the mean of the predicted health outcomeμ, and its concentration indexC _k .

All analyses were weighted (weights provided with the DHS data) and adjusted for the cluster randomized sampling frame (with cluster as the primary sampling unit and household as the secondary sampling unit). The criterion for statistical significance used was α = 0.01.

Health care indicators means and concentration indices were computed in 11 other African countries with DHS data dating from 2007 to 2012 (Figure 1). Kenya was among countries with the lowest SBA (<50%) and the highest inequality in SBA (C* > 0.4). However, with respect to measles immunization coverage, Kenya ranked second to Malawi with the highest coverage (85%) and had a median position in measles coverage inequality.

Nairobi and Central provinces had the highest percentages of SBA deliveries among women while North Eastern, Western and Rift Valley had the lowest (Table 1, Figure 2A). After controlling for the effect of the other variables, Rift Valley was still negatively associated with low uptake of SBA (Table 2). A relative normalized C was computed by province (Figures 2B). With the exception of the Central province and Nairobi, all provinces showed a low SBA coverage (<0.5) and a high inequality in SBA (C* > 0.3). Among this group of provinces, a positive association between coverage and inequality could be observed (R²: 0.79, p = 0.02). Coast and Eastern provinces had the highest inequality. Nairobi was the province with the highest coverage and the lowest inequality.

Rift Valley and Eastern provinces had the highest percentages of measles immunization coverage, and Western and Nyanza, the lowest (Table 1, Figure 3A). Nyanza, North Eastern and Rift Valley had the highest wealth-related inequality in measles immunization (C* > 0.2) (Figure 3B). Nairobi was the only province with a negative C, i.e. an inequality favouring the poor. Coast, Central and Eastern had a high coverage (>85%) and a low inequality (C* < 0.11).

Health care indicators were plotted as a function of wealth quintiles and a normalized relative C was computed, by survey year, in Figure 4. The wealth-related inequality in SBA was high in 1993 and persisted until 2008/09. SBA inequality showed a “mass deprivation pattern”, i.e., the majority of the population had equivalent but deficient access to SBA while small privileged groups (quintiles 4 and 5) had better access to this service. The pattern and the relative C varied little over the survey years.

The wealth-related inequality in measles immunization increased until 2003 and then dropped in 2008/09. The pattern took a “queuing” form, i.e., the general access to measles immunization was better than for SBA, but middle and richer wealth quintiles groups benefited most, while for the poorer groups a large proportion was still not vaccinated. In 2008/09, the queuing pattern seems to have diminished.

The distribution of SBA across population sub-groups is shown in Table 1, and regression coefficients from the multivariate analysis are shown in Table 2. In total, 44% of births were attended by a health professional.

SBA utilization increased with the household wealth quintile and with parent’s education. SBA coverage was four times higher in the richest quintile than in the poorest. Less than 20% of women with no education (13% of women) delivered with skilled assistance. This figure doubled when the mother had attained primary level education and tripled when the mother had attained secondary level education. SBA utilization was two times lower in rural households; however, the effect of rural residence faded when controlling for the other variables. Women from the Kikuyu ethnic group had the highest proportion of SBA uptake. In the multivariate analysis, women from the Kamba, Luhya, Luo and Mijikenda ethnic communities were less likely than Kikuyu to use SBA. Regardless of the socio-economic position of the household, SBA utilization decreased with birth order (p = 0.014), and increased with the number of antenatal care visits. Compared with women who had not attended antenatal care, it was three times greater when women had attended at least one visit, and about five times greater when women had attended the recommended four visits.

The distribution of measles immunization across population sub-groups is shown in Table 1, and regression coefficients from the multivariate analysis are shown in Table 2. Eighty-five percent of 12–23 months children were vaccinated against measles. Inequalities were less noticeable than for SBA, but nevertheless existed. Measles immunization increased with the household wealth quintile and with parent’s education. Children from the richest household wealth quintile and from parents with a secondary or higher education level were close to the 95% target. Coverage was below 80% in the following categories: poorest wealth quintile, parents with no formal education, Luo or Masai communities, father not working or working in the services or agricultural sector, Western province, no skilled antenatal care, birth order 4 and above. However, with the exception of birth order, the effect of socio-economic determinants on measles immunization was not statistically significant in the multivariate analysis.

The relative contributions to the overall C are shown in Table 2. A determinant can contribute to wealth-related inequality in health care both through its association with the health care indicator (regression coefficient) and through its unequal distribution across wealth groups (C). C indicates how unequally the determinant is distributed over wealth: if C > 0, the determinant is more prevalent among the wealthier, and if C < 0, the determinant is more prevalent among the poorer.

In 2008/09, the overall C for predicted SBA was 0.14. The wealth quintile accounted for about 40%. Twenty% of this inequality was explained by mother’s education. Other important contributors were antenatal care (9%), father’s education (8%), province (6%), birth order (6%), and ethnic group (5%).

For predicted measles immunization, the overall C was 0.08. Wealth itself accounted for 60% of this inequality. Other important contributors were birth order (33%), father’s education (22%), mother’s education (6%), and antenatal care (5%). In this decomposition analysis, some determinants had a negative contribution, meaning that they reduced the wealth-related inequality. Rural residence (−19%) was positively associated with measles immunization being more prevalent among the poor (negative C), whereas insurance coverage (−5%) was negatively associated with measles immunization but being more prevalent among the rich (positive C). Parent’s occupation was not a major source of inequality as a whole; only agricultural employee contributed notably to inequality in measles immunization.

DHS are internationally comparable surveys [40] but the comparison with other sub-Saharan countries included DHS from different years between 2007 and 2012, possibly skewing the observed differences. Results of the trends analysis should be interpreted carefully because the available information differed between 1993/1998 and 2003/2008. Since 1999, women were asked to provide information about pregnancies resulting in live births during the five years prior to the survey. Before, the reference period was three years; the number of children included was lower. The study is limited to the information collected through DHS; other factors could have played a role in the analyses. Assessing access (cost and distance) to health facilities might be especially interesting when analysing inequalities in SBA. About 20% of household heads were not the parents of the child; their occupation may have been useful as well.

This study brought to light several considerations when planning interventions aimed at improving health care coverage and equity.

Firstly, the decomposition analysis provided several areas to be targeted in order to reduce inequality in health care. These were: poverty reduction, educational attainment (preferably secondary level and above), antenatal care attendance and parity. Successful interventions targeting the consumer costs for transport and obstetric care barrier (e.g. community loan funds), mother’s education (e.g. community educators), or family planning (e.g. community-based delivery of services) were documented [30].

Secondly, the “mass deprivation” and “queuing” patterns of SBA and measles immunization presented in the trends analysis suggest that a broad strengthening of the whole system, possibly combined with targeting, is required [41]. A program targeting the rural poor was already implemented for measles immunization through the SIA, and was initiated in two districts of Nyanza in 2001 through the Skilled Care Initiative (SCI). SCI consists in the decentralization of routine and emergency obstetric care.

Thirdly, the analysis by province highlighted inequalities between provinces and wealth-related inequality within provinces. A more in-depth analysis determining the location of the most vulnerable sub-groups within provinces would help in better reaching the whole population during interventions.

Finally, each intervention should be consistent with the socioeconomic and political context which plays a proximate role in the process to equity illustrated in the conceptual framework proposed by the Commission on Social Determinants of Health (CSDH) [42]. The Kenyan context seems prone to changes for more equity in health. In the last decade, many initiatives were launched by the Kenyan government to improve social conditions and health, and some had an explicit equity goal [15]. Observations resulting from the study at hand are especially in line with the National Population Policy for Sustainable Development goals: improvement of the standard of living; health through education on how to prevent illness and premature death among mothers and children; sustenance of the on-going demographic transition to further reduce fertility; and responsible parenthood. Moreover, Kenya, as a country partner of the CSDH, was involved in the “Country Work Stream” aiming at turning evidence on the social determinants of health and health equity into effective policies. Whereas the National Reproductive Health Policy that was launched in 2007 did not overtly address issues of social determinants of health, the National Reproductive Health Strategy of 2009–2015 alluded to these. It was noted that the goal of reducing health inequities can only be achieved effectively by involving the population in decisions, mobilization, devolving and allocation of resources. The community strategy was aimed at enhancing community access to health care so as to improve productivity, which in turn would lead to reduction in poverty, hunger, child and maternal deaths [43]. Similarly the Second National Health Sector Strategic Plan of Kenya Annual Operational Plan 6 of July 2010–June 2011 reiterated the need to address equity through the community strategy [44]. However the results of the above efforts are yet to be realized.