Poor quality for poor women? Inequities in the quality of antenatal and delivery care in Kenya

Background: Quality of healthcare is an important determinant of future progress in global health. However, the distributional aspects of quality of care have received inadequate attention. We assessed whether high quality maternal care is equitably distributed by (1) mapping the quality of maternal care in facilities located in poorer versus wealthier areas of Kenya; and (2) comparing the quality of maternal care available to Kenyans in and not in poverty. Methods: We assessed three measures of maternal care quality: facility infrastructure and clinical quality of antenatal care and delivery care, using indicators from the 2010 Kenya Service Provision Assessment (SPA), a standardized facility survey with direct observation of maternal care provision. We calculated poverty of the area served by antenatal or delivery care facilities using the Multidimensional Poverty Index. We used regression analyses and nonparametric tests to assess differences in maternal care quality in facilities located in more and less impoverished areas. We estimated effective coverage with a minimum standard of care for the full population and those in poverty. Results: A total of 564 facilities offering at least one maternal care service were included in this analysis. Quality of maternal care was low, particularly clinical quality of antenatal and delivery care, which averaged 0.52 and 0.58 out of 1 respectively, compared to 0.68 for structural inputs to care. Maternal healthcare quality varied by poverty level: at the facility level, all quality metrics were lowest for the most impoverished areas and increased significantly with greater wealth. Population access to a minimum standard (≥0.75 of 1.00) of quality maternal care was both low and inequitable: only 17% of all women and 8% of impoverished women had access to minimally adequate delivery care. Conclusion: The quality of maternal care is low in Kenya, and care available to the impoverished is significantly worse than that for the better off. To achieve the national targets of maternal and neonatal mortality reduction, policy initiatives need to tackle low quality of care, starting with high-poverty areas. Copyright: We combined data from multiple sources for this analysis. We drew our sample of maternal care facilities in Kenya from the Service Provision Assessment (SPA) survey conducted by the DHS Program in 2010. SPA is a standardized, facility-based cross-sectional survey with direct observation of maternal care provision, designed to be representative of the health system (public and privately run facilities) at national and regional levels. We limited our sample of facilities to those providing at least one maternal care service, whether ANC or delivery care. Demographic data were extracted from two sources: the Oxford Poverty and Human Development Initiative and the 2014 population-representative DHS. The 2010 Constitution established 47 counties in Kenya. We obtained county boundaries from the GADM database version 2.8, a spatial database for global administrative areas, and the estimated 2010 population density from WorldPop (Creative Commons Attribution 4.0 International License).[19] For mapping purposes, we obtained the location of all health facilities in Kenya from Kenya Open Data. Analysis proceeded in two stages: description of poverty and maternal care quality at the county level and assessment of poverty as a predictor of facility quality based on facility catchment areas. We defined poverty using the multidimensional poverty index (MPI). The MPI measures poverty using ten indicators in three dimensions: education (years of schooling and school attendance), health (child mortality and nutrition) and standard of living (cooking fuel, sanitation, water, electricity, floor and asset ownership). The MPI reflects both the incidence and the average intensity of poverty. A person is identified as impoverished if she is deprived in at least one third of the weighted indicators. Using data from the 2008–2009 DHS, the Oxford Poverty and Human Development Initiative estimated MPI per square kilometer (km) for the full country, with accompanying uncertainty estimates based on width of the 95% credible interval.[20] We calculated number of those impoverished, average poverty and average uncertainty per county. Based on data showing that nearly 90% of women in Kenya lived within 5 km of a health facility,[21] we defined facility catchment area using a buffer zone with a 5-km radius and calculated average proportion of individuals in poverty within these areas. We classified counties and catchment areas into five levels (poorest to richest) using even intervals of 20% poverty. As a secondary metric, we extracted household wealth index from the 2014 DHS survey and calculated average wealth per county (weighted with household sampling weights). The 2014 DHS was designed to provide representative summaries at the county level. We classified counties into wealth quintiles based on the thresholds applied in defining wealth quintiles for households. The Institute of Medicine’s canonical report on quality of healthcare proposes that a high quality health system is safe, effective, patient centered, timely, efficient, and equitable.[22] We operationalized these measures using Donabedian’s framework of structure, process and outcome.[23] Structural elements, such as availability of medicines and equipment, represent necessary, but not sufficient, conditions for the delivery of a given quality of health care. Process indicators represent the closest approximation of the actual quality of health care offered, as these include both technical (appropriate delivery of clinical procedures) and interpersonal (client-provider interactions) aspects of healthcare delivery. Outcome indicators measure health improvements attributable to medical care, for example, under-five mortality rate. These measures, however, may be affected by factors other than quality of healthcare delivered that influence outcomes. Using Donabedian’s framework, we identified indicators for structural inputs to maternal care (infrastructure, staffing, and equipment) as well as clinical care processes in antenatal care and delivery care to cover the spectrum of maternal care. Indicators for maternal care structure were constructed using data extracted from the SPA facility audit and provider interviews. This index included thirty items covering infrastructure (e.g., functional water and electricity), staffing (e.g., availability of 24-hour delivery care, staff training in ANC or delivery care), and equipment and supplies, such as iron folate for ANC and injectable uterotonics for delivery. See S1 Fig for details. Facilities offering only one of ANC or delivery care were scored based on general facility infrastructure items and those elements relevant to the service provided. Clinical care processes were assessed using direct observation of ANC and deliveries in a subset of facilities. We created an ANC clinical quality index using forty actions providers are expected to perform during all first ANC visits, drawn from the Focused Antenatal Care Model Checklist;[24] sample items include assessment of client history, counseling on danger signs, and administration of tetanus toxoid injection and HIV test (S2 Fig). To measure clinical quality of delivery care, we applied the quality of the process of intrapartum and immediate postpartum care (QoPIIPC) metric validated by Tripathi et al.[25] to the SPA data, extracting indicators matching eighteen of the twenty QoPIIPC items. These included checking woman’s blood pressure, washing hands before any examination, and timely administration of uterotonic (S3 Fig). Clinical observations were averaged within each facility, weighted with the patient sampling weight rescaled within facility. For each quality index, indicators were averaged to provide a facility summary score from 0 to 1, with missing values excluded. Completeness of each indicator is shown in S1 Table; missingness was minimal for infrastructure and ANC observations, but substantial (up to 32%) for observed deliveries, particularly for items at early stages of labor that frequently occur outside of the health facility. We calculated patient load per facility based on reported delivery clients (including cesarean section as applicable) and reported ANC visits in the past 12 months. We first described quality of care and population in poverty at the county level. We calculated weighted averages of each quality metric per county, weighting facilities by total maternal care visits (deliveries plus ANC visits), total ANC visits, and total deliveries for the maternal infrastructure, ANC quality, and delivery quality indices respectively. We calculated the weighted standard error (SE) of each quality index within county, using the full sample standard deviation (SD) for counties with a single facility, and generated a 95% confidence interval (CI) for each county using this SE. We present summary statistics of county population and health facility access by strata of poverty level. We calculated average quality by level of poverty and used average SE by poverty level to generate 95% CIs for each estimate. A non-parametric test for trend was employed to assess differences in maternal care quality by county poverty level. To quantify population access to quality care, we defined a threshold of at least 0.75 on each index as minimally adequate maternal care quality. Given the lack of universally defined minimum quality standards, we selected this threshold on the premise that women should receive most of these basic items at minimum. We calculated access to adequate maternal care by summing the total population and impoverished individuals within counties with adequate care quality. We repeated this analysis using the bounds of the 95% CI for county quality estimates to quantify uncertainty around population access to quality care. In order to obtain a more precise understanding of quality of care accessible to poor women, we assessed poverty and quality at the facility level; we regressed each quality index on poverty level of the surrounding area, clustered by county to account for adjacent facilities with overlapping catchment areas. We used the poorest areas (>80% poverty) as the reference group for all models except for quality of delivery care: due to the small number of facilities with directly observed deliveries in the poorest quintile (N = 2), we collapsed the two poorest quintiles as the reference category for these models. We conducted multiple sensitivity analyses. To assess sensitivity to the definition of the catchment area, we repeated the analysis redefining the catchment area for hospitals to cover 20 km. To test robustness to uncertainty in the MPI estimates, we excluded facilities with average uncertainty greater than 50%. To assess the reproducibility of the results with an alternative poverty metric, we regressed average quality per county on wealth quintile as measured in the 2014 DHS. Statistical analysis was done using Stata 14.1 (StataCorp, Texas) and mapping using QGIS Version 2.12 (Free Software Foundation, Massachusetts) and ArcGIS 10.3.1 (Esri, California). The original survey implementers obtained ethical approvals for data collection; the Harvard University Human Research Protection Program deemed this secondary analysis exempt from human subjects review.