Do active patients seek higher quality prenatal care?: A panel data analysis from Nairobi, Kenya

Despite poverty and limited access to health care, evidence is growing that patients in low-income countries are taking a more active role in their selection of health care providers. Urban areas such as Nairobi, Kenya offer a rich context for studying these “active” patients because of the large number of heterogeneous providers available. We use a unique panel dataset from 2015 in which 402 pregnant women from peri-urban (the “slums” of) Nairobi, Kenya were interviewed three times over the course of their pregnancy and delivery, allowing us to follow women’s care decisions and their perceptions of the quality of care they received. We define active antenatal care (ANC) patients as those women who switch ANC providers and explore the prevalence, characteristics and care-seeking behavior of these patients. We analyze whether active ANC patients appear to be seeking out higher quality facilities and whether they are more satisfied with their care. Women in our sample visit over 150 different public and private ANC facilities. Active patients are more educated and more likely to have high risk pregnancies, but have otherwise similar characteristics to non-active patients. We find that active patients are increasingly likely to pay for private care (despite public care being free) and to receive a higher quality of care over the course of their pregnancy. We find that active patients appear more satisfied with their care over the course of pregnancy, as they are increasingly likely to choose to deliver at the facility providing their ANC. The study was conducted between February and September of 2015 in 24 peri-urban neighborhoods of Nairobi within Kiambu and Nairobi counties. These densely populated areas surrounding Nairobi are within 12–15 km of the city center and are primarily made up of low-income residential estates shared with industrial enterprises, especially in locations closer to the city center. These areas are characterized by a large number of public, private, and faith-based health facilities ranging from small pharmacies and outpatient care to large hospitals with maternity wards; these facilities also range widely in cost, size and services available. Study neighborhoods were selected based on: 1) the availability of both private and public facilities for ANC and delivery, 2) a composition of primarily lower-income residents, and 3) meeting a minimum level of security. Community health workers were engaged during the selection process to ensure these criteria were fulfilled. The fertility rate in Nairobi slums is 3.5 children per woman (APHRC, 2014) compared to the Nairobi provincial rate of 2.8 (KNBS, 2010). Pregnant women (self-reported gestational age of 5, 6, or 7 months) aged 18 years and above were recruited through convenience sampling during a planned recruitment event within the study neighborhoods. During these events, field staff were stationed in community centers (e.g. markets and pedestrian intersections) where pedestrian traffic was high and where they could easily engage with interested community members and/or pregnant women passing through. Snowball sampling was used to supplement recruitment efforts: interested community members were asked if they knew pregnant women within the community who might be interested in the study and if so, they were encouraged to share the study flyer and/or come to speak with our field staff for more information. Eligible respondents were visited at their residence to obtain informed consent, ensure eligibility, and administer the baseline survey. The baseline survey occurred, on average, at 27.1 weeks gestation (median: 27.9; 95% CI: 26.7–27.5). A midline survey was administered at the respondent’s home or work place and was scheduled to take place during her 8th month of pregnancy, occurring, on average, at 33.7 weeks gestation (median: 33.7; 95% CI: 33.5–33.9). A final survey was scheduled for 2–4 weeks post-partum and occurred, on average, at 3.5 weeks after delivery (median: 2.9; 95% CI: 3.3–3.7). At the baseline visit, all women were asked about any ANC appointments they had for this pregnancy up until the day of the survey. For each subsequent survey, women were asked about any ANC appointments occurring since the last survey. For all ANC visits reported, confirmation of the visit and date was attempted in the ANC book, a small booklet provided at all Kenyan facilities documenting a patient’s ANC history per pregnancy. 71.7% of visits were able to be confirmed in the ANC book, with no significant difference in ANC book confirmation between active and non-active patients (p = 0.626). Prior to administering baseline, a sub-sample of women (n = 111) was randomly selected to be administered an abridged version of the baseline and midline surveys in case the extensive questions related to birth planning were found to influence behavior. Only one of our outcome variables is unavailable for the 27.6% of respondents given the abridged survey. A total of 553 women were surveyed at baseline, 459 at midline and 454 at endline (Fig. 1). Of the 553 surveyed at baseline, 21 women withdrew from the study, 21 could not be tracked for either the midline survey or endline survey or both, and 55 relocated out of the study area. Most relocations were to places outside of Nairobi and were temporary, as it is common to stay with family just before and after the birth of a child. A further 25 women delivered before midline, and there was one maternal death, 21 neonatal deaths and 5 miscarriages. 404 women were surveyed at all three visits. Among these, 2 had no ANC appointments and are dropped from the sample, leaving an analysis sample of 402 women. A total of 1621 ANC visits were reported on in our sample, occurring at a total of 165 different facilities. Survey samples and reasons for attrition. We define “active patients” as those women who switch ANC providers at least once during pregnancy. Of the 402 women in our sample, 139 (34.6%) were active patients. Outcomes were derived from self-reported data (with the occurrence and date of ANC appointments confirmed in booklets 71.7% of the time) about ANC and delivery care collected during the three surveys. These included the frequency, timing, location, and quality of ANC visits, as well as the type of facility attended for ANC (private, public or other). An equally-weighted 6 point quality index was generated to capture whether respondents received various essential services during their visit and included the following components: whether the patients’ weight was checked, blood pressure checked, fundal height measured, whether the baby’s heart rate was measured, whether a urine sample was collected, and whether iron supplements were given. These measures were based on ANC quality measures captured in the 2014 Kenya Demographic and Health Survey and adapted to include baby’s heart rate measurement and iron supplementation based on advice from maternal health experts in Kenya. We explore whether these services were conducted at each ANC visit. We also present results for each component of the index separately. We construct two measures of patients’ perceptions of the quality of care. First, we construct a binary variable equal to one if the respondent stated she would rate the quality of care she received as “excellent” and zero if she rated it “poor”, “fair” or “good”. The second measure is based on a question derived from a facility ranking exercise conducted with women during the baseline and midline surveys. In this exercise women were asked to list all of the facilities they were considering for delivery and were then asked to rank them relative to the other facilities being considered along a number of dimensions, including perceptions of overall quality. From this, we construct a binary variable equal to one if the woman ranked the facility she was using for ANC highest in terms of overall quality and zero otherwise. Finally, we report on whether a woman ultimately delivered at the facility where she attended ANC, a measure of the patient’s satisfaction with care. Since not all ANC facilities offer delivery services, we also report this outcome for the restricted sample of women attending ANC at facilities that offer delivery services. In order to measure changes in the quality of care during pregnancy, we report outcomes for each of the first three ANC visits. 90% of women in our sample had at least three ANC visits, but only 66% had at least four visits. Since we want to demonstrate how care changes over the course of pregnancy for the majority of our sample—without confounding variation from large changes in sample composition—we restrict the analysis to the first three visits. Results for the fourth visit are presented in an appendix for comparison. We test for differences in characteristics of active and non-active patients by running simple linear regressions of the variable of interest (demographic characteristics as well as characteristics of the pregnancy and delivery) on a binary variable for whether the patient is “active” and a constant term. We present the coefficient on the “active” variable, representing the difference in means between active and non-active patients, and the two-tailed p-value for the null hypothesis that the coefficient on “active” is equal to zero. We use the same approach to test for differences in our outcome measures related to quality of care, except that each regression is run for the first, second and third ANC visit separately. Since many of our outcome measures are binary and the quality index is an event count variable, to check the robustness of our results to specification choice we also present alternative specifications with logistic regression for binary outcomes and Poisson regression for the quality index in an appendix table. In all regressions, robust standard errors are clustered at the neighborhood level. The protocol including all study materials was approved in the United States by Harvard T. H. Chan School of Public Health’s Institutional Review Board and in Kenya by the Amref Health Africa (formerly, AMREF) Ethics and Scientific Review Committee.