What prevents pregnant women from adhering to the continuum of maternal care? Evidence on interrelated mechanisms from a cohort study in Kenya

Objectives To examine the determinants of the continuum of maternal care from an integrated perspective, focusing on how key components of an adequate journey are interrelated. Design A facility-based prospective cohort study. Setting 25 health facilities across three counties of Kenya: Nairobi, Kisumu and Kakamega. Participants A total of 5 879 low-income pregnant women aged 13-49 years. Outcome measures Ordinary least squares, Poisson and logistic regression models were employed, to predict three key determinants of the continuum of maternal care: (i) the week of enrolment at the clinic for antenatal care (ANC), (ii) the total number of ANC visits and (iii) utilisation of skilled birth attendance (SBA). The interrelationship between the three outcome variables was assessed with structural equation modeling. Results Each week of delayed enrolment in ANC reduced the number of ANC visits by 3% (incidence rate ratio=0.967, 95% CI 0.965 to 0.969). A higher number of ANC visits increased the relative probability of using SBA (odds ratio=1.28, 95% CI 1.22 to 1.34). The direct association between late enrolment and SBA was positive (odds ratio=1.033, 95% CI 1.02 to 1.04). Predisposing factors (age, household head’s education), enabling factors (wealth, shorter distance, rural area) and need factors (risk level of pregnancy, multigravida) were positively associated with adherence to ANC. Conclusion The results point towards a domino-effect and underscore the importance of enhancing the full continuum of maternal care. A larger number of ANC visits increases SBA, while early initiation of the care journey increases the number of ANC visits, thereby indirectly supporting SBA as well. These beneficial pathways counteract the direct link between enrolment and SBA, which is partly driven by pregnant teenagers who both enrol late and are at heightened risk of complications, stressing the need for specific attention to this vulnerable population. This study is based on a conceptual framework adapted from Andersen’s model of health services utilisation.39 Andersen’s model explains healthcare utilisation with contextual and individual characteristics, health behaviours and outcomes. Our study focuses on the individual rather than the contextual aspects, using individual care journey data to understand the basis of women’s behaviour in using maternal care. Figure 1 shows the adapted framework. Healthcare utilisation is mainly influenced by three categories of individual characteristics: predisposing, enabling and need factors.39 Predisposing factors indirectly influence the use of health services, encompassing demographics and social factors such as gender, age, education, occupation, ethnicity, social network and health beliefs. Predisposing factors available for our analysis are the mother’s age at conception and the household head’s education level. Enabling variables expedite or impede the use of healthcare services. The enabling factors used in this study are wealth status, county indicators and travel time to the facility. Need factors play a fundamental role in the use health services. They capture health-related determinants of the woman’s decision to seek maternal care. This encompasses perceived needs, capturing subjective perceptions about the required health services during pregnancy, as well as evaluated needs, referring to professional and objective measurements, diagnoses and estimated risk levels of the pregnancy.39 The need factors included in the analysis are the risk level as diagnosed by a health professional during pregnancy, having had a previous pregnancy and prior utilisation of maternal care services. The conceptual framework for adherence to maternal care. ANC, antenatal care. Conceptual framework adapted from Andersen’s framework. Kenya is classified as a lower-middle-income country. The country’s total population size was estimated to be 48 million in 2019, with 50.5% women, of which 57.5% were of reproductive age.40 The study is based on data gathered from three counties: Nairobi (urban, estimated population 4.3 million), Kisumu (relatively periurban, estimated population 1.2 million of which 61.8% rural) and Kakamega (mostly rural, estimated population 1.9 million of which 90.0% rural). In combination, they provide a diverse overview of the maternal health-seeking behaviour of pregnant women in both rural and urban areas. The MMR in Kenya is 62% higher than the world average.4 In 2014, only 19.8% of Kenyan mothers initiated their ANC visits in the first trimester.41 Coverage of the recommended four ANC visits was 57.6 %.15 In 2019, 85.7 % of pregnant women gave birth at a healthcare facility using SBA.42 Almost 40 % of all neonatal deaths in Kenya are related to inadequate check-ups for pregnancy complications.43 Attending at least two ANC visits has been shown to decrease the probability of a stillbirth by half in Kenya.44 The data are drawn from the MomCare project, which incentivises ‘access and adherence to care’-journeys through a digitally enabled ‘smart contract’. As an initiative of the PharmAccess Group, MomCare enrols pregnant women in a partly or fully subsidised health insurance programme, offering a ‘health wallet’ on their mobile phone, which they can use to check-in and pay at a selected network of clinics. The health wallet runs on a mobile platform (m-tiba45 by Carepay Limited46) that enables the MomCare analytics engine to collect real-time medical data, send reminders for check-ups and nudges to women to use care, and reward providers financially for quality care provision when women complete their maternal care journey. As such, MomCare promotes transparency over pregnancy status, delivered care and funds allocation across all agents (patients, providers, payers) during the entire care process. MomCare started in November 2017 and was operating in 25 health facilities by the end of the study period in August 2020. These facilities were connected to the m-tiba platform and received support through SafeCare, a quality improvement programme.47 The MomCare bundle covers the following basic maternal care services: four ANC consultations with related lab tests and vitamin complements, ultrasound scan, extra clinic consultations to treat pregnancy-related complications, normal and complicated delivery, two post-natal care consultations and three immunisations for the newborn. The study is designed as a prospective cohort study covering the period of February 2019 to August 2020. The study sampling frame includes all pregnant women who presented at one of the MomCare clinics during the study period, and who were eligible for the MomCare program. According to the eligibility criteria, enrolment should take place within the first 26 weeks of pregnancy, except for teenagers who could enrol at any pregnancy stage. Data were only collected from MomCare clinics, precluding a comparison with non-MomCare clinics. Eligible women presenting at the health facilities are onboarded on MomCare as follows: they register, receive information on the MomCare bundle, consent to participate, and participate in a baseline survey. In total, 11 538 eligible women enrolled in one of the 25 MomCare-connected clinics since programme inception in November 2017 (see online supplemental figure A1 for the sampling strategy). The study period starts from February 2019 onwards, because the baseline questionnaire was not standardised before that moment. As a result, 856 women who enrolled before February 2019 (7.4 %) were excluded from the study sample. To be able to conduct the analyses of enrolment, ANC visits and SBA on the same sample of women, we excluded 4 799 women (40.5 %) from the sample without SBA information (whose pregnancy was less than 42 weeks in August 2020, and who had not yet delivered in a MomCare clinic). This yields a sample of 5 883 women between 13–49 years. Missing observations in one or more outcome variables reduce the final sample further to 5 879. With this sample size, the analysis is powered (at β=0.80 and α=0.05) to correctly estimate the week of enrolment within 0.2 weeks of the true population average, the number of ANC visits within 0.05 visits of the true population average, and SBA within 4.0 % of the true population average.48 bmjopen-2021-050670supp001.pdf The analysis is based on the data collected through the MomCare analytics engine, that is, data collected from the survey at enrolment as described in the Study setting section, and throughout the mother journey via the medical information submitted on m-tiba by the healthcare providers.49 Key advantages of using real-time data collected through the analytics engine are the reduced recall bias and increased probability of accurate reporting,38 50 especially when compared with data based on, for example, Demographic and Health Surveys that rely on women’s retrospective self-reports with recall periods of up to 5 years.51 The MomCare baseline survey recorded information about women’s demographic and socioeconomic characteristics, including age, education of the household head, household size, dwelling information, wealth indicators, and parity as well as obstetric history. Medical records in the MomCare analytics engine contain information about the week of enrolment; the number of ANC visits; diagnoses, drugs and tests associated with each ANC visit; risk-level of the pregnancy; type of delivery and complications during delivery. The analyses are based on three primary outcome variables: the week of enrolment at a MomCare clinic, the total number of ANC visits at a MomCare clinic and having a skilled delivery at a MomCare clinic. The data do not capture visits at non-MomCare facilities. The first visit at the MomCare clinic of 142 women was recorded as a normal check-up rather than an ANC visit. These observations are kept in the dataset but not counted as an ANC visit in the analyses. Explanatory variables are classified into predisposing, enabling and need factors in line with the adapted conceptual framework in figure 1. Age is included as a continuous variable. Education level is a categorical variable measuring the highest completed education level of the household head (at most primary completed, at most secondary completed, tertiary completed). A wealth index was created based on the first loading of a principal component analysis on the total sample using data on households’ ownership of selected assets and dwelling characteristics, presence of electricity, education of household head, household size and means of transportation to the clinic. The population was then ranked based on the wealth index and assigned to three wealth terciles: low, middle and high. Distance to the clinic was measured as a dummy variable equal to 1 for travel time greater than 30 min. County indicators were included for Nairobi, Kisumu and Kakamega (as proxies for urban, periurban and rural). The pregnancy risk level is included as a categorical variable, ranging from low risk1 for normal pregnancy without any additional complications, medium risk2 for pregnancies with non-life-threatening diagnoses (such as urinary tract infections or gestational diabetes) to high risk3 for severe conditions. The risk level is determined by healthcare professionals at MomCare facilities based on medical diagnosis; it measures the maximum risk level attained at any point during enrolment.49 We emphasise that low-risk pregnancies still require the recommended minimum of four ANC visits. Furthermore, ‘previously pregnant’ is a dummy variable equal to 1 if the woman had been pregnant before (multigravida). The analysis investigates four main research questions, that is, what are the determinants of (i) week of enrolment, (ii) number of ANC visits, and (iii) SBA and (iv) how are these outcomes interrelated? The analysis first estimates separate regressions for each of the three outcomes. The model’s parameters are predicted with ordinary least square (OLS) regression, Poisson regression and logistic regression, respectively. Poisson regression is suitable to estimate equation 2 since the null hypothesis of the goodness-of-fit χ2 chi-squared test (H0: X~Poisson) is not rejected. In addition, the mean and variance of the ANC variable (2.971 and 2.975, respectively) show that the dependent variable is not over-dispersed and does not have an excessive number of zeros. The analyses subsequently introduce subsets of explanatory variables (predisposing, enabling and need factors) in a stepwise manner to control for potential confounding effects. Finally, the simultaneous relationships between the three outcome variables are assessed with structural equation modeling (SEM), using the same estimation methods and explanatory variables as for the separate regressions. The determinants of the week of enrolment, the number of ANC visits and utilisation of SBA are estimated with the following consecutive specifications: Where the subscript i indicates the individual. The variables included in Xi are age, education of household head, wealth tercile, distance to clinic dummy, county indicators, pregnancy risk level and previously pregnant, as described in the Variables section. ui, vi and εi are the individual error terms. Standard errors (SEs) are robust to allow for heteroscedasticity. To capture the continuum of care, the interrelationship between the week of enrolment, the number of ANC visits and utilisation of SBA is predicted with the following system of SEM equations: Where ψi and ξi are the individual error terms, and SEs are robust. All analyses are carried out using Stata V.16.0. Each pregnant participant voluntarily consented to join MomCare. MomCare made use of learnings deriving from the collected data and the providers’ experiences to adapt the care bundle to the mothers’ needs. Providers interacted directly with the mothers communicating about MomCare.