Accelerating Kenya’s progress to 2030: Understanding the determinants of under-five mortality from 1990 to 2015

Introduction Despite recent gains, Kenya did not achieve its Millennium Development Goal (MDG) target for reducing under-five mortality. To accelerate progress to 2030, we must understand what impacted mortality throughout the MDG period. Methods Trends in the under-five mortality rate (U5MR) were analysed using data from nationally representative Demographic and Health Surveys (1989-2014). Comprehensive, mixed-methods analyses of health policies and systems, workforce and health financing were conducted using relevant surveys, government documents and key informant interviews with country experts. A hierarchical multivariable linear regression analysis was undertaken to better understand the proximal determinants of change in U5MR over the MDG period. Results U5MR declined by 50% from 1993 to 2014. However, mortality increased between 1990 and 2000, following the introduction of facility user fees and declining coverage of essential interventions. The MDGs, together with Kenya’s political changes in 2003, ushered in a new era of policymaking with a strong focus on children under 5 years of age. External aid for child health quadrupled from 40 million in 2002 to 180 million in 2012, contributing to the dramatic improvement in U5MR throughout the latter half of the MDG period. Our multivariable analysis explained 44% of the decline in U5MR from 2003 to 2014, highlighting maternal literacy, household wealth, sexual and reproductive health and maternal and infant nutrition as important contributing factors. Children living in Nairobi had higher odds of child mortality relative to children living in other regions of Kenya. Conclusions To attain the Sustainable Development Goal targets for child health, Kenya must uphold its current momentum. For equitable access to health services, user fees must not be reintroduced in public facilities. Support for maternal nutrition and reproductive health should be prioritised, and Kenya should acknowledge its changing demographics in order to effectively manage the escalating burden of poor health among the urban poor. For both the main case study3 and the current analysis, we used a Countdown-adapted conceptual framework (online supplementary eFigure 1) to guide our approach to understanding RMNCAH in Kenya. This framework incorporates each of the Countdown-specific domains that will contribute to or detract from progress in RMNCAH: health systems and policy, health financing, coverage, equity and mortality (including the Lives Saved Tool). While we have previously examined coverage, equity and mortality, the current paper will address the remaining domains of health systems, policies and financing over the MDG period. The conceptual framework reflects our broader objectives of understanding RMNCAH change in Kenya; thus, some aspects of maternal and newborn health are reported within these domains, despite the paper’s focus on child health. Raw data from nationally representative Kenya Demographic and Health Surveys (K-DHS) from 1989 to 20142 5–9 were used to generate under-five mortality estimates over the MDG period (1990–2015). For simplicity of interpretation, we refer to the year of the survey (1993, 2003 and 2014) in our results, though U5MR estimates actually reflect child mortality for the survey’s preceding 5-year period. Coverage indicators were defined as per the Countdown to 2015 guidelines.10 A Composite Coverage Index (CCI) was calculated to present an overall picture of intervention coverage in Kenya11 (online supplementary etable 1). CCI is comprised of the following set of eight preventive and curative interventions: (1) demand for family planning satisfied, (2) skilled birth attendance, (3) antenatal care with a skilled provider (ANC), (4) DPT3 vaccination, (5) measles vaccination, (6) BCG vaccination, (7) oral rehydration therapy and continued feeding for children with diarrhoea and (8) care seeking for children with suspected pneumonia. Use of CCI as a summary statistic is a novel feature among standardised Countdown country case studies. K-DHS 2014 was powered at the county level, while all previous DHS surveys are powered for provincial estimates. Statistical packages R and Win bug 14 were used to estimate prevalence, and ArcGIS10 was used to create high-resolution maps for visualisation of CCI across counties. To assess policy and systems changes relevant to RMNCAH in Kenya from 1990 to 2014, we used the following three standardised tools developed by the Countdown Health Systems and Policies Technical Working Group and applied in previous Countdown case studies12–14: (1) Policy and Program Timeline Tool; (2) Health Policy Tracer Indicators Dashboard; and (3) Health Systems Tracer Indicator Dashboard.4 Data for these tools were primarily populated through the review of existing published and grey literature, including peer-reviewed publications, Ministry of Health (MoH) policy and strategy documents, MoH reports, WHO and United Nations agencies reports and databases, for example, data from the WHO Global Maternal, Newborn, Child and Adolescent Health policy indicators database.15 Health workforce and facility data were obtained from the 2004 Service Availability Mapping survey16 and the most recent WHO Service Availability and Readiness Assessment Mapping for Kenya.17 Spider plots were used to visually depict status of policy implementation, whereby each major policy was dissociated into related subpolicies, and implementation level (%) was mapped within the plot. Spider plots were based on up-to-date data from the WHO database.15 To examine the relationship between health systems inputs and service accessibility and delivery, we overlaid health systems data (health workforce, health facility density, health services budget and lifesaving commodities) on a map of CCI. We also conducted key informant interviews with stakeholders from the MoH, Division of Family Health, WHO and the World Bank as part of the methodology for populating, verifying and analysing data in the Policy and Program Timeline Tool.4 For the health financing analysis, we undertook a review of the existing literature to describe the evolution of health financing policy in Kenya. Information from National Health Accounts (NHA) for fiscal years 2005/2006, 2009/2010 and 2012/2013,18 Health Facts and Figures 2014 19 compiled by the Ministry of Health, the Countdown to 2015 database on official development assistance (ODA) for RMNCAH20 and the 2003 and 2013 Kenya Household Health Expenditure and Utilisation Survey21 22 were used in order to determine trends in health spending. Multivariable analyses were performed to determine factors associated with improvements in child survival across the 1993–2014 MDG period. We evaluated data from three K-DHS surveys (1993, 2003 and 2014) to reflect the early (1990–1999), mid (2000–2009) and late (2010–2015) MDG periods. Robust and representative U5MR estimates were available for these survey years. Maternal mortality was variable and available for only two time points (2003 and 2014), and neonatal mortality did not sufficiently decline to perform a robust analysis of change. Due to the nested nature of the K-DHS data, mixed-effects, multilevel logistic regression models were used. Hierarchical model building strategies were used to determine multivariable predictors of U5MR.23 Using evidence-based child survival frameworks,24 25 variables were mapped into two levels that corresponded to intermediate (level 2) and proximal (level 1) determinants of child mortality. Intermediary factors included geographical region, residence, access to improved water and sanitation facilities, maternal and paternal education and household wealth index, a composite measure based on household assets. At the proximal level, we examined individual child and parent characteristics including gender, birth order, birth interval, size at birth, timing of breastfeeding initiation, age of mother at birth, contraceptive use and parity. For each survey (1993, 2003 and 2014), the dependent variable was U5MR, while the independent variables encompassed all intermediate and proximal determinants listed above. Bivariate associations between the determinant and U5MR were assessed using ORs and corresponding p values/95% CIs. Factors statistically significant at p<0.20 were entered into multivariable modelling within the respective level and retained if p<0.15. Multicollinearity was assessed using variance inflation factors (VIF), where variables with VIF >3 were considered collinear.26 Additionally, a correlation matrix for all predictive variables is appended (online supplementary eTable 2). A secondary analysis was conducted to determine the effect of change in the determinants on change in U5MR between surveys. These methods are based on similar analyses by Hong et al 27 conducted using DHS data from three time points in Rwanda. Changes in the distributions of indicators between surveys (2014 compared with 2003 and 2014 compared with 1993) were multiplied by the coefficients of variables obtained from the final 2014 multivariable model of mortality, as described above. The products were then summed, and the results were exponentiated to obtain the change in mortality that was due to changes in the values of the indicators between surveys.27 In other words, this procedure allows you to determine how much of the change in child mortality from 1993 to 2014 is due to shifting proportions of the explanatory variables between surveys. To account for the impact of the HIV/AIDS epidemic on child mortality in Kenya, we have included a dedicated section on HIV/AIDS trends. The first Kenya Aids Indicator Survey (KAIS) took place in 2007, after the peak of the HIV epidemic in Kenya. Before KAIS, the 2003 K-DHS was the first national, household-based survey to include HIV testing.8 As such, we used both K-DHS data and modelled data from the Joint United Nations Programme on HIV/AIDS (UNAIDS)28 to report HIV epidemic trends throughout the MDG period in Kenya (1990–2015). Coverage of antiretroviral treatment (ART) and indicators relating to prevention of mother-to-child transmission were not available from UNAIDS until 2010. The funders of the study had no role in study design, data collection, data analysis, data interpretation or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.