Menu
Menu
Menu
Menu
Background Progress in reproductive, maternal, newborn, and child health (RMNCH) in Kenya has been inconsistent over the past two decades, despite the global push to foster accountability, reduce child mortality, and improve maternal health in an equitable manner. Although several cross-sectional assessments have been done, a systematic analysis of RMNCH in Kenya was needed to better understand the push and pull factors that govern intervention coverage and influence mortality trends. As such, we aimed to determine coverage and impact of key RMNCH interventions between 1990 and 2015. Methods We did a comprehensive, systematic assessment of RMNCH in Kenya from 1990 to 2015, using data from nationally representative Demographic Health Surveys done between 1989 and 2014. For comparison, we used modelled mortality estimates from the UN Inter-Agency Groups for Child and Maternal Mortality Estimation. We estimated time trends for key RMNCH indicators, as defined by Countdown to 2015, at both the national and the subnational level, and used linear regression methods to understand the determinants of change in intervention coverage during the past decade. Finally, we used the Lives Saved Tool (LiST) to model the effect of intervention scale-up by 2030. Findings After an increase in mortality between 1990 and 2003, there was a reversal in all mortality trends from 2003 onwards, although progress was not substantial enough for Kenya to achieve Millennium Development Goal targets 4 or 5. Between 1990 and 2015, maternal mortality declined at half the rate of under-5 mortality, and changes in neonatal mortality were even slower. National-level trends in intervention coverage have improved, although some geographical inequities remain, especially for counties comprising the northeastern, eastern, and northern Rift Valley regions. Disaggregation of intervention coverage by wealth quintile also revealed wide inequities for several health-systems-based interventions, such as skilled birth assistance. Multivariable analyses of predictors of change in family planning, skilled birth assistance, and full vaccination suggested that maternal literacy and family size are important drivers of positive change in key interventions across the continuum of care. LiST analyses clearly showed the importance of quality of care around birth for maternal and newborn survival. Interpretation Intensified and focused efforts are needed for Kenya to achieve the RMNCH targets for 2030. Kenya must build on its previous progress to further reduce mortality through the widespread implementation of key preventive and curative interventions, especially those pertaining to labour, delivery, and the first day of life. Deliberate targeting of the poor, least educated, and rural women, through the scale-up of community-level interventions, is needed to improve equity and accelerate progress. Funding US Fund for UNICEF, Bill & Melinda Gates Foundation.
We did a systematic review of all available electronic data published, in English, between Jan 1, 1990, and May 31, 2015, as well as unpublished data. All data pertained to the situation analysis of RMNCH in Kenya from 1990–2015, including information about socioeconomic development, relevant policies, programme strategies and interventions, and official reports about progress towards MDGs. We estimated trends and the average annual rate of reduction (ARR) of maternal, stillbirth, neonatal (≤1 month), and under-5 mortality for the period 1989–90 to 2014–15. Data sources are listed in the appendix.2, 3, 5, 9, 10, 11, 12, 13, 14 For each mortality outcome, we retained both modelled and sampled estimates when available. We modelled all cause-of-death data with the LiST. Projections of rates to 2030 were based on two scenarios: current trends, estimated with the observed ARR, and the accelerated ARR required to achieve the Sustainable Development Goal targets for maternal (70 per 100 000 livebirths), newborn (12 per 1000 livebirths), and child (25 per 1000 livebirths) mortality.15 To measure health outcomes and trends in maternal, newborn, and child intervention coverage in Kenya between 1989 and 2014, we analysed raw data from the Kenya Demographic and Health Surveys (K-DHS). These surveys provide comprehensive estimates at both national and subnational levels. Average annual rates of change for the periods 1989–2003, 2003–14, and 1989–2014 were calculated for key interventions to better visualise trends in coverage. All coverage indicators were defined according to Countdown to 2015 guidelines.16 Because of small sample sizes at the country level, we used Bayesian spatial models to estimate key RMNCH indicators for these small geographical areas. We estimated Bayesian posterior prevalence rates for K-DHS surveys in 2003, 2008–09, and 2014, using covariates and standardised procedures that are detailed elsewhere.17 We used R (version 3.2.5) and WinBUGS (version 14) to calculate the estimated prevalence, and ArcGIS10 (version 10) to create high-resolution maps for visualisation of coverage of key indicators across counties. We explored key correlates of RMNCH intervention coverage in the period with the greatest growth (2003–14). We did ecological multivariable regression using county as the unit of analysis. Analysed outcomes included coverage of family planning for women married or in union, skilled birth assistance, and full immunisation of children aged 12–23 months. We chose these outcomes on the basis of their unique positions in the continuum of care and different delivery strategies. We modelled outcomes as absolute differences in Bayesian prevalence estimates calculated for each of the 47 counties (difference between 2014 and 2003). Because of a scarcity of available covariate data at baseline (2003), we examined potential determinants using best-estimate fixed values for 2014. We adapted a hierarchical modelling approach, as suggested by Victora and colleagues,18 for similar datasets. We conceptualised three levels of potential determinants for each outcome (appendix). At the distal level (level 3) we examined a range of socioeconomic factors; the intermediary (level 2) variables included health service accessibility factors; and the proximal (level 1) variables were individual, household, and behavioural factors that could affect coverage levels. We explored univariate distributions of outcome and predictors using mean or median, frequency or percent, and histograms, as appropriate. We estimated crude associations between the covariate and the change outcome (and further adjusted for baseline values of the coverage outcome) via ordinary least-squares regression, and evaluated linear slope coefficients and partial R2 values. Covariates that were significant at liberal cutoff (p<0·20) in the crude analysis were incorporated into a series of model-building strategies at the respective conceptual level. We used a combination of highest adjusted R2 modelling and backward elimination model-building strategies to determine the final set of covariates at each level; variables were retained if p values were less than 0·15. In line with Victora's modelling strategy,18 factors that were significant in the distal model were retained in the intermediate model (irrespective of statistical significance), and factors that were significant in the combined set were retained in the proximal model. We iteratively evaluated assumptions of ordinary least-squares regression, including homoscedasticity and linearity, and found no important violations. We examined deviance and influence statistics, and used Akaike and Bayesian information criteria and R2 values to assess final model fit. We examined multicollinearity among predictors using variance inflation factors, whereby variance inflation factors greater than 3 indicated highly collinear variables. The type 1 error rate was 0·05 and we did statistical analyses with SAS (version 9.4). The equity analysis was based on the stratified coverage of 11 interventions across the continuum of care according to household level wealth index. The appendix provides definitions of indicators. Household wealth was divided into five standard quintiles (Q1–5) and was developed on the basis of asset indices. We also calculated a composite coverage index (CCI)19 to present an overall picture of intervention coverage in Kenya. A five-dot plot was developed to visualise differences in intervention coverage with respect to wealth quintiles. All analyses were done in Stata (version 12). Analyses were weighted with DHS sample weights to restrict variability among sampling of different regions and ensure data were representative of the population. We used the LiST to model the effect of scale-up of RMNCH interventions on maternal, neonatal, and child mortality. The appendix provides a complete list of interventions selected for modelling. We obtained the current level of coverage for each intervention from the latest available K-DHS (2014) estimates. When coverage data were unavailable for any intervention, estimates were made on the basis of known coverage of other interventions, as described in the LiST manual. For interventions not available in DHS, we used default data sources (appendix). For the prospective LiST model, 2016 was used as the baseline year and coverage was scaled up in two pragmatic scenarios: from base year to 2025, and from 2025 to 2030. The first scenario assumed scale-up of all interventions at first pragmatic level of coverage until 2025. We defined realistic targets for each intervention: target coverage of 50% if current level was less than 30%, target coverage of 70% if current coverage was above 30% but lower than 70%, and target coverage of 90% if current coverage was above 70%. In the second scenario, we scaled up coverage of interventions from the level attained in the year 2025 to the next pragmatic level by 2030. We also did retrospective LiST analyses to understand the effect of specific interventions on the observed mortality trends between 1993 and 2014. We further assessed the effect of scale-up of 13 community-delivered and primary care interventions to 90% of the current coverage by 2030 in different wealth quintiles, on the basis of the K-DHS 2014 data (appendix). We used LiST to model the cause-of-death structure for maternal, neonatal, and under-5 deaths using coverage levels from DHS and default data sources, when applicable. We also calculated cause-of-death estimates by quintile of socioeconomic status by assuming changes in coverage from the national level to the levels measured in each of the wealth subgroups.20 The institutions that commissioned, funded, or administered the surveys were responsible for all ethics procedures. 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.
