Menu
Menu
Menu
Menu
BACKGROUND: Despite progress toward meeting the Sustainable Development Goals, a large burden of maternal and neonatal mortality persists for the most vulnerable people in rural areas. We assessed coverage, coverage change and inequity for 8 maternal and newborn health care indicators in parts of rural Nigeria, Ethiopia and India. METHODS: We examined coverage changes and inequity in 2012 and 2015 in 3 high-burden populations where multiple actors were attempting to improve outcomes. We conducted cluster-based household surveys using a structured questionnaire to collect 8 priority indicators, disaggregated by relative household socioeconomic status. Where there was evidence of a change in coverage between 2012 and 2015, we used binomial regression models to assess whether the change reduced inequity. RESULTS: In 2015, we interviewed women with a birth in the previous 12 months in Gombe, Nigeria (n = 1100 women), Ethiopia (n = 404) and Uttar Pradesh, India (n = 584). Among the 8 indicators, 2 positive coverage changes were observed in each of Gombe and Uttar Pradesh, and 5 in Ethiopia. Coverage improvements occurred equally for all socioeconomic groups, with little improvement in inequity. For example, in Ethiopia, coverage of facility delivery almost tripled, increasing from 15% (95% confidence interval [CI] 9%-25%) to 43% (95% CI 33%-54%). This change was similar across socioeconomic groups (p = 0.2). By 2015, the poorest women had about the same facility delivery coverage as the least poor women had had in 2012 (32% and 36%, respectively), but coverage for the least poor had increased to 60%. INTERPRETATION: Although coverage increased equitably because of various community-based interventions, underlying inequities persisted. Action is needed to address the needs of the most vulnerable women, particularly those living in the most rural areas.
The study took place in 2012 and 2015 in Gombe State, Nigeria; in the 4 most populous regions of Ethiopia; and in the state of Uttar Pradesh, India. We focused on these 3 diverse settings because they reflect areas of interest of the Bill & Melinda Gates Foundation, a global health funder, and because they represent rural, poor populations experiencing a high number of maternal and neonatal deaths.13 Gombe State in northeast Nigeria has a population of 3.1 million. In 2015, the state had an estimated maternal mortality ratio of 1549 per 100 000 population and a neonatal mortality rate of 35 per 1000 population.14 In Ethiopia, the implementation area covered a population of about 6 million people living in 59 districts in the 4 regions of Oromia, Tigray, Amhara and Southern Nations Nationalities and Peoples. In 2015, the country had an estimated maternal mortality ratio of 353 per 100 000 population and a neonatal mortality rate of 28 per 1000 population.15 In Uttar Pradesh, implementation took place within a population of about 13 million people living in 6 districts. In 2016, the maternal mortality ratio in this state was estimated at 201 per 100 000 population and the neonatal mortality rate at 35 per 1000 population.16 The prioritization process for what was to be implemented in each study setting involved consultation with government and community leaders and was guided by a global recommendation concerning the basic package of care for all women and newborns, emphasizing a core set of interventions that could be delivered at the community or primary health level.17 Following detailed characterization of the individual implementation approaches,18 8 indicators for improvement were identified across the study areas: 4 for access to essential care and 4 for interventions or behaviours recommended for all families (Table 1). Indicators of maternal and newborn health targeted for change in all 3 jurisdictions* Community-based programming differed according to setting; however, all interventions were designed to improve survival, to reach the poor, to be feasible for large-scale implementation and to support government priorities (Box 1). For example, the interventions included an emergency transport scheme to remove access barriers in Gombe, improved use of local data to identify families in need of care in Ethiopia and the establishment of women’s groups in Uttar Pradesh. For further detail, see Appendix 1 (available at www.cmaj.ca/lookup/suppl/doi:10.1503/cmaj.190219/-/DC1). Gombe State, Nigeria* Ethiopia† Uttar Pradesh, India‡ *For further detail, see the Maternal and Neonatal Health Project — Gombe (http://www.sfhnigeria.org/the-maternal-and-neonatal-health-project-2012-2016/).†For further detail, see the Last Ten Kilometers Project (http://l10k.jsi.com/).‡For further detail, see the RGMVP (Rajiv Gandhi Mahila Vikas Pariyojana), a poverty reduction program of the Rajiv Gandhi Charitable Trust (http://www.rgmvp.org/). *For further detail, see the Maternal and Neonatal Health Project — Gombe (http://www.sfhnigeria.org/the-maternal-and-neonatal-health-project-2012-2016/). †For further detail, see the Last Ten Kilometers Project (http://l10k.jsi.com/). ‡For further detail, see the RGMVP (Rajiv Gandhi Mahila Vikas Pariyojana), a poverty reduction program of the Rajiv Gandhi Charitable Trust (http://www.rgmvp.org/). We performed cluster-based household surveys in 2012 and 2015, which involved interviewing women who reported a live birth in the preceding 12 months.19,20 The same methods and sampling frames were applied in both years and covered the entire area of implementation (Appendix 1). We applied multistage random sampling to generate a representative sample of women living in the implementation areas. In Gombe State, clusters were defined as enumeration areas. The enumeration areas were listed alphabetically, and their population size cumulated; areas were then systematically selected with probability proportional to population size. Households in the selected enumeration areas were listed and enumeration areas segmented into groups of about 75 households, with 1 segment in each enumeration area randomly selected for the survey. In Ethiopia, clusters were defined as villages. The 59 implementation districts (woreda) and their subdistricts (kebele) were listed geographically from north to south, and their population size cumulated; subdistricts were then systematically selected with probability proportional to population size. One village was randomly sampled for each selected subdistrict. Within each village, households were listed and villages segmented into groups of about 75 households, with 1 segment in each village randomly selected for the survey. In Uttar Pradesh, clusters were defined as villages. All villages from the 6 implementation districts were listed alphabetically, and their population size cumulated; villages were then systematically selected with probability proportional to population size. All households within selected villages were listed and villages segmented into groups of about 75 households, with 1 segment in each village randomly selected for the survey. The final sample size was sufficient to measure, with 90% power and a 5% level of significance, changes of a minimum of 20 percentage points across the range of indicators, representing the magnitude of change that was anticipated by project partners. In 2015, partly because of declining trends in fertility, 2 changes were made to increase the sample size. In Gombe, the number of clusters was doubled, and in all 3 jurisdictions, cluster size was increased from 50 to 75 households (Table 2). Household survey samples in 2012 and 2015 A modular household questionnaire was applied by trained interviewers. In 2012, there was relatively little guidance available on best practice for measurement of maternal and newborn health.21 We conducted extensive pretesting of questions and pilot testing of survey protocols, and we reviewed existing surveys, including the Demographic and Health Survey.22 In brief, household heads were asked about socioeconomic characteristics, and resident women aged 13–49 years were asked about their access to health care in the past year. Further questions were asked of women who reported a recent birth. As an example, the questionnaire implemented in Ethiopia in 2015 is provided in English in Appendix 2 (available at www.cmaj.ca/lookup/suppl/doi:10.1503/cmaj.190219/-/DC1). After translation and back-translation, surveys were implemented in Hausa in Gombe; in Amharic, Oromifa and Tigrinya in Ethiopia; and in Hindi in Uttar Pradesh. In Gombe and Uttar Pradesh, the data were collected using hand-held digital devices. In Ethiopia, the data were collected in 2012 using paper questionnaires, which were double-entered and reconciled; digital data collection was introduced in 2015. To enhance response rates, survey teams revisited each household up to 3 times for call-backs. Informed, written consent was obtained from all participants. We performed the analyses separately for each study setting. We adjusted coverage indicators for clustering of the segmented villages using the svy command in Stata 14 (StataCorp). We calculated odds ratios (ORs) for the difference in coverage over time using individual-level binomial regression models. For each survey, we used principal components analysis to construct an indicator of relative household socioeconomic status. We divided the resulting continuous index variable into quintiles of households from quintile 1 (poorest) to quintile 5 (least poor). The characteristics of poorest and least poor families in the 3 jurisdictions are illustrated in Appendix 3 (available at www.cmaj.ca/lookup/suppl/doi:10.1503/cmaj.190219/-/DC1). We examined the association between household socioeconomic status quintile and indicator coverage at each time point using binomial regression. We tested linearity of the association between socioeconomic status and indicators using a likelihood ratio test, comparing a model with socioeconomic status quintiles treated as a continuous variable with a model in which quintiles were treated as categories. Where there was no evidence of nonlinearity, we calculated the ORs and 95% confidence intervals (CIs) for a 1-unit change in socioeconomic status. Where there was evidence of nonlinearity, we calculated separate ORs for each socioeconomic status quintile. Where there was evidence of coverage change between survey years, to determine whether the association between household socioeconomic status quintile and indicator coverage changed over time, we included data from both time points in a binomial regression model and tested for an interaction between time point and socioeconomic status quintile. All regression models were at the individual-woman level and included robust standard errors to account for clustering of the data. We used interaction tests to examine whether change was inequitable for multiple indicators. A p value of less than 0.05 was used to indicate statistical significance. In Nigeria, national-level approval was obtained from the National Health Research Ethics Committee, Federal Ministry of Health, Abuja, and in Gombe State from the State Ministry of Health in both Gombe and Abuja. In Ethiopia, national-level support was obtained from the Ethiopian Ministry of Health and ethics approval from the Ministry of Science and Technology; at the regional level, approval was granted by the Regional Institutional Review Boards in Oromia, Tigray, Amhara, and Southern Nations Nationalities and Peoples. In Uttar Pradesh, India, approval was obtained from SPECT-ERB, an independent ethics review board, and written permission was obtained from the National Rural Health Mission of Uttar Pradesh. Ethics approval was also obtained from the London School of Hygiene & Tropical Medicine (reference 6088).
N/A
