Menu
Menu
Menu
Menu
Background: Maternal and perinatal mortality is a global development priority that continues to present major challenges in sub-Saharan Africa. Saving Mothers, Giving Life (SMGL) was a multipartner initiative implemented from 2012 to 2017 with the goal of improving maternal and perinatal health in high-mortality settings. The initiative accomplished this by reducing delays to timely and appropriate obstetric care through the introduction and support of community and facility evidence-based and district-wide health systems strengthening interventions. Methods: SMGL-designated pilot districts in Uganda and Zambia documented baseline and endline maternal and perinatal health outcomes using multiple approaches. These included health facility assessments, pregnancy outcome monitoring, enhanced maternal mortality detection in facilities, and district population-based identification and investigation of maternal deaths in communities. Results: Over the course of the 5-year SMGL initiative, population-based estimates documented a 44% reduction in the SMGL-supported district-wide maternal mortality ratio (MMR) in Uganda (from 452 to 255 maternal deaths per 100,000 live births) and a 41% reduction in Zambia (from 480 to 284 maternal deaths per 100,000 live births). The MMR in SMGL-supported health facilities declined by 44% in Uganda and by 38% in Zambia. The institutional delivery rate increased by 47% in Uganda (from 45.5% to 66.8% of district births) and by 44% in Zambia (from 62.6% to 90.2% of district births). The number of facilities providing emergency obstetric and newborn care (EmONC) rose from 10 to 26 in Uganda and from 7 to 13 in Zambia, and lower- and mid-level facilities increased the number of EmONC signal functions performed. Cesarean delivery rates increased by more than 70% in both countries, reaching 9% and 5% of all births in Uganda and Zambia districts, respectively. Maternal deaths in facilities due to obstetric hemorrhage declined by 42% in Uganda and 65% in Zambia. Overall, perinatal mortality rates declined, largely due to reductions in stillbirths in both countries; however, no statistically significant changes were found in predischarge neonatal death rates in predischarge either country. Conclusions: MMRs fell significantly in Uganda and Zambia following the introduction of the SMGL interventions, and SMGL’s comprehensive district systems-strengthening approach successfully improved coverage and quality of care for mothers and newborns. The lessons learned from the initiative can inform policy makers and program managers in other low- and middle-income settings where similar approaches could be used to rapidly reduce preventable maternal and newborn deaths.
Uganda and Zambia were selected for the SMGL initiative because of their high number of maternal deaths and elevated MMRs, average or below average use of maternal health services (especially around the time of delivery) compared with other countries in the region, government commitments to improving maternal and neonatal survival, and ability to leverage existing U.S. government platforms to promote maternal health and reduce HIV transmission. Before the SMGL initiative, Uganda and Zambia had estimated national MMRs of 438 and 398 maternal deaths per 100,000 live births, respectively.20,21 In Uganda alone, an estimated 4,700 maternal deaths and 35,000 neonatal deaths occurred every year. Four SMGL-supported districts were designated as ‘learning districts’ in both Uganda and Zambia. Just prior to implementation of SMGL in 2011, Uganda SMGL-supported districts had a combined population of 1.75 million, with approximately 330,776 women of reproductive age (WRA; women aged 15–49 years) and an estimated 78,000 live births annually (Table 2).22 Zambia SMGL-supported districts had a smaller combined population of 925,000, with approximately 194,000 WRA and 37,000 annual live births. Whereas the 4 learning districts in Uganda were contiguous and densely populated, the 4 learning districts in Zambia were geographically dispersed and comprised a much larger, but more sparsely populated, geographic area. At SMGL baseline, Uganda learning districts had more hospitals and high-level health centers (HC IVs) with surgical capacity per capita than the learning districts in Zambia.23,24 Hospitals in SMGL-supported districts in both countries are predominantly government-owned, with a few private, faith-based facilities. Both countries had a regional hospital (Fort Portal Regional Referral Hospital in Uganda and Mansa General Hospital in Zambia) that was part of the SMGL initiative, with catchment areas that extended to neighboring non-SMGL-supported districts. Health centers III in Uganda and all health centers in Zambia were mid-level facilities that provide basic maternity and newborn care and limited emergency obstetric care including some, but not all, of the 7 BEmONC signal functions. Assisted vaginal delivery, in particular, was often not performed in mid-level facilities due to concerns about possible adverse events. Health centers II in Uganda and health posts in Zambia are lower-level primary care facilities that provide antenatal, delivery, and postpartum care and refer complicated births to higher-level facilities. Baseline SMGL-Supported District Characteristics Abbreviations: CEmONC, comprehensive emergency obstetric and newborn care; EmONC, emergency obstetric and newborn care; SMGL, Saving Mothers, Giving Life. The SMGL initiative was implemented in phases: Phase 0 (pre-implementation planning in 2011–2012), Phase 1 (June 2012 to December 2013), and Phase 2 (January 2014 to October 2017). Phase 1 consisted of rapidly scaled-up facility and community interventions (“the big push”) to address the 3 delays.15 Phase 2 aimed to continue and consolidate successful interventions introduced in Phase 1; improve quality of care, including care for sick and small newborns; and further refine M&E methods and surveillance activities in the learning districts. Additional districts in both countries adopted the SMGL model, except for the M&E approaches.15 To evaluate SMGL’s impact, comparisons of maternal and perinatal outcomes in the learning districts were made between the 12-month baseline period (June 2011 to May 2012) prior to SMGL implementation and the endline period (January to December 2016). The baseline and endline evaluations used similar M&E approaches to measure progress and outcomes: health facility assessments (HFAs), facility pregnancy outcome monitoring with enhanced identification of maternal deaths, and, at the SMGL-supported district population level, community-based maternal death identification with Reproductive Age Mortality Studies (RAMOS) in Uganda and censuses in Zambia, which included verbal autopsies for suspected maternal deaths (Table 3). Each of the M&E data collection and analytic approaches is described in greater detail below. SMGL Indicator Baseline and Endline Data Sources in Uganda and Zambia SMGL-Supported Districts Abbreviations: HFA, health facility assessment; MDSR, maternal death surveillance and response; POMS, Pregnancy Outcome Monitoring System; RAPID, Rapid Ascertainment Process for Institutional Deaths; RAMOS, Reproductive Age Mortality Studies. Each country’s baseline and endline evaluations used similar M&E approaches to measure progress and outcomes. At SMGL baseline and endline, each country conducted health facility assessments (HFAs) in all facilities that provided childbirth care in the SMGL-supported districts, using a modified version of the standard EmONC HFA questionnaire originally developed by the Averting Maternal Death and Disability program at Columbia University.25 The HFAs gathered data on maternity care infrastructure, human resources, and adherence to safe motherhood protocols and practices, drugs, equipment, and supplies. The HFAs also characterized facility EmONC status—defined as performance of 7 BEmONC or 9 CEmONC signal functions in the 3 months prior to the HFAs—and assessed capacity and use of transport for emergency referrals. The number of health facilities performing deliveries varied in each country over the 5-year initiative. The HFA results presented here were compiled only from those facilities that maintained delivery capacity from baseline to endline (105 in Uganda and 110 in Zambia). The HFA baseline results were used to document baseline status and identify programmatic needs for SMGL. The results also informed the distribution of human and financial resources to strengthen infrastructure and other facility capacities, particularly during Phase 1 of the initiative.26 The results of the endline HFAs, conducted in November 2016, were used to assess changes in infrastructure and capacity at the end of the initiative and to guide planning for post-SMGL sustainability. Baseline and endline indicators of changes in health care facility infrastructure, availability of medications and supplies, and EmONC functions and labor management were calculated as the percentages of all facilities that reported positive responses on the HFA indicators, with the exception of indicators that are reported as complete enumerations. HFA results documented baseline status and program needs and informed the distribution of human and financial resources to strengthen infrastructure and other facility capacities. Individual and aggregated retrospective pregnancy outcome data, including identification of maternal deaths in facilities, were collected periodically by trained health facility staff and SMGL M&E personnel in both countries using enhanced data collection tools. In 2011, Uganda and Zambia had just started to use an electronic aggregated health service data platform (District Health Information System, version 2), which did not cover all health facilities. To address this, SMGL M&E teams developed standard abstraction forms and operation procedures for ongoing data collection of health service and outcome indicators. SMGL-supported facility monitoring led to improvements in tracking routine service delivery indicators as part of the newly established district data platform. In Uganda, SMGL-supported facilities that provided CEmONC implemented individual-level Pregnancy Outcome Monitoring Surveillance (POMS) data collection on maternal and newborn outcomes, including information on obstetric surgeries. As part of POMS, a package of standard tools was developed and used to obtain comprehensive maternal and reproductive health information: (1) electronic abstraction of all individual pregnancy outcomes found in labor and delivery registers, (2) abstraction forms to triangulate data on complications and obstetric surgeries from multiple sources, and (3) standard operation procedures to perform data abstraction and data entry. Because SMGL-supported facilities used ward-specific log books rather than centralized health records, POMS data from hospitals and HC IVs were triangulated with patient logs from various sources—such as labor and delivery, postpartum, female ward, surgical, admission/discharge registers, and hospital morgues—within each health facility.27 Trained SMGL M&E and clinical staff collected information on maternal characteristics, type of delivery, pregnancy outcomes, and up to 3 maternal complications at the time of each delivery. The most immediately life-threatening complication was used to analyze maternal morbidities and calculate case fatality rates (CFRs) from direct obstetric causes. Although data on early pregnancy outcomes—spontaneous and induced abortions and ectopic pregnancies—were also individually collected, they were not included in the calculation of the severe direct obstetric complications and CFRs unless they led to maternal demise. This approach was used to ensure that only severely complicated early pregnancy outcomes are examined and yield conservative estimates of met need for obstetric complications and CFRs. In Uganda, SMGL-supported facilities triangulated POMS data and patient logs within each facility. In lower-level delivery facilities in Uganda, aggregated outcome data from maternity registers were collected at baseline. By Phase 2 of SMGL, the individual-level POMS approach was expanded to all delivery facilities and individual delivery and pregnancy loss data were collected every 3 months using a Microsoft Access-based electronic data management system. Starting in 2013, the Ugandan Ministry of Health, in collaboration with the implementing partners, introduced an ongoing maternal death surveillance and response (MDSR) system in SMGL-supported health facilities and communities, with the goal of more accurately identifying and ascertaining maternal deaths. In Uganda, detection of facility maternal deaths was enhanced using the Rapid Ascertainment Process for Institutional Deaths (RAPID) methodology,28 in which all health facility records related to deaths among WRA were reviewed. RAPID data collection was conducted periodically in hospitals and HC IVs by Ugandan and U.S. Centers for Disease Control and Prevention obstetricians, and collected data were cross-checked with POMS data. While conducted separately, RAPID enhanced the capacity of facility-based MDSR to identify and review additional facility maternal deaths. In Zambia, aggregate facility maternal and perinatal outcome data were collected at baseline by SMGL M&E teams from each implementing partner in conjunction with the baseline HFA data collection. After SMGL interventions were introduced, monthly collection of aggregated facility outcomes data continued through the end of Phase 1 (December 2013). Facility data abstraction forms were used to compile aggregated data primarily from maternity registers. However, data abstraction forms and the completeness of case detection varied among implementing partners. In Phase 2, the periodicity of data abstraction changed from monthly to quarterly and a unified electronic data abstraction tool was implemented. In Zambia, facility data abstraction forms were used to compile aggregated data primarily from maternity registers. Starting in mid-2015, enhanced case detection and an audit of each maternal death became mandatory in all Zambian health facilities as part of newly implemented national maternal mortality surveillance. SMGL monitoring included audited maternal deaths in the estimation of endline facility maternal mortality. In both countries, and in accordance with the global MDSR guidance,29 baseline and endline measurements of maternal deaths in SMGL-supported facilities were derived by cross-checking multiple facility and community data sources (as further described) in order to capture a complete list of maternal deaths. Maternal deaths captured in these sources include those due to direct and indirect obstetric causes. Direct, indirect, and cause-specific MMRs in facilities were calculated as the number of cause-specific maternal deaths per 100,000 live births. WHO guidelines for using the 10th revision of the International Classification of Diseases to the classification of maternal mortality (ICD-MM) were applied to determine the underlying cause of death.30 However, this was complicated by the inclusion of the deaths of pregnant or postpartum women who lived outside SMGL-supported districts—and, thus, were not exposed to the SMGL interventions—in the count of maternal deaths in the SMGL-supported facilities. Facility-based pregnancy outcome data were used to estimate other standard indicators of monitoring emergency obstetric care, such as the cesarean delivery rate, met need for emergency obstetric care, the direct obstetric CFR, and the facility MMR.5,10 The cesarean delivery rate was defined as the proportion of deliveries by cesarean delivery of total district births. Met need for emergency obstetric care in all facilities was defined as the proportion of all women expected to have developed severe obstetric complications (estimated at 15%)10 who were treated in any health facility, and the met need for emergency obstetric care in EmONC facilities was represented by the proportion of expected severe obstetric complications that were treated in a fully functioning EmONC facility. The direct obstetric CFR was defined as the proportion of all women admitted to all facilities and to EmONC facilities with a given severe complication who died before discharge. The facility MMR was calculated as the number of maternal deaths per 100,000 live births in SMGL-supported facilities. Throughout the SMGL initiative, the number of stillbirths and predischarge neonatal deaths among babies weighing ≥1000 grams were monitored from information recorded in facility maternity registers. In contrast to the monitoring of maternal deaths for SMGL, identification of perinatal deaths was not enhanced through triangulation of multiple data sources, audits were less widespread, and underlying medical and nonmedical causes were not consistently available. Individual-level data on maternal and delivery characteristics, Apgar scores, and birthweight were available only in Uganda; similar data were collected as aggregate counts in Zambia. Although facility HFAs reported whether they had performed neonatal resuscitation, information about successful neonatal resuscitation was not available for either country. The facility perinatal mortality rate was calculated as the number of stillbirths and predischarge neonatal deaths among births delivered in facilities divided by the total number of births (live births and stillbirths) in SMGL-supported facilities. Similarly, the total facility stillbirth rate (SBR) was calculated as the total number of facility stillbirths per 1,000 facility births. In Uganda, where timing of fetal death was captured, it was possible to calculate the intrapartum SBR as the number of intrapartum stillbirths (those occurring after the onset of labor but before birth) divided by the total number of births per 1,000 births. Finally, the predischarge neonatal mortality rate (NMR) was calculated as the number of facility neonatal deaths divided by the total number of facility live births per 1,000 live births. In Uganda, retrospective RAMOS were conducted in SMGL-supported districts to capture community-level maternal deaths at baseline, end of Phase 1, and endline. At baseline, trained village health teams used community registers to identify and compile lists of WRA deaths in the prior 18-month period. Deaths were investigated using a 1-page screening tool to identify WRA who had been pregnant during the 2 months preceding death. Caretakers of women who died while pregnant or postpartum were interviewed using a standardized verbal autopsy protocol, which explores circumstances and potential causes of maternal death.29,31 At the end of Phase 2, the trained interviewers used an expanded RAMOS questionnaire that collected data on household composition, lifetime and recent pregnancy events among all WRA residing in the household, and all deaths in the household since January 2016. Households that reported WRA deaths were further asked to identify if deaths occurred during pregnancy, delivery, or postpartum using the baseline 1-page screening tool. Verbal autopsy teams conducted interviews with caregivers to women whose deaths were associated with pregnancy. At both baseline and endline, verbal autopsy data were analyzed independently by 2 physicians trained to assign underlying cause of death, with a third physician opinion sought when no consensus on cause of death could be reached. They then issued a consensus standard WHO death certificate for each verbal autopsy. Only maternal deaths that occurred during the baseline and endline periods were included in the analyses. In Uganda, retrospective RAMOS were conducted in SMGL-supported districts to capture community-level maternal deaths at baseline, end of Phase 1, and endline. In Zambia, community-level maternal mortality data were collected using household population censuses conducted in 2012 and 2017. The primary aim of the censuses was to assess the baseline mortality for WRA—including maternal mortality—and the change between the 2 time points. To enable calculation of maternal mortality rates and ratios, the household census data provided the number of WRA, the number of WRA deaths, and the number of live births in the population within the 12-month period before each census. For the 2012 census, the recent period used for WRA deaths and births was March 2011–February 2012; for the 2017 census, the recent period was July 2016–June 2017. A series of questions was asked about each person who was a usual member of the household and had died recently (since October 1, 2010, for the 2012 census and since January 1, 2016, for the 2017 census), including the age, sex, and dates of birth and of death. For each death of a woman aged 12 to 49 years, additional questions were asked about whether the woman had died when pregnant, during childbirth, or within 2 months after the end of a pregnancy. For each death of women aged 12 to 49 years, a verbal autopsy interview was conducted with a member of the household to record information about the circumstances, signs, and symptoms experienced by the deceased before she died. Teams of trained physicians reviewed the verbal autopsy interview responses and coded them to assign causes of death within both the baseline and endline censuses. To compensate for underreporting of deaths in reported numbers of all deaths of women aged 15 to 49 years and on maternal deaths from the baseline and endline censuses, standard adjustments to the data were made. We compared census-based measurements of population, births, and deaths in the 4 SMGL-supported districts with external sources and assessed that they were incomplete, particularly at baseline.20,21,32–34 We adjusted the 4 district mortality completeness using the General Growth Balance method.35 The adjustment factors were derived from fitting a line to a series of observed and predicted mortality rates for different age groups using the most recent national censuses and the United Nations Census Pregnancy-Related Mortality (CensusPRM) workbook for estimating maternal mortality from census data.36 The proportion of deaths among women of reproductive age that are due to maternal causes was estimated using the verbal autopsy data and applied to adjusted numbers of deaths to WRA to obtain the estimated number of maternal deaths. Likewise, the proportions of maternal deaths due to specific causes were applied to the estimated number of all maternal deaths in a reporting period to estimate the number of maternal deaths by cause. In order to establish a more comprehensive count of maternal deaths in facilities, community maternal death data in both countries were cross-checked with deaths reported through facility monitoring. A probabilistic match between information from verbal autopsies and from facility monitoring using place, cause, and month of death was completed. If a facility death was reported in a verbal autopsy but was not matched to a death recorded in the facility’s monthly monitoring statistics, the death was classified as an additional facility-based death and added to the facility count of maternal deaths. Population-based MMRs were computed using information collected through verbal autopsies in Uganda and Zambia at baseline and endline. Total and cause-specific MMRs were calculated after classifying causes of maternal death in accordance with ICD-MM.29 Zambia baseline verbal autopsy data were reclassified at the endline using ICD-MM, which was initially only used at endline. This resulted in an increase in the counts of Zambia maternal deaths identified in the baseline census and a corresponding increase in the baseline facility-based maternal mortality previously published.16 Direct, indirect, and cause-specific population MMRs were calculated as the number of cause-specific deaths in the SMGL-supported districts per 100,000 live births to WRA in these districts. The annual rate of SMGL MMR reduction (ARR) was calculated as: ARR=log(MMRendline/MMRbaseline)/5*100. This is consistent with WHO methodology to estimate MMR ARRs both globally and at the country level.1 Population-based MMRs were computed using information collected through verbal autopsies in Uganda and Zambia at baseline and endline. Response rates for verbal autopsies were very high in both countries. In Uganda, only 6 suspected maternal deaths identified in the baseline RAMOS and 2 deaths in the endline were not followed by an interview due to household dissolution or relocation. There were no refusals to participate in the baseline and endline RAMOS studies. In Zambia, several suspected maternal deaths were not followed by verbal autopsies (11 at baseline and 18 at endline). Refusals were encountered from 2 and 5 households, respectively. However, population maternal mortality data from Zambia are adjusted estimates based on the application of the General Growth Balance method to compensate for underreporting of WRA deaths and the estimated proportion of deaths among WRA that are due to maternal causes to derive maternal deaths. Calculation of population MMRs and selected EmONC indicators requires external population data. District-wide censuses in Zambia (2012 and 2017) and Uganda (2013 and 2017) were conducted by SMGL to enumerate households, population, and WRA. Enumerations were projected back to estimate the 2011 population using the inverse growth coefficient derived from the intercensal population growth rate provided by the countries’ national statistics bureaus.32,33 The baseline number of live births in Zambia districts was estimated by applying crude birth rates to the baseline district populations—directly derived from the 2010 national census. The endline live births were estimated by applying district-specific facility delivery rates calculated from the 2017 SMGL census to the endline district population. In Uganda, the number of live births was estimated by applying age-specific fertility rates among WRA enumerated in 2013 and 2017 in Uganda districts. For both countries, we calculated MMRs in facilities using the number of live births in facilities as the denominator and population-based MMRs using the estimated number of live births in the SMGL-supported districts. District-wide censuses in Zambia and Uganda were conducted by SMGL to enumerate households, population, and WRA to facilitate calculation of population MMRs. The results shown here were based on 4 district data analyses performed for each country. They were based on the total population and total number of health facilities in the SMGL-supported districts in each country. They were not a sample and are not representative of a larger population in the country. The pregnancy outcomes in facilities, including institutional mortality rates and ratios, were based on complete enumeration of deaths identified in facilities, so they were not subject to sampling error. However, the rates and ratios may be affected by random variation and changes in case detection.37 The following statistical tests were used when testing the difference between the Phase 0 and Phase 2 results. For the mortality rates and ratios, the error was modeled assuming deaths and births to be distributed according to a Poisson distribution. A z statistic, z=√SE(MMRbaseline)2 + SE(MMRendline),2 was used to calculate the P value of the difference between the baseline and endline MMRs, both in facilities and when comparing population MMRs.38 Similarly, changes in other core indicators, based on complete counts of events during the 2 periods, were also estimated using z statistics for significance testing. Finally, for the indicators that capture facility functionality, infrastructure, and availability of supplies, the McNemar’s test, which is appropriate for dichotomous responses for matched pairs of data collected at different time points, was used to test for significant differences.39 Results were considered significant if P<.05. The study protocol was approved by, and complied with, Uganda and Zambia Ministries of Health procedures for protecting human rights in research, and was deemed nonresearch by the U.S. Centers for Disease Control and Prevention Human Research Protection Office of the Center for Global Health. Written informed consent was obtained for respondents in all households and among women for the census and RAMOS interviews. For the verbal autopsies, written consent among the caregivers of the deceased subjects was obtained after informing the caregivers about the purpose and public health importance of the research, selection procedures, voluntary participation, and confidentiality. Interviews were scheduled no sooner than 6 weeks after the death occurred.
N/A
