The Harvard community has made this article openly available. Please share how this access benefits you. Your story matters

Developing an Implementation Research Program for Quality and Equity: Exploring the Context, Adaptation, and Measurement Challenges of Maternal and Child Health Implementation Research in Rural Nepal The Harvard community has made this article openly available. Please share how this access benefits you. Your story matters Citation Harsha, Alex Kathryn. 2016. Developing an Implementation Research Program for Quality and Equity: Exploring the Context, Adaptation, and Measurement Challenges of Maternal and Child Health Implementation Research in Rural Nepal. Doctoral dissertation, Harvard Medical School. Citable link http://nrs.harvard.edu/urn-3:hul.instrepos:27007741 Terms of Use This article was downloaded from Harvard University s DASH repository, and is made available under the terms and conditions applicable to Other Posted Material, as set forth at http:// nrs.harvard.edu/urn-3:hul.instrepos:dash.current.terms-ofuse#laa

Dedication With this thesis, I would like to celebrate the progress we have made towards building a better, more equitable healthcare system in Achham, Nepal. There is so much more to be done, and I hope these chapters help illuminate a path forward. To that end, I dedicate this work to Possible, our patients, and all those around the world who fight for social justice and healthcare for all. 2

Acknowledgments In real-world practice, problems do not present themselves as givens. They must be constructed from the materials of problematic situations which are puzzling, troubling, and uncertain In the varied topography of professional practice, there is a high, hard ground where practitioners can make effective use of research-based theory and technique, and there is a swampy lowland where situations are confusing messes incapable of technical solution. The difficulty is the problems of the high-ground, however great their technical interest, are often relatively unimportant to clients or to the larger society, while in the swamp are the problems of greatest human concern.there are those who choose the swampy lowlands. They deliberately involve themselves in messy but crucially important problems and, when asked to describe their methods of inquiry, they speak of experience, trial and error, intuition, and muddling through. - from Schön, Donald A. The Reflective Practitioner: How Professionals Think in Action. New York: Basic Books, 1983: pp40-43 The journey into the swamplands was harder and more rewarding than I imagined. For their support along the way, I would like to thank the entire team at Possible and in particular Drs. Duncan and Sheela Maru. Dr. David Citrin, Isha Nirola, Poshan Thapa, Anant Raut, Dr. Al Ozonoff and Dr. Ryan Schwarz. Your tenacity, wisdom, humility and humor carried us forward and helped me find greater understanding of both our purpose and our path. Duncan and Sheela, you pushed me to think and write with clarity, and to find my place as a physician-researcher. Thank you. Thanks as well to David, for your keen anthropologic eye and ethical voice both of which I hope you ll find echoed here throughout. Poshan and Isha, thank you both for your leadership, guidance, and willingness to do the hard work of partnership. Anant, thanks for your quiet support and your dedication to improving our data systems for our patients. Al, thank you for generously answering my (many) biostatistics questions. And Ryan, your insightful questions and incisive comments always help me hone my thinking. Thank you as well to my parents and family, who have always been there even when I m far away. Finally, I would like to thank my wonderful husband, Finnoh Bangura, without whom I might still be banging my head at my desk. Your unending support, thoughtful critiques and quick wit give me life. Thank you all. 3

TABLE OF CONTENTS Glossary... 5 Abbreviations... 6 Introduction... 7 Ethics statement... 9 Implementation Research: Framing the Epistemological Challenges... 9 Study setting and background... 12 Chapter 1 Maternal and Child Health Utilization in Rural Nepal: Institutional Birth in the Context of Expanding Surgical Obstetric Access... 15 Background... 15 Methods... 17 Results and Discussion... 19 Limitations... 21 Conclusions... 22 Chapter 2 Adaptation in Implementation Research: A Case Study of Group Antenatal Care... 24 Background... 24 Methods... 25 Results and discussion... 26 Inner setting: Organizational structure, culture code and social networks... 28 Intervention characteristics: Source and Evidence... 29 Outer setting: external policy & incentives... 32 Implementation Climate: Compatibility as a function of Adaptability and Relative Priority 33 Implementation Effectiveness: Research and the Learning Climate-Values Fit... 35 Reflective action to improve the learning climate... 38 Government re-engagement: balancing fidelity and fit... 39 Implementation and Effectiveness Outcomes: Preliminary Results... 40 Conclusions... 41 Chapter 3 - Measurement challenges in implementation research: The case of child mortality... 45 The Need and Challenge of Child Mortality Estimation... 45 Under-two mortality assessment in health systems research trials... 47 Our experience in rural Nepal... 49 Towards continuous mortality assessment systems for child survival studies... 50 Conclusions... 53 Summary and Suggestions for Future Work... 54 References... 56 Tables and figures... 70 Supplement A: Organizational Strategy, Care Delivery Design, and Impact Evaluation at Possible... 82 4

Glossary Term Basic Emergency Obstetric Care (BEmOC) Census Comprehensive Emergency Obstetric Care (CEmOC) Effectiveness Efficacy Implementation Implementation Climate Practitioner ethnography Under-five mortality Definition A set of six signal functions for management of pregnancy and delivery complications, including pharmacologic treatments and assisted delivery Refers here to exhaustive sampling methods rather than full population demographic enumeration A set of nine signal functions for management of pregnancy and delivery complications, including all basic functions cesarean section, blood transfusion, and sick newborn care How well a particular intervention works in practice (as compared to efficacy) How well a particular intervention works in controlled settings (as compared to effectiveness) The act of putting an intervention into practice Shared set of beliefs, expectations, and motivations regarding a specific innovation implementation within an organization Ethnographic method and writing style aiming to help practitioners to describe and understand the culture of their workplace The number of child deaths occurring before age five divided by the number of live births in a particular time period 5

Abbreviations ANC: Antenatal Care BH: Bayalpata Hospital (a public-private hospital in Achham, Nepal run by Possible) CHP: Community Health Program (at Possible) CHW: Community Health Worker IBR: Institutional Birth Rate (proportion of deliveries in prior 2 years occurring in a health facility) IMR: Infant Mortality Rate (number of child deaths occurring before 12 months of age over the number of live births in a particular time period) KI: Key informant interview NMR: Neonatal Mortality Rate (number of child deaths occurring before 28 days of age over the number of live births in a particular time period) U2MR: Under-Two Mortality Rate (number of child deaths occurring before 24 months of age over the number of live births in a particular time period) U5MR: Under-Five Mortality Rate (number of child deaths occurring before 5 years of age over the number of live births in a particular time period) 6

Introduction In this paper, I examine the emerging field of implementation research as a multidisciplinary field of study (1) and an urgent call to action to facilitate the adoption of evidence into practice and to develop responsive healthcare systems that address healthcare quality and equity (2-4) particularly to accelerate progress in maternal and child survival (5). Over the past eighteen months, Possible, a nonprofit that runs a rural hospital in a partnership with the government of Nepal, has been developing an implementation research program embedded within expanding direct healthcare delivery capacity. This is distinct from, say, using the hospital or community as a partner site for efficacy-oriented studies or from creating an implementation research program within an already well-formed organizational or healthcare delivery framework (2). Possible made this decision as part of a core strategy to build a learning, adaptive healthcare system in rural Nepal (6). This experience highlights some of the challenges of defining, creating strategy for, and executing on implementation research in resource-limited settings. Globally, maternal and child mortality are major indicators of quality and equity, and have suffered from a major implementation gap in resource-limited settings (7, 8). Thus, I ground my exploration of implementation research challenges in the context of maternal and child health in rural Nepal. The outline of the paper is as follows: to begin, I provide a brief overview of the implementation research field, highlighting some of the broad epistemological challenges faced by this growing multidisciplinary field. I also offer a description of the study setting and organizational background. In the first chapter, I review the evidence driving Possible s multilevel strategy for maternal and child healthcare improvement in rural Nepal and present an analysis of institutional birth factors in the context of Possible s expansion of emergency obstetric services in the study area. While not an implementation research study itself, the study findings provided the impetus for the organization s first hybrid effectiveness-implementation research study (type 1) (9) involving group antenatal care. 7

In the second chapter, I present an analysis of the challenges attendant to this nascent implementation research program and intervention that reflect deep tensions between fidelity and adaptation. I, along with my team members, reflect on these challenges through a practitioner ethnography of implementation (10) that draws on Klein and Sorra s foundational implementation effectiveness theory (11). Briefly, this theory holds that implementation effectiveness is determined by the climate of implementation (i.e. the system of expectations, support and rewards that encourage implementation) and the degree of fit between the intervention s attributes and the organization s values. I extend Klein and Sorra s theory to include research dynamics as an element of the intervention that may shape implementation climate and thus affect the perceived innovation-values fit of an intervention. In the final chapter, I address a significant limitation of implementation research in resource-limited settings: the development of reliable data systems for routine outcomes monitoring. I explore these challenges with a review of existing child mortality assessment techniques, providing an argument for under-two mortality as a reliable indicator for health systems trials at the sub-national level. I then outline our experiences developing continuous surveillance methods to collect these data, which suggest such an approach is feasible within an integrated healthcare systems approach to service delivery and quality improvement. Together, these chapters explore the development of an implementation research program in rural Nepal, particularly as it pertains to maternal and child health interventions to improve health equity and quality of care in a severely underserved environment. I attempt to document the evidence supporting our efforts to implement maternal and child health interventions and to delineate two of the major challenges of implementation research in this setting, namely, resolution of fidelity-adaptation concerns in iterative design and measurement of reliable and robust population health outcomes. Through this analysis, I offer our experiences as an example of the importance of continual reflection and further integration of 8

implementation research design and measurement strategies into in the messy world of healthcare systems strengthening in resource-limited settings. Ethics statement All data presented herein are part of either (1) the institutional birth study, or (2) the group antenatal care study, both approved by the Nepal Health Research Council (#98/2011 and #133/2014, respectively) and the Partners Human Research Committee (2012P000856/BWH and 2015P000058/BWH, respectively). The Institutional Research Committee of Kathmandu University School of Medical Sciences/Dhulikhel Hospital (IRC-KUSMS) also approved the group antenatal care protocol (81/14). In addition, all research activities are reviewed by Possible s Community Advisory Board on a quarterly basis. For the institutional birth study, all participants gave verbal consent and no patient identifiers were recorded. For the group antenatal care study, all patient participants gave written consent; their data is presented here as a preliminary analysis only. Due to the small organization and the nature of the reflexive ethnography, participant-authors gave verbal consent with the understanding that their identity would be protected as far as possible in the final manuscript. Implementation Research: Framing the Epistemological Challenges Implementation research is a growing field of study exploring how and why evidence-based practices and innovations are successfully carried out in real-world contexts, often described as filling a know-do gap (12). Rather than capitulate to Schon s depiction of these questions as a swampy lowlands where problems are not amenable to scientific inquiry, researchers, policymakers and practitioners attempt to leverage the scientific method to both improve on and learn from implementation efforts (13). Implementation research addresses the challenges of health services implementation by drawing from fields such as clinical research, individual and organizational psychology, management, public policy, and even manufacturing. Due to the complexity of the endeavor and the diverse epistemology, researchers have struggled to define the field consistently (14-20). 9

Some progress has been made in recent years towards defining implementation research s scope and methods. Remme et. al (2010) propose a taxonomy of three domains: first, operational research, which aims to answer locally relevant and immediately useful questions about how to improve interventions and their implementation in a specific context; second, implementation research, which addresses both locally and broadly applicable questions about the effectiveness of different implementation strategies for specific services; and third, health systems research, which is concerned with questions of broad relevance for healthcare quality improvement at the systems level (17). Peters et al (2013) organize the approaches along a similar spectrum considering the centrality of implementation to the research question, and the degree of generalizability afforded by the design (1). In particular, hybrid effectivenessimplementation trials can be broken down along this spectrum. Moving from lowto-high centrality of implementation, these are: type I trials, which focus on clinical effectiveness while secondarily collecting implementation measures (often through process evaluations) that may help guide adaptation or uptake in other settings; type II trials, which cross-compare clinical effectiveness and implementation strategies, often through factorial designs, to improve both for interventions with high face validity in the setting in question; and type III trials, which emphasizes the effect of implementation strategies on intervention outcomes in different settings (9). Another way to organize the scope of implementation research questions may thus be along a fidelity-adaptation continuum, along which the generalizable and the specific are in constant tension (21). Fidelity is defined as the degree to which an intervention is carried out as designers intended, and is often conceptualized primarily as adherence, despite definitions that include dose, delivery quality, patient responsiveness, and program differentiation (22). Importantly, adaptation does not always stand in direct contrast to fidelity. While sometimes framed as nonadherence (23), adaptation is also understood as improvement on a model (15, 24) 10

during pre-implementation phases, and flexibility in implementation strategies (25) after feasibility has been assessed. However, the two remain in tension, as core model components can be lost through adaptation. Framed along this spectrum, operational research questions are those that are concerned with how to optimize interventions for specific contexts a formula that likely requires a measure of context-specific adaptation and evaluation (26, 27). More generalizable questions (i.e. those in Remme s implementation research (17)) are likely to emphasize how to increase fidelity of interventions in multiple contexts, (28) while the most generalizable studies may control for fidelity or employ it as a measure of internal validity (29). This spectrum is a useful heuristic because the degree of fidelity sought and achieved by a study is critical to the interpretation of evidence (22, 25, 29-31). A number of studies have found that, in general, high implementation fidelity is an important factor in intervention effectiveness (22, 32) but in practice, adaptation is often necessary (24, 33). An over-emphasis on fidelity and internal validity may thus reduce implementation effectiveness and the relevance of both the intervention and research (34). The tension between fidelity and adaptation is greatest when interventions are under-specified, as implementers and researchers cannot make educated guesses about whether adaptation will compromise an intervention s impact (25, 31, 35). Regardless, pressure for adaptation may outweigh the evidence for core intervention components (36). Evaluation of the relative emphasis on internal and external validity is as critical in implementation research as in other fields; more so, perhaps, due to the multidisciplinarity and pragmatism required to implement complex health interventions (37). Balancing multiple perspectives and applying evidence with limited guideposts are significant challenges in any setting, but resource-limited settings are especially fraught due to lack of support and infrastructure for evidence-driven, iterative implementation (2). Certainly, it can and has been done in similar contexts (38), however we believe our experience developing an 11

implementation research program highlights some of the potential challenges navigating fidelity-adaptation decisions and measuring valid, relevant outcomes. Study setting and background The interventions and research presented here take place in collaboration with the non-profit organization Possible in the catchment area surrounding Bayalpata Hospital. The hospital operates as a public-private partnership between the Ministry of Health of Nepal and Possible s Nepal-based sister NGO, Nyaya Health Nepal. It is located in the district of Achham, a hilly region of far-western Nepal. Achham is served by one major road; the hospital is approximately 12 hours from the nearest tertiary care facility and domestic airport, and more than 30 hours from Kathmandu by road. Communities in Achham are dispersed with an estimated population density of 153 people per square kilometer (39) and the hospital s catchment area primarily serves the 14 nearest village clusters (approximately 30,000 people) but routinely sees patients from neighboring districts including those who have traveled several days for care. The setting encapsulates many of the challenges faced globally in geographically isolated, low-resource settings. Working with the government of Nepal, Possible designs and implements an integrated healthcare model focused on providing quality longitudinal care from frontline health workers to the hospital. Possible works within the district health system, which is comprised of several tiers of health facilities and administration. Bayalpata Hospital is designated as a public hospital under a performance-based financing agreement and reports required service delivery and public health information to the District Health Office and subsequently the Ministry of Health. The District Health Office oversees the functioning of the district hospital (located 2-3 hours from Bayalpata Hospital by road), supervises the local health facilities (known in Nepal as health posts or sub-health posts) and collaborates with Possible on community health activities. Local health facilities are run by mid-level practitioners who are both administrators and basic medical care providers. Maternal health and preventive child health services (i.e. vaccine campaigns) are 12

provided by auxiliary nurse midwives (ANMs). Six of the 14 local health facilities in Possible s catchment area are designated as basic emergency obstetric care facilities. In 2013, Possible expanded its maternal and neonatal health services from basic to comprehensive emergency obstetric care (40) and introduced other surgical, laboratory and x-ray services at the hospital (41). Shortly thereafter, Possible began re-factoring its Community Health Program (CHP) and focusing on implementation of a community health systems strengthening intervention. These efforts involved: development of group antenatal care program for improved maternal and neonatal health service delivery, hiring new leadership for the CHP, and greater professionalization (e.g. increased training, promotion to full-time status) of Possible s community health worker (CHW) team. Simultaneously, Possible began developing an embedded implementation research program. The process involved three major institutional changes. The first was the establishment of an in-house Impact Team to develop data systems, including an electronic medical record and mobile data collection tools, to monitor hospital and community-based programming and guide implementation research. Subsequently, Possible established of the Health Systems Design Group (42), a research arm of the organization that provides academic partnerships and supports principal investigators. Finally, the team convened a Community Advisory Board, which includes District Health officials and staff, to provide regular feedback and guidance (42). The CHP aims to improve healthcare quality and access through local health facility infrastructure improvements, delivery of group care at those facilities, and routine home visits for patient outreach and follow-up. The Impact Team and CHP work closes together to define and measure progress towards these goals. Maternal and child health are main priority areas, due to continued high mortality in the region. 13

Thus, two of the program s four key indicators, institutional birth rate 1 and antenatal care coverage 2, relate to maternal and neonatal health service; a third is under-two child mortality 3. I will address each of these priority areas in turn while drawing attention to the role of implementation research and its attendant challenges. Chapter 1 explores the determinants of institutional birth in the context of expanding comprehensive emergency obstetric care through a quasi-experimental study that does not address implementation concerns. Grounded in this evidence regarding demand for institutional birth, Chapter 2 describes the adaptation of a group antenatal care intervention aiming to improve antenatal care coverage and institutional birth rates, highlighting some potential pitfalls of implementation research in this setting. Chapter 3 discusses the call to measure child mortality and the significant lack of adequate methods for doing so in resource-limited settings. I present an argument for focusing on under-two mortality rather than the typical under-five mortality, and outline a proposal for and some initial experiences with a continuous surveillance approach to measurement. 1 Institutional birth rate is defined as the proportion of deliveries in the prior two years that occurred in a health facility. 2 Antenatal care coverage is defined as the percentage of women who had four or more antenatal care visits during their most recent pregnancy in the prior two years. 3 Under-two child mortality is defined as the number of deaths occurring before age two divided by the number of live births occurring in the past two years. 14

Chapter 1 Maternal and Child Health Utilization in Rural Nepal: Institutional Birth in the Context of Expanding Surgical Obstetric Access Background The greatest lifetime risk for a mother and her baby occurs during childbirth; eight hundred women die from preventable causes related to childbirth every day (43). More than 40% of the world's 535,900 annual maternal deaths are related to intrapartum complications, and these deaths are closely linked to the world's 1.02 million annual intrapartum neonatal deaths and 904,000 birth asphyxia-related neonatal deaths (44). Ninety-nine percent of these deaths occur in developing countries (43). Nationally, neonatal, infant, and under-five mortality rates are 33 (one in 27), 46 (one in 22) and 54 (one in 19) deaths per 1,000 live births, respectively (45). The Far-Western region is home to the highest under-five mortality rates in the country, at 82 (one in 12) deaths per 1,000 live births (45). Increasing institutional birth rates is a central strategy in reducing maternal and neonatal mortality, yet progress in doing so remains challenged by several factors globally. Nepal, one of South Asia s most impoverished countries, is a paradigmatic case of the challenges in achieving institutional birth and reducing maternal mortality. In 2011, maternal mortality in Nepal was estimated at 281 per 100,000, and only 35% of births took place in a health facility (46). The Three Delays Model offers a useful framework for understanding the barriers to achieving institutional birth: the first delay occurs with the decision to seek care, the second in arrival at a health facility, and the third in the provision of appropriate care (47). The first and second delays health-care decision-making and mobilization for care are often thought of as demand problems. Maternal age, parity, education and household wealth have all been positively associated with usage of services (48). These factors suggest that experience with labor, awareness of danger signs, autonomy, and financial support all increase demand for maternal health care. Government-supported financial incentives and outreach programs 15

have been in place in Nepal since 2009, and attributed to increased institutional birth (49). However, rural Nepal is a patriarchal society where a woman often must defer health care decisions to her husband or his family members, particularly the mother-in-law (50). Several studies have found that lack of support from this older generation is a key barrier to utilization of maternal health services (50-53), and may be more important than cost or lack of awareness (53). While financial, educational and participatory action-based community mobilization strategies have shown success in improving utilization in Nepal and elsewhere, there is limited evidence that increased demand alone improves maternal and neonatal outcomes (54, 55). The third delay, the provision of appropriate care, is a supply problem addressed by increasing availability of trained providers and well-equipped facilities. To significantly reduce maternal mortality, the WHO recommends at least four basic emergency obstetric care (BEmOC) facilities and one comprehensive emergency obstetric care (CEmOC) facility per 500,000 people (56). Table 1 compares available services at these facilities. While evidence for impact of intrapartum interventions and emergency obstetric care on maternal outcomes remains inadequate in low-resource settings (44, 57), it is estimated that universal CEmOC coverage would avert 519,000 intrapartum-related neonatal deaths per year (85% of deaths) (58). The three delays are clearly interrelated, as decisions about when and where to seek care are limited by available services. Others have found that increasing emergency obstetric services leads to increased institutional birth and that women consider service availability when deciding where to deliver (59-61). Importantly, a recent quasi-experimental study comparing demand- and supply-side interventions demonstrated that when financial incentive programs were followed by increased access to BEmOC, communities saw the greatest gains in institutional birth, particularly for the most impoverished groups (62). 16

Health systems strengthening requires attention to all three care delays in concert (8, 58, 63, 64). However, the dynamics of how maternal healthcare-seeking behaviors change as CEmOC access increases had not been studied. Possible s expansion to CEmOC in 2013 thus provided an opportunity to evaluate this evolution. Understanding these dynamics has important implications for quality improvement, health education, and outreach as obstetric care access increases. Methods Here, I present the data and analysis from a prospective observational study on the impact of Possible s CEmOC expansion on institutional birth in rural Nepal. While not an implementation research study per se, the questions addressed here lay the foundation for the maternal and neonatal health intervention and implementation research explored in the following chapter. For this pre- and post-intervention cohort study on the impact of CEmOC expansion on institutional birth in rural Nepal, we surveyed an exhaustive sample of all women less than six weeks postpartum during three-month periods before and after CEmOC expansion. The sample included two populations of postpartum women who: (1) delivered in the community (at home or a clinic) and (2) either delivered or received services for postpartum complications in the hospital. Women presenting to the hospital were identified, consented, and interviewed by nursemidwives, while women delivering in the community were identified by government community health workers (CHWs, known locally as Female Community Health Volunteers) and interviewed by the hospital s paid CHWs. All postpartum women within six weeks of birth and living in the catchment area were eligible for participation in the study. Women referred to another facility after arriving at Bayalpata Hospital were later excluded due to difficulty determining outcomes in this dispersed population. There were no other exclusion criteria. Informed consent was obtained with either a signature or thumbprint. Women received NRs 100 (approximately $1 USD) compensation for participation. Government CHWs received NRs 50 for each woman identified and nurse-midwives 17

received NRs 100 for each survey administered. Bayalpata CHWs received no additional compensation, as these activities are part of their employment contract. In addition to demographic questions, women were asked about their choice of delivery location, their beliefs around safe delivery practices, the factors important in their decision-making process, and their satisfaction with their choice. Factors were elicited through open-ended questioning and later coded into themes; dummy variables were then used to enumerate the frequency of each factor within and between groups. All other questions were multiple choice. All quantitative data were coded on paper and manually entered into an excel spreadsheet. Identifiers were not collected. The pre-intervention data were coded and analyzed separately for a previous study on drivers of institutional birth (40). After completion of the post-intervention surveys, the data were added to a pooled dataset and re-analyzed. Data were analyzed with JMP version 11 (SAS Institute Inc, Cary, NC 2013). Bivariate analysis of demographics across periods and factors associated with institutional birth, within and across periods, using Fisher s exact test for categorical variables (e.g. caste, parity, delivery location, priority factors), and the Wilcoxon rank-sum test as all continuous variables (e.g. age, income, ropani, distance, travel cost) had non-normal distributions. Significance was determined by an alpha-cutoff of 0.05. For logistic regression analysis, all significant variables in bivariate analysis (listed in Tables 2 and 3) were entered into the model and refined using manual backward elimination, again with an alpha cut-off of 0.05. Those variables that did not meet the cut-off are not reported in the final model. To qualitatively assess the impact of CEmOC expansion on institutional birth, a single open-ended question was posed to each participant: Tell me the story of your birth. Responses were transcribed in shorthand in Nepali and translated into English. Two investigators analyzed responses from the pre- and post-expansion data through immersion crystallization (65). The social contextual model, which illuminates pathways by which social and contextual factors lead to differing health outcomes or health behaviors, informed the analysis (66). Based on the model, 18

factors were categorized as modifying or mediating factors on individual, interpersonal, organizational, community, or societal levels. Modifying factors are those that affect the outcome independently of the intervention pathway. Mediating factors are on the pathway between the intervention and the outcome. Analysis was undertaken separately in 2012 and 2014 and the results compared to explore how birth stories changed with the expansion to CEmOC. Results and Discussion We surveyed 98 and 133 women pre- and post-expansion, respectively, including 21 women living outside the catchment area who delivered at Bayalpata Hospital. Bayalpata CHWs do not follow women outside the catchment area population and thus similar women who delivered at home were not a part of the sample. To reduce bias, these 21 out-of-catchment women were excluded from final analysis, resulting in 77 pre-expansion and 133 post-expansion respondents. In 2012, 21 women reported delivering in the hospital, 2 in a health post (lower level facility), and 54 at home. In 2014, 79 women reported delivering in the hospital, 23 in a health post (lower level facility), and 31 at home. According to hospital records, there were 29 and 85 eligible hospital births in during months of data collection in 2012 and 2014, respectively. While we do not have a goldstandard comparator for health post and home deliveries, this data suggest 79% of the pre-expansion and 93% of the post-expansion hospital delivery groups were surveyed with a total of 89% coverage. Monthly income, literacy, antenatal visit completion, and autonomy in delivery-care decisions were are all significantly higher for the post-expansion group, but the groups were similar in terms of age, distance to facility, median land ownership, parity, and caste as shown in Table 2. Institutional birth increased significantly after CEmOC implementation (30 to 77% overall) at both hospital (27 to 59%) and village clinic levels (3 to 17%, p<0.01 for all) as shown in Figure 1. We report the results of bivariate analysis of significant 19

factors in Table 3. The number of women who believed the hospital is the safest delivery location and who prioritized safety in decision-making increased postexpansion. Prioritization of distance decreased, while prioritization of cost increased post-expansion. Median travel cost to the facility marginally decreased (300 to 260 NRs, p<0.01) and slightly more women received the governmentsponsored incentive payment (100% vs. 91%, p=0.03). Finally, post-expansion, more women reported prior knowledge of service availability and prioritization of services. Logistic regression indicated exposure to CEmOC availability, belief that hospital is the safest delivery location, safety prioritization, and higher monthly income predict institutional birth (AUC=0.83). There was a significant interaction of safety prioritization and time, such that pre-expansion women who prioritized safety were seven times more likely to deliver in an institution than those who did not, as shown in Table 4. Satisfaction with birth experiences increased (87% to 99%, p<0.01). Women delivering in an institution post-expansion were more likely to be satisfied with their delivery care (OR 13, p=0.04). There was no significant difference in perceived adequacy of staff, supplies, or facilities in the pre- and post-expansion institutional birth groups. In comparing the pre- and post-expansion birth stories, there were several notable differences as demonstrated in Table 5. Mediating factors on an individual and interpersonal level, such as perceptions of safety, knowledge of services, and the awareness of a potential need for services were increasingly common postexpansion. Some explained motivation for hospital births in the context of prior or current birth complications, demonstrating a perceived risk of home birth that could be addressed in the hospital due to CEmOC. For example, one woman noted, I knew that this hospital provided a complete set of services, just like other hospitals. Women who gave birth in the village clinic often did so as a secondary option, but largely reported positive impressions of safety and quality at these basic facilities. 20

We identified referrals, birth planning, and preparedness as organizational and societal-level factors driving institutional birth. For example, one woman said, per the suggestion of the [CHW] I also decided to deliver with skilled healthcare personnel. More women in the post-expansion group mentioned detailed birth planning involving the hospital or village clinic. As one woman described, I completed four antenatal care visits. I had also arranged for money and clothes. When the labour pain started, we called the jeep [for the hospital]. Pre-expansion, very few women expressed similar birth plans. There were three referrals in the pre-expansion group from the hospital to higher-level facilities. All led to noninstitutional births because of difficulties and delays with transportation. After CEmOC implementation, there were no similar referrals. Five women had cesarean delivery and one woman received a blood transfusion. Modifying factors at the individual and interpersonal level, as shown in Figure 2, included family (particularly mother-in-law) and partner support, access to financial resources, means of transport to an institutional setting, and gendered work responsibilities. Many women relayed the importance of family and partner support; its absence highlighted the lack of autonomy many experience. In both years, women who failed to have institutional birth reported challenges finding assistance with travel, often due to inadequate family or community support and planning. Nonetheless, the government-provided financial incentive for travel was a societal level modifying factor that motivated women to have an institutional birth in both time periods. Limitations There are several significant limitations to this small study that encourage caution in generalizing conclusions about the effect of implementing CEmOC on institutional birth rates in rural areas. It is important to note that, given the broad acceptance that cesarean deliveries save lives, a randomized, controlled trial would be unethical. As such, questions must be answered via non-randomized approaches like the one taken here. Given that the study was an observational pre/post design, 21

the effect of prior trends is not accounted for in the analysis. For example, increased income may represent a secular development trend not fully captured by sociodemographics data. In addition, village clinic quality improvements were not assessed and remain possible confounders given the increase in births at those facilities. We attempted to address these issues through qualitative analysis of individual birth stories. While we aimed to reach all eligible postpartum women, the sampling technique introduces possible bias because women who delivered at home may have been more difficult to identify or reach by community health workers and we have no gold-standard comparator to validate our coverage of these women. The hospitalbased data does suggest we achieved 89% coverage of those cases, and thus are unlikely to underestimate our primary outcome of institutional birth rate if birth rates remained relatively stable between 2012 and 2014. The use of community health workers and nurse midwives as enumerators also increases bias in selfreported perceptions of and preferences for institutional birth. We assume that sampling and self-report biases would be equal across time periods and thus less likely to affect questions of change in institutional birth rates. Conclusions The substantial increase in institutional birth seen in our quantitative analysis can be partly understood by the mediating social and contextual factors identified in our qualitative analysis. An increased perception of hospital as a more safe and desirable place to give birth was notable in post-expansion birth stories. Local health facility deliveries were less desirable alternatives but were often more achievable due to travel constraints and generally viewed favorably by women who experienced them. This effect suggests broader normalization of institutional delivery and greater trust in the healthcare system as a whole. Normalization of institutional birth may explain why perception of hospital safety strongly predicts institutional birth in both time periods while prioritization of safety only did so prior to CEmOC expansion. That is to say, once institutional birth is normalized, 22

safety may still be important but its priority may become more implicit in social norms framing institutional birth as the standard. While demand-generating activities have proven critical to increased institutional birth rates (55), we demonstrate that CEmOC expansion can drive significant demand for institutional births in an impoverished community with previously low access. After CEmOC expansion, women appeared to perceive more benefits of institutional birth and incorporate it into a normative framework that encouraged planning for the extra costs and contingencies required to achieve it. This effect also cascaded down to BEmOC local health facilities, with women exhibiting greater trust in the healthcare system overall. We believe that the demand-generating capacity of CEmOC services should thus be taken into account when considering allocation of resources for maternal and neonatal health. These findings support greater expansion of CEmOC services in rural/underserved areas even when institutional birth rates are low, as the services are likely to increase both utilization and safety. By increasing demand for institutional births while also making those births safer, surgical obstetric expansion may thus have a greater impact on childbirth-related mortality than demand-generating or BEmOC expansion approaches alone. 23

Chapter 2 Adaptation in Implementation Research: A Case Study of Group Antenatal Care Background The study findings presented in Chapter 1 reveal great improvements in institutional birth rates after surgical obstetric service expansion but also illuminate a remaining need to support demand and access. Increasing social support for institutional birth, encouraging birth preparedness, advancing gender equality through women s empowerment activities, and further developing transportation resources and should be key targets of future interventions to encourage institutional birth and decrease maternal mortality in this setting. To address the social barriers to care identified in the study, Possible began developing a group antenatal care aims to improve social support and birth planning to improve institutional birth and ultimately improve maternal and neonatal survival. However, the institutional birth study described in the previous chapter notably does not address implementation issues affecting emergency obstetric care delivery, factors that may affect both the internal validity and external validity of the results. Ideally, conclusions drawn about the impact of CEmOC on demand would reflect how implementation variables (e.g. quality or coverage) modify the response. Similarly, the evidence for the impact of CEmOC would be made more meaningful through an exploration of how it was implemented in this context. To address these and similar limitations of traditional research studies, Possible began simultaneously developing an implementation research program. The embedded implementation research program aims to address pragmatic challenges of healthcare service delivery in settings of under-utilization, and group antenatal care was the team s first implementation-effectiveness study. Here, I present an ethnographic case study of this developing implementation research program through the lens of the group antenatal care study, and focusing on the challenges of iterative design in the context of fidelity and adaptation concerns. 24

The following ethnographic case study builds on prior research by explicitly and reflexively exploring how the implementation climate affected intervention adaptation in both intended and unintended ways (34). Applying methods including participant observation, documentation review, and key informant interviews, we aim to: (1) Provide the contextual basis for analysis of the prospective Group ANC study data, which includes quantitative outcomes and qualitative analysis of program acceptability, feasibility and impact. (2) Explore the dynamics of implementation research in our context and their impact on intervention adaptation and implementation. Methods For this case study of implementation research in action, we choose an ethnographic approach for its emphasis on the social milieu of an intervention (67) the individual, organizational and sociocultural perceptions, beliefs and motivations that contributed to the adaptation of this intervention. Ethnography is increasingly used in implementation research to understand the process of implementation through multiple, often competing, perspectives in a unique organizational, social and historical context (27, 68, 69). Direct, long-term observation of and interaction with implementation efforts yields deep contextual data that may illuminate dynamics critical to implementation outcomes that would otherwise remain hidden (70). We deviate somewhat from traditional ethnographic methods by intentionally blurring the line between the researcher and the researched (71). The first author was embedded as a participant observer who contributed substantively to the development of the intervention and its evaluation while observing program and organizational processes. All other authors were also directly responsible for program implementation, and contributed to the analysis presented here both as subjects of formal key informant interviews and as critical, reflexive observers. While this approach likely reduces the objectivity of the data presented, we believe, 25

consistent with some arguments (10, 72, 73), that this practitioner ethnography approach is necessary to understand the dynamics at hand. We rely on several sources of data including field notes, internal documentation, meeting notes and other communications over the organization s online project management tool (74), and key informant interviews with program leadership. We undertook qualitative analysis through the recursive process of immersion crystallization (65). We coded field notes and internal communications data for themes and created a set of open-ended key informant questions based on these themes to further explore the dynamics of program implementation and research development from the perspectives of those directly involved. All data was then reanalyzed and, where possible, mapped to the Consolidated Framework for Implementation Research (CFIR) (75). The CFIR has been successfully used: (1) as potential factors for inductive analysis (76) or improved implementation design when paired with a theory of change as part of participatory action research (33, 77); (2) as an analytic framework for post-hoc deductive analysis (78); and (3) as an organizational tool to facilitate an explanation of post-hoc analyses (79). We use the third approach here, to place our findings in the larger implementation research discourse and identify potentially generalizable issues affecting implementation outcomes. (See Table 6 for CFIR constructs and relevant definitions.) Results and discussion Broadly, and perhaps as expected, tension between fidelity and adaptation was a central feature of the implementation process for this nascent intervention. The pressure for adaptation was exerted primarily by the outer setting s external policy and incentives for maternal and child health. Given incompatibility with government antenatal care scheduling protocol, fidelity-adaptation tensions were high. Broadly, the two sides fell along research versus implementation lines. For implementers, the priority on fidelity was perceived as inappropriate, a perception driven by the intervention s source (research) and relative priority (over other CHP programs and over government partnership). For researchers, the priority on fidelity was 26

important, a perception driven by a belief in the evidence base and a lack of supporting evidence for government s scheduling system. The implementation climate was affected by this debate because priority on fidelity was reinforced through implicit and explicit behaviors that created a learning climate in which some implementers did not feel valued or safe to challenge the situation. This learning climate was in direct conflict with organizational values regarding role of research in design, which is driven by attentiveness to historical trends in extractive research as well as a priority on service. This climate and the pre-adaptation intervention also conflicted with an over-arching narrative that we are a mutually supportive and pragmatic team that works within a government system to produce improved patient outcomes. These dynamics parallel Klein and Sorra s (1996) influential model of implementation effectiveness in which the implementation climate and the innovation-values fit within a particular organization determines its ability to implement an intervention (11). Klein and Sorra (1996) define the innovationvalues fit as the extent to which targeted users perceive that use of the innovation will foster (or, conversely, inhibit) the fulfillment of their values (11). This definition is encompassed by the CFIR s compatibility construct, defined as the degree of tangible fit between meaning and values attached to the intervention by involved individuals, how those align with individuals own norms, values, and perceived risks and needs, and how the intervention fits with existing workflows and systems (75). For simplicity, we will use compatibility to represent both concepts. Similarly, the CFIR s definition of implementation climate as the absorptive capacity for change, shared receptivity of involved individuals to an intervention and the extent to which use of that intervention will be rewarded, supported, and expected within their organization (75) draws heavily on Klein and Sorra s model (80). A sub-construct of implementation climate, the learning climate represents the constellation of attitudes, beliefs, behaviors and organizational structures that create safety for innovation, failure and mutual learning (75). Specifically, a 27

supportive learning climate requires (a) humble leadership, (b) respect of team members as knowledgeable partners, and (c) adequate opportunity for reflection and evaluation of the change process (75, 80). We extend Klein and Sorra s model here to reflexively explore how implementation research interacts with the learning climate to shape perceptions of intervention adaptability and, ultimately, compatibility. In other words, organizations have values about what needs to be done and for whom, but also about how those interventions should be implemented and evaluated; these values must also be considered in the implementation research discourse. In the sections that follow, we will evaluate Possible s organizational culture and values, the intervention characteristics, the implementation climate for Group ANC, and the compatibility conflicts that arose out of an interaction between these and the implicit power dynamics of the organization s implementation research program. Inner setting: Organizational structure, culture code and social networks Possible was established as Nyaya Health in 2006 and started delivering care at Bayalpata in 2009. All healthcare delivery and government contracts and Nepali staff are managed, legally speaking, by the Nepali non-profit organization. A USbased non-profit provides funding and staff support. The two organizations operate with a single org chart. In 2013, Possible undertook a re-structuring exercise that led to greater professionalization of the team and the development of an explicit organizational culture code that guides strategy, communications and implementation. The culture code (Figure 3) comprises ten key identity statements that seek to align a cross-cultural, dispersed team on core values of service, innovation, efficiency, transparency and a strong work ethic. Possible promotes a supportive learning culture as part of its code to embrace challenge, be transparent and encourage mutual growth through professional intensity and personal support (81). These beliefs are enacted by supporting open communications at every level, including broad use of an online project management platform (74) and other transparent systems (82), formalized 28

upward feedback systems, systems-level morbidity and mortality conferences (83) and efforts to publish transparent accounts of organizational failures or challenges (84). Possible s commitment to implementation research is a formalized part of this learning culture that supports [the] fundamental service obligations to the communities in which [it] works (85) and helps the organization serve as a model for adaptive healthcare systems in Nepal and globally. As such, rigorous evaluation of interventions is considered the grease that facilitates quality improvement and scaling of health systems interventions (86), creating a virtuous cycle of health service delivery (see Figure 4). Possible thus endorses the idea that embedded implementation research programs are necessary to improve quality and equity in resource-limited settings (87). It is important to note that much of the strategy, intellectual formulation, and piloting of the initial Group ANC intervention preceded the development of the implementation research program presented here. The team as a whole was relatively inexperienced with respect to implementation in this specific context and implementation research in general. In addition, the CHP experienced significant leadership turnover during the initial intervention design phase. Finally, the simultaneous development of a population health surveillance system (see Chapter 3) was another implementation research activity that relied on both the Impact and CHP teams, and which created additional stress. Intervention characteristics: Source and Evidence The intervention was designed by the study PI in collaboration with several hospital midwives (non-community Healthcare staff) and Community Healthcare staff after a 2013 study on the determinants of institutional birth in the hospital s catchment area (40). The study found that while institutional birth has become normalized, lack of social support, birth planning and resources remain key barriers for women (40). Given the wealth of information on the success of CenteringPregnancy in highresource settings (88, 89), and participatory action women s groups in low-resource 29

settings (90, 91), we hypothesized that a group care model would address those barriers. CenteringPregnancy is a model of facilitated prenatal care groups with stable composition that include group health assessments and self-care activities, education, support and socialization, and ongoing evaluation of outcomes. (See Figure 5 for core components.) The model appears to improve maternal and neonatal outcomes including improved maternal satisfaction, fewer preterm births, and increased birth weight, particularly when implemented with skilled facilitation (92, 93). The initial design of the group care model tested in this study (Group ANC) aimed to maintain high fidelity to all 13 components, but reduced the visits from 10 to 6 to more closely align with the schedule for government antenatal care incentives (94). In an effort to maintain fidelity, the intervention included two extra visits in addition to the government-required four (one during the 7 th month and one between 1-2 months postnatal). As such, the content was also condensed for each of the six visits and adapted to the government guidelines for antenatal counseling and testing (94). The intervention was also designed to extend the CenteringPregnancy model by including a participatory action process for addressing barriers to maternal healthcare access, particularly poverty and lack of resources (95). The participatory learning and action group model takes participants through a shared process of identifying problems, planning solutions, taking action, observing results and reflecting on the impact and the need for further action (90). Women s groups employing this model in low-income countries including Nepal have demonstrated impact on maternal and neonatal health-seeking behaviors and outcomes through increased confidence and strengthened social support (91, 96, 97). Due to this growing body of evidence, the World Health Organization now recommends such groups (98). Together with two senior nurse-midwives, the research team piloted intervention content with women recruited from nearby villages and/or who presented to the hospital for routine antenatal care. The PI and a Nepali collaborator, both 30

physicians, assisted the nurse-midwives and gathered feedback from them regarding the feasibility and acceptability of the content and group structure. Due to the high turnover during this time, most team members were not directly involved in the development of the intervention and felt the intervention s source limited their ownership of it: [Group ANC] something that we didn't quite have a say in in terms of whether or not it was something we were going to implement or we thought it was a good idea or we were very familiar with the evidence base. It was kind of 'read some of these articles, this is what we're doing' and it was kind of 'jump right in'. (Impact Director, KI, December 7, 2015) Initially, this perception did not create significant tension. The team, including newly hired community health leadership, did in fact jump right in, perceiving that the patient-centeredness, empowerment, and efficiency that group care appears to provide were compatible with the organization s values and the needs of the community (Impact Director, KI). Local government healthcare providers and officials were convened to discuss the overall program design and give consent for working within their clinics and twelve nurse-midwives and CHWs received a twoday training on the model and content. Together with these government partners, we led demonstration groups in each of the six village clusters. This pilot served as an assessment of interest and reach, as all pregnant women in the village clusters were invited to attend. Interest was gauged as very high, with more than 100% of expected pregnant women (average of 43) attending in each village cluster. However, as groups are intended to involve only 5-12 women at a time, it was difficult to assess the receptivity of women to the ideal group environment or gather feedback about content and structure. To refine the intervention, we planned a three-month pilot and run-in period before the initiation of study enrollment in November 2014. The CHP and research staff worked closely together to launch this phase. The Impact and research staff focused on creation of monitoring tools and the CHP staff engaged the government partners, securing an agreement with the District Health Officer that stated the group visits 31

would be eligible for the government incentive program. Throughout, there was significant overlap of roles given the small team and the open, horizontal organizational structure. All team members attended groups to observe and provided input on logistics, documentation, content and facilitation. Outer setting: external policy & incentives Due to the organization s deep commitment to strengthening the public healthcare system and the dependency of group ANC on the participation of government health providers, the most critical implementation challenges centered on conflicts with the government s external policy and incentives. Several major issues were highlighted, many similar to previously cited CenteringPregnancy implementation barriers (36, 93, 99), including poor engagement with local clinic nurse-midwives who were often attending to other duties during group sessions, the need for more training and support for facilitators to increase the quality of group discussions and improve participatory action, and poor visit documentation. While nurse-midwives were happy with the group ANC sessions because they feel like the group ANC sessions were helping them deliver care to pregnant mothers, (Research Director, KI, December 14, 2015) they also felt the program negatively affected their ability to fulfill certain government obligations such as administrative reporting and achievement of full antenatal care coverage. The national safe motherhood program provides a financial incentive to women who complete four visits, but only if those visits are completed during the fourth, sixth, eighth and ninth month of gestation (94). The government tracks coverage of antenatal care by the percentage of expected pregnant women that fulfill these requirements, and rewards the clinics with the highest rates. Thus when women fell outside the specified gestational age windows during group visits, both the women and providers would lose incentives. Problematically, these windows are often defined and calculated inconsistently, making it difficult to predict when a woman would be considered eligible for the incentive. To achieve fidelity to the CenteringPregnancy model the groups were 32

composed of women who would deliver in the same month; to maximize incentive eligibility the visits were scheduled on different days each month. The nursemidwives appreciated the gestational age focus but felt the scheduling was too chaotic and preferred fixed dates (Focus group, December 23, 2015). Unfortunately, government officials were not receptive to more flexible eligibility policies that would accommodate both stable gestational age group scheduling and fixed monthly visits. This problem thus illuminated how deeply fidelity-adaptation tensions could be felt, and how complex these tensions can be when working under an implementation research paradigm and within a public healthcare system. Implementation Climate: Compatibility as a function of Adaptability and Relative Priority After the first three-month pilot, the challenges of implementation were significant enough to delay study enrollment while we began considering re-design options. The focus of fidelity-adaptation tensions centered on the scheduling issues; with the other issues identified (variable nurse-midwife engagement, facilitation, and documentation) there was a general consensus that improving these with fidelity to the model would lead to better outcomes, or at least a better understanding of whether group ANC could be successful in this context. Secondary analysis of fidelity data for Centering Pregnancy RCTs suggests that process fidelity to the core components listed in Figure 5 has a greater impact on maternal and neonatal health outcomes than content fidelity of the facilitated discussion material (93). However, due to the significant restructuring of clinic space and time required to hold groups, adaptations that compromise process fidelity, such as reducing staff facilitators or broadening gestational age ranges in groups to increase participation, may be necessary (36). With no published studies of antenatal care groups that do not adhere to the CenteringPregnancy model, and one small CenteringPregnancy pilot in Africa (100), there is little evidence to guide adaptation particularly in low-resource environments. 33

As the Director of Research explains, we didn't have a good idea of what the ideal group sessions look like or what are the components (KI, December 14, 2015). This lack of clarity heightened the fidelity-adaptation tensions around scheduling because it was unknown whether the stable gestational age group was a component of the model that could be adapted without compromising fidelity to the overall concept. The lack of scientifically validated evidence to evaluate the adaptation options or to support the rigidity of the government system frustrated the research staff greatly, and suggestions about adopting a monthly drop-in mixed gestational age group model to accommodate it were met with resistance. In addition, due to poor documentation of visits, there was little objective evidence to assess whether the problem affected a significant number of women. For researchers, fidelity to the gestational age group model remained paramount in the face of this uncertainty. Fidelity was less of a concern for the implementation team, particularly with respect to the gestational age group model. As the program director gained experience in the organization and began framing a broader strategy for the community health program, the scheduling challenges and the time spent dealing with them weighed heavily. The intervention s relative priority became more problematic, and concerns regarding the role of research in the implementation began to surface. For example, here the CHP Director explains, I am struggling with this a bit. I feel a lot of pressure to put my workplan second so we can push ahead with the research/group ANC needs. I am 100% a team player but I am confused on how much I am supposed to compromise our own work plan for the research studies Perhaps this is a misguided perception on my part. (To Impact Director, September 18, 2014) In addition, the community health program staff began getting increasingly negative feedback from the government partners regarding the fact that many women were not meeting government requirements through the group visit program. 34

Implementation Effectiveness: Research and the Learning Climate-Values Fit One driver of these compatibility tensions was the perception that the researchers had authority over intervention decisions, and limited implementers ability to adapt the intervention. This perception created a learning climate in which team members did not feel valued equally as partners in the iterative design process, which both inhibited effective partnership and conflicted with the organization s value on pragmatic, useful and non-extractive research. As the Impact Director explains, the sense that 'this is for research, this is for career-building, this is for publications' was an initial barrier to overcome in terms of working together as a team on the design. It masked the sense that we solve for the patients Or that we work with government for partnership building. (KI, December 12, 2015) The organization s culture of mutual support and horizontal, open-access structure did not eliminate this dynamic, as cultural deference towards doctors, particularly Western doctors from prestigious institutions, is pervasive in both Nepali and Western attitudes (101). Despite reassurance from leadership that program implementers were responsible for adapting the intervention as needed, these perceptions persisted. For example, here the CHP Director responds to a strategy proposed by one of the team members to improve the participatory action component of Group ANC by requesting input from the PI about the intervention: At this time we are not in a position to organize the mothers groups Is this compulsory? I am concerned about the type of oversight this intervention will need. (CHP Director to PI, September 8, 2014) The Director s use of the term compulsory suggests a belief that the research team was primarily responsible for making program design decisions. Though the PI worked to dispel this belief, with assurances such as you have the best sense of what the CHWLs can take on at this time, (September 14, 2014) suggestions and recommendations from the research team continued to be perceived as directives (CHP Director, KI, December 21, 2015). These dynamics compounded the 35

implementation challenges by simultaneously limiting the range of perceived adaptation options available and limiting the perceived power of the implementation team to make any changes. While implicit, these power dynamics were enforced during re-design discussions when the organization s leadership mediated the debate around stable gestational age group fidelity. For the research team, the case for fidelity came primarily from a belief in the evidence base rather than a priority on internal validity, and the lack of evidence to guide alternative (i.e. non-centeringpregnancy) designs. The leadership acknowledged the practical challenges and complicated government partnership dynamics of the scheduling system but ultimately promoted fidelity to the original model while encouraging data collection to monitor the impact of the scheduling scheme on participants government incentive eligibility. Thus, the experience of the CHP staff s struggle to coordinate the schedule and their tense interactions with government partners was unintentionally discounted, leading to a perception that the organization prioritized research over programs. The team planned a re-launch (Research Assistant to PI, December 23, 2014) to improve implementation strategies through new facilitator manuals that included a structured birth planning curriculum and worksheet, and simplified documentation for supervision and patient care. In recognition of the scheduling challenges, the team also blocked certain dates to minimize conflict with nurse-midwives other scheduled duties. Unfortunately, the strategy proposed did not address the fundamental conflict with the government s protocols. In the following exchange, the CHP Director and PI demonstrate two opposing points of view: At this point, given all the barriers we've faced, should we consider using a mixed group guide for all interventions VDCs? I think this will minimize most of the challenges we've been facing - and may be the most scale-able model at this point."(chp Director to PI, January 31, 2015) "As others have mentioned, if this model works and we can show that it works better than the traditional model, we will have leverage for further negotiations 36

with the government. I don't feel that this is the right juncture to pull-back/stop with our efforts to implement this model."(pi to CHP Director, February 1, 2015) This approach created greater conflict with the District Health Office and nursemidwives under their authority, and reduced their engagement in the process. As the Research Director writes of a planning meeting, they were less interested to know what we were doing; rather were more interested to suggest us to follow the government protocol"(march 4, 2015). Thus, the intervention began to feel unsustainable and the validity and feasibility of the entire model came into question (Research Director, KI, December 14 2015). Was this an appropriate, outcomesoriented and scale-able choice for the public healthcare system (CHP Director, KI, December 21, 2015)? These compatibility questions raised by the CHP Director reflected conflict with the organization s mission to build efficient, simple systems. Team collaboration started to break down. We were failing as an organization because people's feelings were hurt and we weren't delivering the right care that we wanted and we weren't evaluating it very carefully... Even though on the one hand we celebrated that as the nature of implementation research, I don't think we acknowledged [how] that was going to feel as an organization in terms of org culture. (Impact Director, KI, December 7, 2015) After tensions built significantly within the organization and with government partners, the CHP Director made an executive decision to delay any further action on the intervention in order to focus on other programmatic priorities. While the organizational leadership respected this decision, it did not resolve the fundamental authority issues because some team members did not trust that the Director was allowed to make that decision (CHP Director, KI, December 21, 2015). At an impasse and lost in internal conflict (Research Director, KI, December 14, 2015) we struggled to negotiate appropriate adaptations in light of both the internal setting s stretched capacity to manage the complex intervention and the outer setting s logistical challenges and political will to accommodate it. The way forward 37

required a repairing of the relationship between research and implementation team members and a re-engagement with government partners around iterative design. Reflective action to improve the learning climate Recognizing that traditional views of academic research and cultural deference to physician-researchers (101) were pervasive forces in design and iteration discussions challenged our self-conception as a mutually supportive and transparent organization with individuals dedicated to social justice and action, with research as one component of that effort. Perhaps because this position was core to the organization s values, the researchers struggled to understand the negative impressions of their efforts. This misalignment was worsened by a lack of openness about the degree of dissatisfaction among implementers (Research Director, KI, December 14, 2015). As the CHP Director explains: it wasn't until much later that [other staff] came up to me and were like 'you know what, I'm sorry I didn't speak up, it was wrong we were pulling allnighters trying to figure this out like all the time, we weren't working on any other care delivery programs, and, uh we were coming to you and complaining that this is not sustainable. (KI, December 21, 2015) Once realized, however, we attempted to face the issue head on. The team leadership collaboratively revised an implementation research strategy document to codify the language and process of implementation research and iterative design. Reiterating conversations that had taken place over many months, they sought to ensure that program implementers felt validated and empowered to make adaptation decisions. This process included adding something that says care delivery teams programs leadership always drives program and science always follows. Really clear-cut signpost language that we could all say we agreed upon It was really this clarifying but also group therapeutic process. (Impact Director, KI, December 7, 2015) In addition, the organization formally adjusted its reporting structure to place PIs under the supervision of the relevant program directors. In this case, that meant the 38

PI began having regular one-on-one meetings with the CHP Director, who helped to ensure the research efforts were aligned with broader programmatic priorities. This structural adaptation improved the implementation process by formally clarifying the responsibilities for decision-making regarding program adaptation. The organization s perspective had not changed; research had long been framed as the grease and not the engine of health systems improvement (6, 86). However, the reflective process that accompanied the drafting of the document and adapting the organizational reporting structure allowed the perceived conflict between the implementation research dynamics and the organization s values to be broken down in more meaningful ways. (See Supplement A for the document in its entirety.) Government re-engagement: balancing fidelity and fit With much of the internal tensions resolved and greater alignment around the adaptability of the intervention, we were able to focus on the external challenges of partnership with the government. Given both the implementation team and government partners strong preference for a simplified scheduling approach that aligned with government protocol, the implementation team proposed another iteration of the model with a fixed monthly ANC Day that women are invited to attend according to their gestational age. This proposal was well received by the government and they made a number of suggestions regarding the logistics. As the CHP Program Manager at the time explains, after we involved them heavily in the design process itself, they were a lot [easier]. They themselves understand the program now and there is no resistance or negative feedback from them in any way. (January 28, 2016) With this input and approval from the government, we thus revised the program towards a drop-in model that women attend as desired to meet the governmentdefined eligibility windows. This adaptation loosens CenteringPregnancy s requirement for stable groups but maintains some gestational-age focus by separating these drop-in groups into a second trimester (4 th -6 th month) and a third trimester (8 th -9 th ) group each month. The model largely abandons the participatory 39

action cycle because of this loss of group stability and further reduction in visit number. Instead, group problem solving takes place through creating individual birth plans for each participant. The adaptations have made the model more intuitive to the government health workers and thus made them more comfortable in implementing (CHP Program Manager, January 28, 2016). Through this process, the CHP has made tremendous strides in building the rapport relationship with the [District Health Office] and they've had to put a lot of effort into that -- just as much effort into that as into the real thinking behind what the components of the program look like, like how do you build a relationship with the government at the district level. (Impact Director, KI, December 7, 2015) While it was often an incredible struggle, our efforts to iteratively design an acceptable intervention eventually served to further both the intervention and the organizational partnerships with government healthcare providers. In retrospect, the scale of this success overshadows our daily frustrations, miscommunications and tensions. Implementation and Effectiveness Outcomes: Preliminary Results Overall, the adapted model maintains a focus on creating a supportive, empowering atmosphere and providing high-quality counseling and basic diagnostics because we agreed these are likely the essential intervention components. Fidelity to these components is measured through routine group checklists on quality of facilitation, content coverage and exam coverage. Quality of facilitation is primarily assessed through likert scale ratings of the group dynamics rating from either a class (0) or peer group discussion (5). Preliminary analysis of this data suggests that facilitation of groups is improving, with an increase from a median 2 to a median 4 rating since April 2015. Content coverage is assessed through documentation of the topics discussed. We have found that nurse-midwives and CHWs are consistently able to cover 100% of the content at least once over a four-visit cycle, though topics such as sex, self-esteem and violence are less frequently discussed. Exam coverage is 40

measured as the proportion of women receiving labs and ultrasounds before four and nine months, respectively. This data is not available for review at the time of writing. The feasibility of the intervention has greatly improved due to the simplicity of the scheduling system, which has also increased the compatibility with government protocols and organizational workflows as described above. Group care is also acceptable to women, as demonstrated by median group attendance of 6 (IQR 3-12) and preliminary data suggesting it is significantly more enjoyable than individual care (see Tables 7 and 8 for preliminary results). In addition, per on-going focus groups, group participation seems to generate a sense of empowerment and selfrespect that women find meaningful. With regard to effectiveness, preliminary analysis of our small pilot cohort data (51 and 56 women in control and intervention groups respectively) demonstrates no significant difference in institutional birth rates or antenatal care coverage. We plan to primary outcomes through yearly census survey data, which will be re-collected February-April of 2016, and which will have sufficient power to compare the intervention and control groups. Conclusions As others have found, implementing group antenatal care is challenging due to the degree of practice change required, from adjusting staffing and scheduling systems to facilitating an empowering learning experience rather than didactically imparting medical knowledge (36, 93, 99). In rural Nepal, these challenges were compounded by working within a public healthcare system where strict policy and incentives around antenatal care scheduling are not compatible with a stable gestational age group model. The necessity of adaptation in this environment is clear, but few roadmaps exist. In our experience, the process of adaptation was fraught with conflict that reflected an imperfect implementation climate, particularly with respect to a learning climate in which some team members perceived an implicit authority of researchers or 41

research-based priorities over program leaders, government partners and their experiences and priorities. This conflict developed largely as a result of a contentious implementation challenge: to adapt the gestational age groupings to accommodate government partners or to work around them and advocate for adaptation of government policy. On the one hand, there is no evidence to indicate whether gestational age matching affects the strength of group antenatal care, and on the other, the implementation of government policy around timing of care was both inconsistent and assessed to be more practical than evidence-based. To give up gestational age groupings threatened the potential effect of group antenatal care, but to push the government staff to flexibly implement the national protocol in order to accommodate group antenatal care threatened a nascent partnership that had implications for the broader community health program. Due largely to the complexities of team-building in a dispersed group with diverse educational, professional, and cultural backgrounds, a narrative arose in which this implementation challenge reflected power imbalances inherent to practice-based research. The implementation team felt: (1) the push for fidelity to a stable gestational age group model was rooted in research goals; (2) it lacked appropriate attention to government partnership, and (3) it was not safe to be fully transparent about the challenges due to traditional views regarding the authority of researchers. Inadvertent messaging from organization leadership that this aspect of fidelity was more important than partnership with government staff reinforced these perceptions. The research team struggled to respond to the implementation team s concerns because: (1) their desire for fidelity stemmed from a belief that stable gestational age groups are important for group antenatal care success (creating a supportive and empowering environment for participants), rather than a priority on research per se; (2) the government partnership challenges were seen as surmountable; and (3) they shared the belief that research should support but not drive program design and so were less attuned to the perception that they held such authority. 42

Problematically, this learning climate was in direct conflict to core organizational values around mutual support, transparency and the purpose of research. Extending Klein and Sorra s implementation-effectiveness model, then, both the implementation climate and the innovation had reduced fit in this setting and ultimately, these incompatibilities amplified one another to threaten the viability of the intervention. Thus, resolution ultimately required adaptation of both the intervention and our approach to and understanding of the role of implementation research in health systems strengthening. Our success navigating these challenges to develop a feasible, acceptable intervention and improve the dialogue around implementation research serves as an example of the power of commitment to a strong learning climate in which all team members feel valued and safe to fail forward (102). While it would be inappropriate to reduce all the group antenatal care implementation challenges to power imbalances and reify further the distinction between research and practice, we suggest that attentiveness to these dynamics may ease some of the tensions that inevitably arise when attempting to apply scientific evidence and methods in the swampy lowlands of healthcare systems in Nepal and globally (13). In fact, because implementation research toes the line between the scientific and the practical, it necessarily abuts these difficult power dynamics and often contrasting priorities (35). These forces may be particularly relevant in low-resource settings where there exists a long history of extractive research practice and an emphasis on global generalizability when considering the value of local health development work (103). If implementation science purports to address how and why interventions succeed in real-world contexts (1), our initial experiences suggest that the very act of researching such questions is a moderating force that must be considered as such when analyzing implementation outcomes. The effect of the researchimplementation relationship is perhaps the most clear with regard to data-driven facilitation strategies, such as process measure feedback loops that help program implementers calibrate their performance and improve program fidelity and 43

quality. These types of activities are often explicitly acknowledged as both implementation strategies and evaluation activities (38). We suggest further, however, that implementation research is not immune to power dynamics based in pervasive beliefs that scientific knowledge should be privileged over practical experience (101) and that these dynamics may affect implementation in unexpected and unintended ways by shaping the learning climate. Explicit attention to these power structures and biases is critical to understanding how implementation, and particularly adaptation, evolves when research is a core element. 44

Chapter 3 - Measurement challenges in implementation research: The case of child mortality A major challenge in implementation research on complex health systems interventions in resource-limited settings centers on the routine measurement of key population health outcomes. As described previously, Possible has defined three maternal and child health-related outcomes for its core implementation research: institutional birth, antenatal care coverage, and under-two child mortality. Here, I present the rationale and developing methods for assessing this third measure in our setting. In brief, we have chosen to use the under-two mortality rate as our metric for mortality measurement, for the following reasons: 1) as overall childhood mortality declines, deaths concentrate among children under the age of two; 2) two-year cohorts are shorter thus possibly more malleable in the short term of intervention trials; and 3) two-year cohorts are smaller, making prospective census cohorts as feasible in small populations as the cross-sectional sample surveys required to estimate under-five mortality. Finally, we propose the ultimate goal for assessing under-two mortality in implementation science trials should be to develop a continuous mortality surveillance system that assesses deaths as they occur. The Need and Challenge of Child Mortality Estimation Improving child survival is a central function of health systems globally. There remains a substantial implementation gap in translating evidence-based interventions into health systems improvements. Trials that test health systems innovations are central to meeting this gap. A robust measure assessing child mortality should form a primary outcome of these trials. Presently, countries use the under-five mortality rate (U5MR) to assess progress at reducing child mortality. Under-five mortality has been broadly applied as a metric across countries and over timelines of many years. It is the basis of Millennium Development Goal (MDG) 4, providing a target for reduction of child mortality by two thirds globally. While 45

useful for cross-country comparisons, U5MR exhibits problems in collection and analysis for intervention trials. In Nepal, like much of the developing world, one of the major barriers to collecting accurate population-based metrics at a sufficient level of spatio-temporal precision is infrequent census updates (usually every 10 years) and the absence of effective vital registration systems. It is estimated that in the in the district we work, only 35% of children under five have birth certificates, with a national average of 60% by age five (45, 104). In the absence of vital registration systems, estimates of child mortality are created from surveys of mothers about all the children they have had, both living and deceased, with either: (1) the direct method using full birth histories, which include birth and death dates for each child, or (2) the indirect method using summary birth histories (all children ever born and ever died) and some proxy measure of exposure (i.e. mother s age or time since first birth) to impute ages of those children (105, 106). Both methods rely on life table models of underlying child mortality patterns to estimate overall mortality risk before age five, and then to derive infant mortality (105). Most sources of national mortality estimates use the full birth history method as it most closely replicates a vital registration system, if one assumes that coverage and recall are not biased. Collection of full birth histories is time consuming, however, and the results are susceptible to so-called age-heaping around 12 months, which biases estimates of infant mortality upwards (107). Direct and especially indirect estimation require several assumptions about underlying fertility and mortality risk patterns that make estimation of recent trends in small populations (15,000 households or less) untenable when U5MR dips below 100 (108, 109). Infant mortality estimation from the same datasets is even more vulnerable to inaccurate assumptions (107). Furthermore, attempts to make U5MR more precise also reduce its applicability to real populations (110). For example, standard methods exclude data from the youngest mothers (ages 15-24 years old) because children born to these mothers are high-risk first births and are 46

likely from lower socioeconomic groups (107). Because these mothers have the highest proportion of young children their exclusion means both that recent mortality estimates cannot be reliably calculated and, perhaps equally as important, the estimate no longer reflects the true mortality risk of children born to those women (111). When the goal is to measure impact of specific interventions on reduction of mortality risk, introducing this kind of bias is problematic. Finally, until recently, U5MR was reported without confidence intervals necessary for ascertainment of significant differences between groups or across time.(106) With the development of new methods for constructing confidence bounds, it is now evident that estimates for high mortality contexts such as Nepal (defined as more than 40 deaths per 1000 births) have wide confidence intervals (112), limiting our ability to use these data for accurate comparisons across time or populations. While valuable for standardized cross-country comparison, it is less useful as a trial or quality improvement metric. We began to confront this challenge while studying an integrated health systems intervention that includes facility-based improvements, community health workerled home visits, as well as group antenatal care (see Chapter 2). Child mortality is a key indicator for these assessments and the complexities of its estimation in small, dispersed populations such as ours have shifted our focus onto the development of continuous surveillance methods that provide robust mortality data for children under age two. This context has shaped our thinking on the proposal we provide in this piece. We propose U2MR be used as the basis for health systems trials, and discuss our experiences assessing this measure in rural Nepal. We then outline the development of our continuous surveillance methods that can approximate a vital registration system. Under-two mortality assessment in health systems research trials As Nepal reaches the end of the MDG era, focus has shifted from addressing underfive mortality to infant and neonatal mortality (94). The infant mortality rate (IMR) and neonatal mortality rate (NMR) are defined as the number of deaths per 1000 47

live births occurring before 1 year and before 28 days, respectively. In 2011, Nepal reached its U5MR target of 54, but neonatal and infant mortality improvements continue to lag behind overall under-five mortality successes. Infant mortality improvement has slowed since 2006, falling just shy of the target of 36 (113). This pattern largely reflects stagnant neonatal mortality rates over the same period, suggestive of the many challenges in addressing perinatal and infant mortality due to complications such as birth asphyxia, low birth weight, and malnutrition. The pattern of U5MR decline in Nepal is not unique. Historically, transitions in childhood mortality have imparted significant changes in distribution of deaths such that as overall U5MR decreases, the proportion of deaths occurring in the perinatal and neonatal period increase. For example, when estimates of U5MR are less than 60, approximately 75% of deaths are concentrated in the infant period, with NMR/U5MR ratios of greater than 0.5 (107). Diarrhea and lower respiratory tract illness causes decrease almost linearly with decreasing U5MR, while vaccinepreventable illness and other causes remain stable at 25-35% of deaths (114). Thus, the proportion of infant (<1 year) versus child (1-5 years) mortality reflects both programmatically and epidemiologically relevant trends regarding underlying causes of death and the current state of the health system. U5MR does not give any indicator of mortality breakdown and could reflect many combinations of infant and child mortality. However, focusing solely on infant mortality is not ideal, as it is vulnerable to overestimation due to age-heaping at year one (107), does not capture as many vaccine-preventable illness causes of death and represents yet a smaller population for which sampling error is more problematic. Applying these data to Nepal, we estimate that over 80% of under-five mortality occurs under age two (107). Considering also the prioritization of health programming in Nepal focused on newborn health and the golden thousand days (conception to age two) (115), we believe under-two mortality (U2MR) provides a more robust and relevant estimate for our research trials and those in similar contexts. Two-year cohorts emphasize neonatal mortality risk but also include more 48

exposure time during which infectious diseases are a major cause of death. Similarly, the larger cohort attenuates problems with age-heaping and sampling error associated with IMR. Compared to U5MR, however, the estimation window is short enough to use for iterative improvement in intervention design, and can be assessed through targeted census methods collecting only the most recent two-year birth history for all women (i.e. birth and death dates of all children born in the prior two years). This last point is critical as it reduces data collection time and inaccurate age recall associated with full birth histories, and the estimation problems associated with summary birth histories in small populations (116). As it has for our program, this targeted census approach can evolve to continuous surveillance methods that essentially equate to vital registration. Our experience in rural Nepal In early 2015, we completed a census of our 7,500-household catchment area to establish baselines for U2MR, IMR, and NMR, as well as institutional birth rates, antenatal care coverage, and post-partum contraceptive prevalence rates. We developed the survey instrument from the Demographic Health Survey and Multiple Indicator Cluster Survey (117), for an Android-based mobile phone running an Open Data Kit application (118). We translated the tools into Nepali and data was stored on SurveyCTO s secure, cloud-based servers (119). Community health workers registered and mapped over 7500 households and collected targeted birth histories from all 1223 women reporting a birth in the previous two years. We achieved approximately 70% coverage of the eligible population, as 15% of households and 20% of identified women were unavailable due to migration or work responsibilities. Six months after the first round, these missing households and women were re-visited to finalize the baseline census. A major challenge that arose during that phase was verifying the identity of women who were previously unavailable (and not re-collecting data on previously enrolled women). This challenge highlighted the need for more reliable identification system. 49

In our initial dataset of 1237 births, the number of reported under-two deaths was surprisingly low. We estimate U2MR at 31 per 1000 live births and, interestingly, all deaths were reported before or at 1 year of age. Figure 6 is a heatmap visualizing the distribution and concentration of these deaths. To assess the degree of error, our health workers surveyed government health volunteers who report new pregnancies and deliveries to government facilities (but lack sufficient oversight to ensure full documentation). While imperfect, these data suggest we missed roughly 50% of deaths but captured more than 100% of stillbirths further illustrating the challenges with measuring child mortality through birth history survey methods or relying on government-reported data. Overall, our health workers felt women were willing to disclose child deaths, but occasionally neglected to report stillbirths, early neonatal deaths, and more recent deaths. Improved probing questions and screening for prior pregnancies instead of deliveries would likely improve birth history data in the future. As we transition to continuous household surveillance of pregnancies, births and deaths, however, this recall bias becomes less problematic. Towards continuous mortality assessment systems for child survival studies Using traditional estimation methods and sample survey data, U2MR is slightly more difficult to estimate than U5MR. Given the paucity of applicable life tables for age cohorts of less than five years, the baseline mortality risk for similar populations between age one and two years is often unknown. In addition, under-two deaths are fewer than under-five deaths, requiring a larger relative sample size. However, whether for under-two or under-five mortality, in small populations sample surveys are often underpowered and census methods more appropriate. For example, a traditional survey aiming to assess a 20% change in under-five mortality requires a sample of nearly 7000 women essentially our entire eligible population (108, 120). As our aim is to develop an implementation science program that is responsive and sustainable for ongoing health systems innovation, we argue that these census methods can and should be designed for continuous surveillance due to the increased spatio-temporal precision and real-time value offered compared to 50

cross-sectional, retrospective datasets. (See Figure 7 for a visualization of these differences.) With continuous surveillance of the entire population, any combination of mortality indicators could theoretically be calculated. Again, however, we emphasize undertwo mortality as the highest-value indicator both for applicability and practicability. Based upon the Nepal government s projections for our catchment area, approximately 10% of women of childbearing age deliver a child every year (45). Given Nepali women space births by a median 2-3 years and have 3 children, we roughly estimate that 16-18% of women have children under age two, 35-40% have children under age five and 90% have ever had children (121). Continuous surveillance of all women with children under age five would thus more than double the women followed, overburdening our health workers who have primary patientcare responsibilities with likely less impact on overall child survival. After completing the first census round, we are now transitioning to a continuous surveillance program in which we screen all eligible women for pregnancy and follow all pregnant women and maternal-infant dyads up to 2 years postpartum. With a mobile phone-enabled screening algorithm, CHWs visit every household in the catchment area (7,500, roughly 400 per CHW) every three months to screen women of childbearing age for unexplained amenorrhea or new pregnancy. Those who screen positive are given a rapid urine pregnancy test and enrolled in follow-up testing or antenatal care as needed. Our CHWs successfully piloted the mobile pregnancy-screening tool with 200 women in August and September of 2015. Of these women, 14 (6% total) reported a known pregnancy of greater than five months gestation 4, and 24 (12% total) screened at risk with either unexplained amenorrhea for more than six weeks (n=21, 11% total) or no return of menses three or more months post-pregnancy (n=3, <1% total). Of these at-risk women, two tested positive with a rapid urine test (8%), 10 tested negative (42%), and 12 women refused testing (50%). The two women with newly identified pregnancies 4 Note, 12 of 14 women who reported pregnancies were also tested to confirm, with a 100% positive rate. 51

were dated between 4-6 months gestation by last menstrual period. Of those who tested negative, 3 women (30%) had only 6-12 weeks of amenorrhea, and would need a repeat test to confirm pregnancy status. In addition, of the women who refused, three (25%) had only 6-12 weeks of amenorrhea, suggesting at least three potential pregnancies could have been missed. The remaining 9 women (75%) had not menstruated in more than 7 months. Together, these data suggest that our pilot protocol has not yet been refined sufficiently to identify pregnancies at an early gestational age. The rate of positive testing among women with 4-6 months unexplained amenorrhea (40%) does imply women may be unlikely to report suspected pregnancies even in the second trimester, further suggesting that pregnancy screening may increase antenatal service uptake. Tabulated data of test results by screening category and by (suspected) gestational age are presented in Tables 9 and 10, respectively. As identified in the re-census phase, a major issue for continuous surveillance centers on reliable identification of households, women and children. Working with government partners, we are beginning to enumerate households with durable markers (tin plaques) and map our catchment area. We are also working with CommCare (122) and SimPrints (123) to improve upon the piloted tools using a case management application with biometrics integration for fingerprinting identification of mothers and children. We anticipate that implementation of this tool will begin in May 2016. Once implemented, community health workers will document births and deaths during routine visits to provide pregnant women and children under age two with preventive care and referrals for group antenatal and pediatric care or urgent facility-based services. These data will be documented with mobile forms integrated into the electronic medical record recently implemented at the hospital. Through routine outreach we plan to create a relatively real-time vital registration system that allows us to assess mortality with spatio-temporal precision and accuracy. Over time, we will also implement a verbal autopsy program assessing cause of death through standardized family interviews (124-126), further deepening our understanding of healthcare system gaps. 52

Conclusions Entering the post-mdg era, the focus of mortality measurements must shift towards the development of robust impact assessments that are relevant to changing mortality patterns and feasible for community-level program implementation. Given the relatively rare nature of childhood mortality and the absence of effective vital registration systems in much of the developing world, the precision of child mortality estimation is greatly affected by sample size and estimates cannot easily be disaggregated to district or sub-district levels. As overall childhood mortality declines, deaths concentrate in the neonatal and infant period. Given that two-year cohorts are shorter and likely subject to less recall bias, we believe under-two mortality will be a more useful metric for our health systems trials work. Furthermore, while precise sample-survey mortality estimation is unlikely to be feasible for frequent assessments of small populations, the establishment of continuous household surveillance that follows pregnant mothers and children through age two is feasible for the measurement of under-two mortality in quality improvement programs and research studies on child survival. 53

Summary and Suggestions for Future Work Implementation research for health systems strengthening is a challenging, lessthan-clearly defined enterprise with great importance for the development of responsive, learning healthcare systems that can overcome critical implementation gaps in resource-limited settings. In this thesis, I have presented a broad overview of the development of such a program in rural Nepal with the organization Possible. Through the past two years of study and iterative improvement, we have assessed community needs regarding institutional birth, implemented and adapted an evidence-based model of group antenatal care, and begun building out a sensitive mobile data collection system for monitoring implementation and measuring intervention outcomes including under-two mortality. This process evolved from an initial non-implementation research study; using a quasi-experimental design, we demonstrated CEmOC expansion is associated with significant demand generation for institutional births in a community with BEmOC coverage but limited prior access to CEmOC. We found this effect was mediated by increased safety perceptions and possibly normalization of the practice, and moderated by a lack of social support and birth planning for mobilization of (limited) resources. The study presented, while not focusing on implementation outcomes per se, illuminates the broader agenda for Possible s implementation research program namely, iterative improvement of the healthcare system for equity and quality through careful study of the impact of interventions and the remaining gaps in quality or coverage. In response to the evidence outlined in Chapter 1, we began a hybrid (type I) trial of group antenatal care. Perhaps unsurprisingly given the relatively inexperienced and dispersed team, we faced significant challenges navigating adaptation decisions during this process. While the intervention was ultimately found to be feasible and acceptable, the process of iterative design was fraught with conflict regarding the appropriate emphasis on fidelity versus adaptation. Though the value of implementation research largely rests in finding the balance between internal and external validity that will produce the most useful, actionable knowledge for health 54

systems improvement, we find that much like other fields it is not devoid of power imbalances that may implicitly preference research-oriented perspectives over practice-oriented ones. In our experience, these power dynamics and misaligned priorities created tension in the implementation and learning climate, which further reduced the innovation-values fit of our group antenatal care intervention. As such, I propose that other implementation researchers consider these dynamics explicitly in the evaluation of implementation effectiveness. Future work in this area could involve testing of the relationships between research activities and partnerships on the learning climate and innovation-values fit of interventions through cross-case comparisons. Practically speaking, attentiveness to these concerns may ease some tensions in implementation research and improve the partnerships necessary for successful iterative design in resource-limited settings. Finally, to push forward our capacity to measure effectiveness in hybrid studies such as the one described in Chapter 2, I address the challenge of measuring child mortality outcomes in resource-limited settings, a barrier we faced as we began to develop the implementation research program described herein. Given the limitations of child mortality measurement in small populations that often necessitate exhaustive samples, the inherent recall bias in birth histories, and the long observation windows required for under-five mortality measurement, we propose under-two mortality may be a more feasible measure for sub-national health systems trials. We believe it sufficiently captures relevant epidemiological trends in child survival to guide maternal and child health programming and can be more reliably captured through truncated birth histories. Through iterative adaptation of this method, we have arrived at a continuous surveillance model we believe will robustly measure under-two mortality, as well as a number of other key program indicators prospectively for real-time quality improvement and outcomes assessment. Over the next several years, we plan to refine our methods and evaluate its reliability and validity in this setting as we expand to other service areas. We encourage researchers in similar settings to do the same, and to explore the reliability and validity of under-two mortality at different levels of analysis. 55

References 1. Peters DH, Tran N, Adam T. Implementation Research in Health: A Practical Guide. Alliance for Health Policy and Systems Research, World Health Organization, 2013. 2. Zachariah R, Ford N, Maher D, Bissell K, Van den Bergh R, van den Boogaard W, et al. Is operational research delivering the goods? The journey to success in lowincome countries. The Lancet Infectious Diseases. 2012;12(5):415-21. doi: 10.1016/s1473-3099(11)70309-7. 3. Zachariah R, Harries AD, Ishikawa N, Rieder HL, Bissell K, Laserson K, et al. Operational research in low-income countries: what, why, and how? The Lancet. 2009;9. 4. USAID, editor Statement on Advancing Implementation Research and Delivery Science. Global Health Systems Research Symposium; 2014; Cape Town, South Africa: WHO. 5. Whitworth J, Sewankambo NK, Snewin VA. Improving Implementation: Building Research Capacity in Maternal, Neonatal, and Child Health in Africa. PLoS Medicine. 2010;7(7):e1000299. doi: 10.1371/journal.pmed.1000299. PubMed PMID: PMC2897765. 6. Maru DS, Andrews J, Schwarz D, Schwarz R, Acharya B, Ramaiya A, et al. Crossing the quality chasm in resource-limited settings. Globalization and Health. 2012;8(41). 7. Hussein J, Newlands D, D'Ambruoso L, Thaver I, Talukder R, Besana G. Identifying practices and ideas to improve the implementation of maternal mortality reduction programmes: findings from five South Asian countries. BJOG. 2010;117(3):304-13. doi: 10.1111/j.1471-0528.2009.02457.x. PubMed PMID: 20015302. 8. Bhutta ZA, Darmstadt GL, Haws RA, Yakoob MY, Lawn JE. Delivering interventions to reduce the global burden of stillbirths: improving service supply and community demand. BMC Pregnancy Childbirth. 2009;9 Suppl 1:S7. doi: 10.1186/1471-2393-9-S1-S7. PubMed PMID: 19426470; PubMed Central PMCID: PMCPMC2679413. 9. Curran GM, Bauer M, Mittman B, Pyne JM, Stetler C. Effectivenessimplementation hybrid designs: combining elements of clinical effectiveness and 56

implementation research to enhance public health impact. Med Care. 2012;50(3):217-26. doi: 10.1097/MLR.0b013e3182408812. PubMed PMID: 22310560; PubMed Central PMCID: PMCPMC3731143. 10. Barton TD. Understanding Practitioner Ethnography. Nurse Researcher. 2008;15(2). 11. Klein KJ, Sorra JS. The Challenge of Innovation Implementation. The Academy of Management Review. 1996;21(4):1055-80. 12. Pakenham-Walsh N. Learning from one another to bridge the know-do gap. BMJ. 2004;329(7475):1189. doi: 10.1136/bmj.329.7475.1189. 13. Schon DA. The Reflective Practitioner: How Professionals Think in Action. New York: Basic Books; 1983. 14. McKibbon KA, Lokker C, Wilczynski NL, Ciliska D, Dobbins M, Davis DA, et al. A cross-sectional study of the number and frequency of terms used to refer to knowledge translation in a body of health literature in 2006: a Tower of Babel? Implement Sci. 2010;5:1-11. 15. Chaudoir SR, Dugan AG, Barr CHI. Measuring factors affecting implementation of health innovations: a systematic review of structural, organizational, provider, patient, and innovation level measures. Implementation Science : IS. 2013;8:22-. doi: 10.1186/1748-5908-8-22. PubMed PMID: PMC3598720. 16. Allotey P, Reidpath DD, Ghalib H, Pagnoni F, Skelly WC. Efficacious, effective, and embedded interventions: implementation research in infectious disease control. BMC Public Health. 2008;8:343. doi: 10.1186/1471-2458-8-343. PubMed PMID: 18826655; PubMed Central PMCID: PMCPMC2567977. 17. Remme JH, Adam T, Becerra-Posada F, D'Arcangues C, Devlin M, Gardner C, et al. Defining research to improve health systems. PLoS Med. 2010;7(11):e1001000. doi: 10.1371/journal.pmed.1001000. PubMed PMID: 21124816; PubMed Central PMCID: PMCPMC2993153. 18. Peters DH, Adam T, Alonge O, Agyepong IA, Tran N. Republished research: Implementation research: what it is and how to do it: Implementation research is a growing but not well understood field of health research that can contribute to more effective public health and clinical policies and programmes. This article provides a broad definition of implementation research and outlines key principles for how to do it. British Journal of Sports Medicine. 2014;48(8):731-6. doi: 10.1136/bmj.f6753. 57

19. Lewis CC, Fischer S, Weiner BJ, Stanick C, Kim M, Martinez RG. Outcomes for implementation science: an enhanced systematic review of instruments using evidence-based rating criteria. Implement Sci. 2015;10:155. doi: 10.1186/s13012-015-0342-x. PubMed PMID: 26537706; PubMed Central PMCID: PMCPMC4634818. 20. Tabak RG, Khoong EC, Chambers DA, Brownson RC. Bridging research and practice: models for dissemination and implementation research. Am J Prev Med. 2012;43(3):337-50. doi: 10.1016/j.amepre.2012.05.024. PubMed PMID: 22898128; PubMed Central PMCID: PMCPMC3592983. 21. Durlak JA, DuPre EP. Implementation matters: a review of research on the influence of implementation on program outcomes and the factors affecting implementation. Am J Community Psychol. 2008;41(3-4):327-50. doi: 10.1007/s10464-008-9165-0. PubMed PMID: 18322790. 22. Dusenbury L, Brannigan R, Falco M, Hansen WB. A review of research on fidelity of implementation: implications for drug abuse prevention in school settings. Health Educ Res. 2003;18(2):237-56. 23. Breitenstein SM, Gross D, Garvey C, Hill C, Fogg L, Resnick B. Implementation Fidelity in Community-Based Interventions. Research in nursing & health. 2010;33(2):164-73. doi: 10.1002/nur.20373. PubMed PMID: PMC3409469. 24. Galavotti C, Kuhlmann AK, Kraft JM, Harford N, Petraglia J. From innovation to implementation: the long and winding road. Am J Community Psychol. 2008;41(3-4):314-26. doi: 10.1007/s10464-008-9172-1. PubMed PMID: 18297390. 25. Kilbourne AM, Neumann MS, Pincus HA, Bauer MS, Stall R. Implementing evidence-based interventions in health care: application of the replicating effective programs framework. Implement Sci. 2007;2:42. doi: 10.1186/1748-5908-2-42. PubMed PMID: 18067681; PubMed Central PMCID: PMCPMC2248206. 26. van Daele T, van Audenhove C, Hermans D, van den Bergh O, van den Broucke S. Empowerment implementation: enhancing fidelity and adaptation in a psycho-educational intervention. Health Promot Int. 2014;29(2):212-22. doi: 10.1093/heapro/das070. PubMed PMID: 23257062. 27. Bunce AE, Gold R, Davis JV, McMullen CK, Jaworski V, Mercer M, et al. Ethnographic process evaluation in primary care: explaining the complexity of implementation. BMC Health Serv Res. 2014;14(607). 28. Chandler C, DiLiberto D, Nayiga S, Taaka L, Nabirye C, Kayendeke M, et al. The PROCESS study: a protocol to evaluate the implementation, mechanisms of effect 58

and context of an intervention to enhance public health centres in Tororo, Uganda. Implement Sci. 2013;8(113). 29. Gearing RE, El-Bassel N, Ghesquiere A, Baldwin S, Gillies J, Ngeow E. Major ingredients of fidelity: a review and scientific guide to improving quality of intervention research implementation. Clin Psychol Rev. 2011;31(1):79-88. doi: 10.1016/j.cpr.2010.09.007. PubMed PMID: 21130938. 30. Carroll C, Patterson M, Wood S, Booth A, Rick J, Balain S. A conceptual framework for implementation fidelity. Implement Sci. 2007;2:40. doi: 10.1186/1748-5908-2-40. PubMed PMID: 18053122; PubMed Central PMCID: PMCPMC2213686. 31. Slaughter SE, Hill JN, Snelgrove-Clarke E. What is the extent and quality of documentation and reporting of fidelity to implementation strategies: a scoping review. Implement Sci. 2015;10:129. doi: 10.1186/s13012-015-0320-3. PubMed PMID: 26345357; PubMed Central PMCID: PMCPMC4562107. 32. Perrin KM, Burke SG, O'Connor D, Walby G, Shippey C, Pitt S, et al. Factors contributing to intervention fidelity in a multi-site chronic disease self-management program. Implement Sci. 2006;1:26. doi: 10.1186/1748-5908-1-26. PubMed PMID: 17067388; PubMed Central PMCID: PMCPMC1636065. 33. Leykum LK, Pugh JA, Lanham HJ, Harmon J, McDaniel RR, Jr. Implementation research design: integrating participatory action research into randomized controlled trials. Implement Sci. 2009;4:69. doi: 10.1186/1748-5908-4-69. PubMed PMID: 19852784; PubMed Central PMCID: PMCPMC2770984. 34. Okwaro FM, Chandler CI, Hutchinson E, Nabirye C, Taaka L, Kayendeke M, et al. Challenging logics of complex intervention trials: community perspectives of a health care improvement intervention in rural Uganda. Soc Sci Med. 2015;131:10-7. doi: 10.1016/j.socscimed.2015.02.032. PubMed PMID: 25748110. 35. Proctor EK, Landsverk J, Aarons GA, Chambers DA, Mittman B. Implementation Research in Mental Health Services: an Emerging Science with Conceptual, Methodological, and Training challenges. Adm Policy Ment Health. 2009;36(1). doi: 10.1007/s10488-008-0197-4. 36. Novick G, Sadler LS, Knafl KA, Groce NE, Kennedy HP. In a hard spot: providing group prenatal care in two urban clinics. Midwifery. 2013;29(6):690-7. doi: 10.1016/j.midw.2012.06.013. PubMed PMID: 22884892; PubMed Central PMCID: PMCPMC3498538. 59

37. Miller WL, Crabtree BF, Harrison MI, Fennell ML. Integrating mixed methods in health services and delivery system research. Health Serv Res. 2013;48(6 Pt 2):2125-33. doi: 10.1111/1475-6773.12123. PubMed PMID: 24279834; PubMed Central PMCID: PMCPMC4097838. 38. Hirschhorn LR, Semrau K, Kodkany B, Churchill R, Kapoor A, Spector J, et al. Learning before leaping: integration of an adaptive study design process prior to initiation of BetterBirth, a large-scale randomized controlled trial in Uttar Pradesh, India. Implement Sci. 2015;10:117. doi: 10.1186/s13012-015-0309-y. PubMed PMID: 26271331; PubMed Central PMCID: PMCPMC4536663. 39. NPHC. National Population and Housing Census 2011 (National Report). Kathmandu, Nepal: Central Bureau of Statistics, National Planning Commission Secretariat, 2012. 40. Maru S. Determinants of Institutional Birth in Rural Nepal. 2014. 41. Maru DS-R, Schwarz R, Schwarz D, Andrews J, Panizales MT, Karelas G, et al. Implementing surgical services in a rural, resource-limited setting: a study protocol. BMJ Open. 2011;1(1):e000166. doi: 10.1136/bmjopen-2011-000166. PubMed PMID: PMC3191574. 42. Group HSD. Healthcare Systems Design Group: Implementation Research for Patient-Centered Design: Possible; 2015. Available from: http://www.hsdg.partners.org. 43. WHO. Maternal Mortality Fact Sheet. Geneva, Switzerland: World Health Organization, May 2012. 44. Hofmeyr GJ, Haws RA, Bergstrom S, Lee AC, Okong P, Darmstadt GL, et al. Obstetric care in low-resource settings: what, who, and how to overcome challenges to scale up? Int J Gynaecol Obstet. 2009;107 Suppl 1:S21-44, s-5. Epub 2009/10/10. doi: 10.1016/j.ijgo.2009.07.017. PubMed PMID: 19815204. 45. NDHS. Nepal Demographic and Health Survey. Ministry of Health and Population (MOHP) Nepal; 2011-2012. 46. NDHS. Nepal Demographic and Health Survey, 2011. Kathmandu, Nepal: Population Division, Ministry of Health and Population, 2012. 47. Thaddeus S, Maine D. Too far to walk: maternal mortality in context. Soc Sci Med. 1994;38(8):1091-110. Epub 1994/04/01. PubMed PMID: 8042057. 60

48. Gabrysch S, Campbell OM. Still too far to walk: literature review of the determinants of delivery service use. BMC Pregnancy Childbirth. 2009;9:34. Epub 2009/08/13. doi: 10.1186/1471-2393-9-34. PubMed PMID: 19671156; PubMed Central PMCID: PMCPmc2744662. 49. Witter S, Khadka S, Nath H, Tiwari S. The national free delivery policy in Nepal: early evidence of its effects on health facilities. Health policy and planning. 2011;26(suppl 2):ii84-ii91. doi: 10.1093/heapol/czr066. 50. Simkhada B, Porter MA, van Teijlingen ER. The role of mothers-in-law in antenatal care decision-making in Nepal: a qualitative study. BMC Pregnancy Childbirth. 2010;10:34. Epub 2010/07/03. doi: 10.1186/1471-2393-10-34. PubMed PMID: 20594340; PubMed Central PMCID: PMCPmc2910658. 51. Simkhada B, van Teijlingen ER, Porter M, Simkhada P. Major problems and key issues in Maternal Health in Nepal. Kathmandu University medical journal (KUMJ). 2006;4(2):258-63. Epub 2008/07/08. PubMed PMID: 18603913. 52. Brunson J. Confronting maternal mortality, controlling birth in Nepal: the gendered politics of receiving biomedical care at birth. Soc Sci Med. 2010;71(10):1719-27. Epub 2010/08/18. doi: 10.1016/j.socscimed.2010.06.013. PubMed PMID: 20713304. 53. Morrison J, Thapa R, Basnet M, Budhathoki B, Tumbahangphe KM, Manandhar DS, et al. Exploring the first delay: a qualitative study of home deliveries in Makwanpur district Nepal. BMC Pregnancy & Childbirth. 2014;14(89). 54. Hurst TE, Semrau K, Patna M, Gawande A, Hirschhorn LR. Demand-side interventions for maternal care: evidence of more use, not better outcomes. BMC Pregnancy Childbirth. 2015;15(1):297. Epub 2015/11/17. doi: 10.1186/s12884-015-0727-5. PubMed PMID: 26566812; PubMed Central PMCID: PMCPmc4644345. 55. Lee AC, Lawn JE, Cousens S, Kumar V, Osrin D, Bhutta ZA, et al. Linking families and facilities for care at birth: what works to avert intrapartum-related deaths? Int J Gynaecol Obstet. 2009;107 Suppl 1:S65-85, S6-8. doi: 10.1016/j.ijgo.2009.07.012. PubMed PMID: 19815201; PubMed Central PMCID: PMCPMC3428847. 56. WHO. Monitoring emergency obstetric care: a handbook. Geneva, Switzerland: World Health Organization, 2009. 61

57. Paxton A, Maine D, Freedman L, Fry D, Lobis S. The evidence for emergency obstetric care. Int J Gynaecol Obstet. 2005;88(2):181-93. Epub 2005/02/08. doi: 10.1016/j.ijgo.2004.11.026. PubMed PMID: 15694106. 58. Lawn JE, Kinney M, Lee AC, Chopra M, Donnay F, Paul VK, et al. Reducing intrapartum-related deaths and disability: can the health system deliver? Int J Gynaecol Obstet. 2009;107 Suppl 1:S123-40, s40-2. Epub 2009/10/10. doi: 10.1016/j.ijgo.2009.07.021. PubMed PMID: 19815205. 59. Rana TG, Chataut BD, Shakya G, Nanda G, Pratt A, Sakai S. Strengthening emergency obstetric care in Nepal: The Women's Right to Life and Health Project (WRLHP). Int J Gynaecol Obstet. 2007;98(3):271-7. Epub 2007/07/03. doi: 10.1016/j.ijgo.2007.05.017. PubMed PMID: 17604032. 60. Islam MT, Hossain MM, Islam MA, Haque YA. Improvement of coverage and utilization of EmOC services in southwestern Bangladesh. Int J Gynaecol Obstet. 2005;91(3):298-305; discussion 283-4. doi: 10.1016/j.ijgo.2005.06.029. PubMed PMID: 16226760. 61. NSMP. Study on Utilisation of Emergency Obstetric Care (EmOC) in Selected Districts of Nepal. Kathmandu, Nepal: Institute of Medicine, Prepared for the Nepal Safer Motherhood Project, 2004. 62. Amudhan S, Mani K, Rai SK, Pandav CS, Krishnan A. Effectiveness of demand and supply side interventions in promoting institutional deliveries--a quasiexperimental trial from rural north India. Int J Epidemiol. 2013;42(3):769-80. doi: 10.1093/ije/dyt071. PubMed PMID: 23918850. 63. Dickson KE, Simen-Kapeu A, Kinney MV, Huicho L, Vesel L, Lackritz E, et al. Every Newborn: health-systems bottlenecks and strategies to accelerate scale-up in countries. The Lancet. 2014;384(9941):438-54. doi: 10.1016/s0140-6736(14)60582-1. 64. Sherr K, Cuembelo F, Michel C, Gimbel S, Micek M, Kariaganis M, et al. Strengthening integrated primary health care in Sofala, Mozambique. BMC Health Serv Res. 2013;13 Suppl 2:S4. doi: 10.1186/1472-6963-13-S2-S4. PubMed PMID: 23819552; PubMed Central PMCID: PMCPMC3668215. 65. Borkan J. Immersion/Crystallization. In: Crabtree BF, Miller WL, editors. Doing Qualitative Reseasrch, 2nd ed. Thousand Oaks, CA: Sage Publications; 2014. p. 179-94. 62

66. Sorensen G, Emmons K, Hunt MK, Barbeau E, Goldman R, Peterson K, et al. Model for incorporating social context in health behavior interventions: applications for cancer prevention for working-class, multiethnic populations. Preventive medicine. 2003;37(3):188-97. Epub 2003/08/14. PubMed PMID: 12914824. 67. Ybema S, Yanow D, Wels H, Kamsteeg F. Organizational Ethnography: Studying the Complexity of Everyday Life. London: Sage; 2009. 68. Evans C, Lambert H. Implementing community interventions for HIV prevention: insights from project ethnography. Soc Sci Med. 2008;66(2):467-78. doi: 10.1016/j.socscimed.2007.08.030. PubMed PMID: 17920740. 69. Parker R, Ehrhardt AA. Through an Ethnographic Lens: Ethnographic Methods, Comparative Analysis, and HIV/AIDS Research. AIDS and Behavior. 2001;5(2):105-14. 70. Tomoaia-Cotisel A, Scammon DL, Waitzman NJ, Cronholm PF, Halladay JR, Driscoll DL, et al. Context matters: the experience of 14 research teams in systematically reporting contextual factors important for practice change. Ann Fam Med. 2013;11 Suppl 1:S115-23. doi: 10.1370/afm.1549. PubMed PMID: 23690380; PubMed Central PMCID: PMCPMC3707255. 71. Down S, Hughes M. When the subject and the researcher speak together: Co-producing organizational ethnography. In: Ybema S, Yanow D, Wels H, Kamsteeg F, editors. Organizational Ethnography: Studying the Complexity of Everyday Life. London: Sage; 2009. 72. Schein E. Legitimating Clinical Research in the Study of Organizational Culture. In: Management MSSo, editor. 1991. 73. Manias E, Street A. Rethinking ethnography: reconstructing nursing relationships. Journal of Advanced Nursing. 2001;33(2):234-42. 74. Asana. [cited 2016 January 17]. Available from: http://www.asana.com/product. 75. Damschroder LJ, Aron DC, Keith RE, R. KS, Alexander JA, Lowery JC. Fostering implementation of health services research findings into practice: A consolidated framework for advancing implementation science: a consolidated framework for advancing implementation science. Implement Sci. 2009;4(50). 63

76. Damschroder LJ, Lowery JC. Evaluation of a large-scale weight management program using the consolidated framework for implementation research (CFIR). 2013;8(51). 77. Grol R, Bosch MC, Hulscher M, Eccles MP, Wensing M. Planning and studying improvement in patient care: the use of theoretical perspectives. Milbank Q. 2007;85:93-138. 78. Iott I, Gerrish K, Booth A, Field B. Testing the Consolidated Framework for Implementation Research on health care innovations from South Yorkshire. J Eval Clin Pract. 2013;19:915-24. 79. Sorensen JL, Kosten T. Developing the tools of implementation science in substance use disorders and treatment: application of the Consolidated Framework for Implementation Research. Psychol Addict Behav. 2011;25:262-8. 80. Weiner BJ, Belden CM, Bergmire DM, Johnston M. The meaning and measurement of implementation climate. Implementation Science : IS. 2011;6:78-. doi: 10.1186/1748-5908-6-78. PubMed PMID: PMC3224582. 81. Our For-Impact Culture Code: Possible; [cited 2016 January 31]. Available from: http://possiblehealth.org/wp-content/uploads/2015/10/possible-culture- Code.pdf. 82. Smith-Rohrberg Maru D, Sharma A, Andrews J, Basu S, Thapa J, Oza S, et al. Global Health Delivery 2.0: Using Open-Access Technologies for Transparency and Operations Research. PLoS Med. 2009;6(12). doi: 10.1371/. 83. Schwarz D, Schwarz R, Gauchan B, Andrews J, Sharma R, Karelas G, et al. Implementing a systems-oriented morbidity and mortality conference in remote rural Nepal for quality improvement. BMJ Qual Saf. 2011;20(12):1082-8. doi: 10.1136/bmjqs-2011-000273. PubMed PMID: 21949441; PubMed Central PMCID: PMCPMC3228264. 84. Schwarz R, Maru DS, Schwarz D, Acharya B, Acharya B, Rajbhandari R, et al. Deliverying Health Care in Severely Resource-Constrained Settings: On Partnership. Narrative Inquiry in Bioethics. 2012;2(2). 85. Bae J, Baffour-Awuah C, Chantanapongwanij O, Ottensmann S. Nyaya Health: From Nepal, to Global Health. Operations Strategy Final Project2012. 64

86. Maru DS, Farmer PE. Human rights and health systems development: Confronting the politics of exclusion and the economics of inequality. Health and Human Rights. 2012;14(2). 87. Walley J, Khan MA, Shah SK, Witter S, Wei X. How to get research into practice: first get practice into research. Bulletin of the World Health Organization. 2007;85(6):424-. doi: 10.2471/blt.07.042531. 88. Rising SS. Centering Pregnancy: An interdisciplinary model of empowerment. Journal of Nurse-Midwifery. 1998;43(1):46-54. 89. Massey Z, Rising SS, Ickovics J. CenteringPregnancy group prenatal care: Promoting relationship-centered care. Journal of obstetric, gynecologic, and neonatal nursing : JOGNN / NAACOG. 2006;35(2):286-94. doi: 10.1111/j.1552-6909.2006.00040.x. PubMed PMID: 16620257. 90. Project PC. Community Mobilisation through Women s Groups to Improve the Health of Mothers and Babies: Good Practice Guide. BADAS, Ekjut, Women and Children First (UK), UCL. 2013. 91. Prost A, Colbourn T, Seward N, Azad K, Coomarasamy A, Copas A, et al. Women's groups practising participatory learning and action to improve maternal and newborn health in low-resource settings: a systematic review and metaanalysis. The Lancet. 2013;381(9879):1736-46. doi: 10.1016/s0140-6736(13)60685-6. 92. Ickovics JR, et al. Group prenatal care and preterm birth weight: results from a matched cohort study at public clinics. Obstetrics & Gynecology. 2003;102.5(Part I):1051-7. 93. Novick G, Reid AE, Lewis J, Kershaw TS, Rising SS, Ickovics JR. Group prenatal care: model fidelity and outcomes. American journal of obstetrics and gynecology. 2013;209(2):112 e1-6. doi: 10.1016/j.ajog.2013.03.026. PubMed PMID: 23524175; PubMed Central PMCID: PMC3726551. 94. MoHP. Annual Report, Department of Health Services 2070/2071 (2013-2014). Teku, Kathmandu: Ministry of Health and Population Nepal, Department of Health Services, 2014. 95. Hoope-Bender PT, Kearns A, Caglia J, O. T, A. L. Group Care: Alternative models of care delivery to increase women's access, engagement, and satisfaction. Women and Health Inititiative, Maternal Health Task Force, Harvard School of Public Health, 2014. 65

96. Manandhar DS, Osrin D, Shrestha BP, Mesko N, Morrison J, Tumbahangphe KM, et al. Effect of a participatory intervention with women's groups on birth outcomes in Nepal: cluster-randomised controlled trial. The Lancet. 2004;364(9438):970-9. doi: 10.1016/s0140-6736(04)17021-9. 97. Morrison J, et al. Understanding how women's groups improve maternal and newborn health in Makwanpur, Nepal: a qualitative study. Int Health. 2010;2(1):25-35. 98. WHO. WHO recommendation on community mobilization through facilitated participatory learning and action cycles with women s groups for maternal and newborn health. Geneva, Switzerland: World Health Organization, 2014. 99. Baldwin K, Phillips G. Voices Along the Journey: Midwives' Perceptions of Implementing the CenteringPregnancy Model of Prenatal Care. J Perinat Educ. 2011;20(4):210-7. doi: 10.1891/1058-1243.20.4.210. PubMed PMID: 22942623; PubMed Central PMCID: PMCPMC3210635. 100. Patil CA, E; Klima, C; Kaponda, C; Leshabari, S; Vonderheid, S; Kamanga, M; Norr, K. CenteringPregnancy-Africa: A pilot of group antenatal care to address Millennium Development Goals. Midwifery. 2013. doi: 10.1016/j.midw.2013.05.008i. 101. Wallerstein N, Duran B. Community-Based Participatory Research Contributions to Intervention Research: The Intersection of Science and Practice to Improve Health Equity. American Journal of Public Health. 2010;100 (Supp 1)(51). doi: 10.2105/AJPH.2009. 102. Maxwell J. Failing Forward: Turning Mistakes Into Stepping-stones for Success. Nashville, Tenn.: Thomas Nelson; 2000. 103. Adams V, Craig S, Samen A. Alternative accounting in maternal and infant global health. Global Public Health: An International Journal for Research, Policy and Practice. 2015. 104. UNCF. Every Child's Birth Right: Inequities and trends in birth registration. New York: United Nations Children's Fund, 2013. 105. Hill K, You D, Inoue M, Oestergaar M. Child Mortality Estimation: Accerlated Progress in Reducing Global Child Mortality 1990-2010. PLoS Med. 2012;9(8):e1001303. 66

106. Levels and Trends in Child Mortality: Estimates Developed by the UN Interagency Group for Child Mortality Estimation. New York: UNICEF, World Health Organization, The World Bank, United Nations, 2012. 107. Guillot M, Gerland P, Pelletier F, Saabneh A. Child Mortality Estimation: A Global Overview of Infant and Child MOrtality Age Patterns in Light of New Empirical Data. PLoS Med. 2012;9(8):e1001299. 108. L. KE, Arnold F, Williams BG, Nahlen BL, Snow RW. Monitoring trends in under-5 mortality rates through national birth history surveys. Int J Epidemiol. 2004;33(6):1293-301. 109. Silva R. Child mortality estimation: consistency of under-five mortality rate estimates using full birth histories and summary birth histories. PLoS Med. 2012;9(8):e1001296. doi: 10.1371/journal.pmed.1001296. PubMed PMID: 22952436; PubMed Central PMCID: PMCPMC3429405. 110. Pedersen J, Liu J. Appropriate Time Periods for Child Mortality Estimates from Full Birth Histories. PLoS Med. 2012;9(8). 111. Rajaratnam JK, Tran LN, Lopez AD, Murray CJ. Measuring under-five mortality: validation of new low-cost methods. PLoS Med. 2010;7(4):e1000253. doi: 10.1371/journal.pmed.1000253. PubMed PMID: 20405055; PubMed Central PMCID: PMCPMC2854123. 112. Alkema L, New JR. Progress toward global reduction in under-five mortality: a bootstrap analysis of uncertainty in Millennium Development Goal 4 estimates. PLoS Med. 2012;9(12):e1001355. doi: 10.1371/journal.pmed.1001355. PubMed PMID: 23239945; PubMed Central PMCID: PMCPMC3519895. 113. UN. Nepal Millennium Development Goals, Progress Report 2013. In: Commision NP, editor. Kathmandu, Nepal2013. 114. Rao C, Adair T, Kinfu Y. Using historical vital statistics to predict the distribution of under-five mortality by cause. Clinical medicine & reserach. 2011;9(2). 115. MoHP. Multi-Sector Nutrition Plan, for Accelerating the Reduction of Maternal and Child Under-nutrition in Nepal 2013-2017 (2023). Kathmandu, Nepal: National Planning Commision, 2012. 67

116. Chi BH, Tih PM, Zanolini A, Stinson K, Ekouevi DK, Coetzee D, et al. Reconstructing the PMTCT Cascade Using Cross-sectional Household Survey Data: The PEARL Study. J Acquir Immune Defic Syndr. 2015;70(1). 117. Hancioglu A, Arnold F. Measuring coverage in MNCH: tracking progress in health for women and children using DHS and MICS household surveys. PLoS Med. 2013;10(5):e1001391. doi: 10.1371/journal.pmed.1001391. PubMed PMID: 23667333; PubMed Central PMCID: PMCPMC3646216. 118. Hartung CLA, Anokwa Y, Tseng C, Brunette W, Borriello G. Borriello G. Open Data Kit: Tools to Build Information Services for Developing Regions. ICTD 20102010. 119. SurveyCTO. [cited 2015 September 8]. Available from: http://www.surveycto.com/index.html. 120. MoHP. Estimated Target Population Fiscal Year 2071/72 (2014-2015). Kathmandu, Nepal: Health Management Information Section, Department of Health Services, Ministry of Health and Population, 2014. 121. Rutstein SO. Trends in Birth Spacing. Calverton, Maryland, US: ICF Macro, 2011. 122. Chatfield A, Javetski G, Lesh N. CommCare Evidence Base. Dimagi, Inc., 2013. 123. SimPrints. Mobile Biometrics for Development 2016. Available from: http://www.simprints.com. 124. Engmann C, Jehan I, Ditekemena J, Garces A, Phiri M, Mazariegos M, et al. An alternative strategy for perinatal verbal autopsy coding: single versus multiple coders. Trop Med Int Health. 2011;16(1):18-29. doi: 10.1111/j.1365-3156.2010.02679.x. PubMed PMID: 21371206; PubMed Central PMCID: PMCPMC3927400. 125. Lawn JE, Osrin D, Adler A, Cousens S. Four million neonatal deaths: counting and attribution of cause of death. Paediatr Perinat Epidemiol. 2008;22(5):410-6. doi: 10.1111/j.1365-3016.2008.00960.x. PubMed PMID: 18782248; PubMed Central PMCID: PMCPMC3428888. 126. Murray CJ, Lopez AD, Black R, Ahuja R, Ali SM, Baqui A, et al. Population Health Metrics Research Consortium gold standard verbal autopsy validation study: design, implementation, and development of analysis datasets. Popul Health Metr. 68

2011;9:27. doi: 10.1186/1478-7954-9-27. PubMed PMID: 21816095; PubMed Central PMCID: PMCPMC3160920. 69

Tables and figures Table 1. Signal functions of Basic Emergency Obstetric Care facilities and Comprehensive Emergency Obstetric Care Facilities, per Monitoring emergency obstetric care: a handbook. 2009, World Health Organization, Geneva, Switzerland. Abbreviations: BEmOC, Basic Emergency Obstetric Care; CEmOC, Comprehensive Emergency Obstetric Care. Table 2. Sample characteristics and demographics Sample Demographics Pre-expansion Group (2012) Post-expansion Group (2014) P- values i,j Total (n) 77 133 - Age, median (IQR) i 25 (21-28) 22 (20-26) 0.1 Distance (hours), median (IQR) a,i 2 (1-2) 2 (1-2) 0.37 Income, median (IQR) b,i 1000 (0-5000) 5000 (3000-7000) <0.01 Ropani, median (IQR) c,i 5 (2-12) 5 (2-7) 0.01 Upper caste, n (%) d,j 41 (53%) 80 (60%) 0.36 Some literacy, n (%) e,j 43 (55%) 114 (86%) <0.01 Multiparity, n (%) f,j 57 (74%) 83 (62%) 0.08 ANC visits Adequate, n (%) g,j 53 (69%) 115 (86%) <0.01 Autonomy, n (%) h,j 25 (32%) 65 (49%) 0.02 a. Distance is defined as the number of hours required to travel from the respondent s home to the hospital using the fastest mode of transport available to the respondent b. Income measured in Nepali Rupees (NRs); regression analysis was done per 1,000 NRs. c. Ropani is a local measure of farming land. d. Upper caste is any non-dalit (untouchable) caste. e. Some literacy is defined as either completion of elementary schooling or any self-reported ability to read in Nepali or English. 70

f. Multiparity is the number of respondents who had more than one previous birth. g. Adequacy of ANC visits is defined in accordance with the Nepali government s minimum of four visits. h. Women who reported themselves as either the primary or the joint decision-maker were coded as Autonomous compared to women who reported their husbands, fathers or mothers-in-law as the primary decision-makers. i. P-values for non-parametric continuous variables were calculated using Wilcoxon rank sum test. j. P-values for categorical variables were calculated using Fisher s exact test. 71

Table 3. Factors of institutional births, compared across time and between birth location in each time period Institutional Birth Factors Home (n=54) 2012, n (% of yes respondents) Facility (n=23) P- Home values a (n=31) 2014, n (% of yes respondents) Facility (n=102) P- 2012 values a (n=77) Total, n (% all respondents) 2014 (n=133) P- values a Hospital is safer 47 (61.0) 22 (28.6) 0.42 27 (20.9) 102 (79.1) <0.01 69 (89.6) 129 (97.0) 0.03 Priority on safety 6 (23.1) 20 (76.9) <0.01 18 (22.0) 64 (78.0) 0.68 26 (33.8) 82 (61.7) 0.01 Priority on cost 2 (100) 0 (0) >0.99 5 (12.5) 35 (87.5) 0.07 2 (2.6) 40 (30.1) <0.01 Priority on distance 32 (91.4) 3 (8.6) <0.01 9 (29.0) 24 (72.7) 0.63 35 (45.5) 33 (24.8) <0.01 Priority on CEmOC 1 (100) 0 (0) >0.99 4 (8.9) 41 (91.1) <0.01 1 (1.3) 45 (33.8) <0.01 Knowledge 10 of CEmOC 3 (23.1) >0.99 3 (3.2) (76.9) availability a. P-values are calculated using Fisher s exact tests. 90 (96.8) <0.01 13 (20.3) 93 (86.9) <0.01 Table 4: Institutional Birth Study: Results of a logistic regression model for institutional birth. Regression Term Estimate Std Error Odds Ratio 95% CI P- value CEmOC availability 0.86 0.24 5.6 2.2-15 0.01 Income (per 1000 NRs) 0.07 0.03 1.1 1.0-1.1 0.01 Hospital Safety 1.9 0.69 45 4.8-1300 <0.01 Safety priority 1.02 0.24 7.7 3.2-21 <0.01 CEmOC availability x Safety priority a -1.06 0.24 0.1 <0.01 Safety priority pre-intervention 1.02 7.7 Safety priority post-intervention -0.04 0.9 a. Interaction term represents the effect of reporting safety as a factor on the likelihood of institutional birth in each time period. 72

Table 5. Institutional Birth Study: Examples of women s birth stories pre- and postexpansion, as translated from Nepali transcriptions. Pre-expansion (2012) Post-expansion (2014) Home All day I worked on the farm. At 7pm, labor pain started. At 12am, female baby was born at home. -- I didn't know about the hospital and I don't have anyone who could carry me to the hospital. -- I planned to go to the hospital to give birth. I knew about the 1000 Rs [government incentive]. My home condition is very bad and I have no support for people to bring me to the hospital. I had a long course of labor pain, but couldn't find anyone to carry me to the hospital, so I delivered at home. I had 4 ANC checks - one at hospital and rest 3 at the HP. I had planned to deliver at the HP, but my labor started suddenly and by the time people had gathered to take me to the HP, I had already delivered. I am planning to deliver my next baby at the hospital though. -- I had planned to deliver at the hospital and I was on my way as well. But I delivered mid-way. There wasn't any safe birth kit, no clean cloths. So, it was very difficult. Facility I started having labor pain and since the hospital is nearby I walked to the hospital and had my baby safely delivered. -- A mother of two wanted to give birth in the hospital because they were close by and because the mother thought it would be safer. Her previous children were born at home. Belly pain started late into the night and the family called the ambulance. Unfortunately, the ambulance was not working so it was not able to pick her up. The family got together and found a stretcher. She was carried on a stretcher during the night, one hour away from the hospital. She delivered in the early morning at Bayalpata Hospital. [as translated] I wanted to go to BH to deliver, but there was nobody to help me to the hospital [so I went to the health post instead]. My husband is in India and there is just an old mother-in-law at home. But I did complete all 4 ANCs, took my iron tabs regularly and also the Immunization. -- I had delivered my last 2 babies at home, and there were no problems. But third time there was spontaneous abortion. This time, the water broke early, so, I came to the hospital. I got the services that I had desired here and delivered a healthy baby. I am very happy. -- I had thought that if I can't deliver normally then I would deliver via operation. But I could deliver normally, so I am very happy. I also got very good service in the hospital. Because I didn't have enough money I couldn't afford to travel to the hospital. But here I was given [government incentive] money for return travel. 73

Table 6. Consolidated Framework for Implementation Research (CFIR) domains, constructs, with definitions of key constructs in this analysis. As defined by Damschroder, et al. in Fostering implementation of health services research findings into practice: A consolidated framework for advancing implementation science. Implement Sci. 2009;4:50. doi:10.1186/1748-5908-4-50. Domains Constructs Definitions Intervention Characteristics Inner Setting Intervention Source Evidence Strength & Quality Relative advantage Adaptability Complexity Structural characteristics Networks & Communications Culture Implementation Climate External Policy & Incentives Perception of key stakeholders about whether the intervention is externally or internally developed. Stakeholder s perceptions of the quality and validity of evidence supporting the belief that the intervention will have desired outcomes. Stakeholders perception of the advantage of implementing the intervention versus an alternative solution. The degree to which an intervention can be adapted, tailored, refined, or reinvented to meet local needs. Perceived difficulty of implementation, reflected by duration, scope, radicalness, disruptiveness, centrality, and intricacy and number of steps required to implement. The social architecture, age, maturity, and size of an organization. The nature and quality of webs of social networks and the nature and quality of formal and informal communications within an organization. Norms, values and basic assumptions of a given organization. The absorptive capacity for change, shared receptivity of involved individuals to an intervention and the extent to which use of that intervention will be rewarded, supported, and expected within their organization. A broad construct that includes external strategies to spread interventions including policy and regulations (governmental or other central entity), external mandates, recommendations and guidelines, pay-for-performance, collaboratives, and public or benchmark reporting. 74

Table 7: Cohort demographics for Group Antenatal Care Study participants. Demographics were obtained at enrollment in group antenatal care or home visits. Standard ANC + home visits Group ANC + home visits Total respondents 51 56 - Lower caste (Dalit/Miya) 21 (41%) 24 (43%) 1 P- value Illiterate a 6 (12%) 17 (30%) 0.03 Household characteristics b, median (IQR) Household size 5 (3-6) 6 (4-8) 0.02 Land units owned 4 (2-8) 5 (3-10) 0.3 Months of subsistence 6 (3-12) 7 (3-12) 0.72 Monthly expenses (NRs) 7,000 (4,000-11,000) 5,500 (3,000-10,000) 0.49 Location of Prior Birth Facility 16 (43%) 24 (60%) 0.17 a. Illiteracy is defined as no reported informal or formal schooling. b. Household socioeconomic characteristics have four components, which we are currently developing into an index. Household size is the number of people eating out of the household s kitchen at least 5 days a week. Land units owned is measured in local units of farming land (kattha), similar to one acre. Months of subsistence is defined as the number of months per year a family can eat what they grow (on their land and/or rented land) and is a measure of food security. Monthly expenses is defined as the reported amount of nepali rupees spent during the month on household necessities and household goods (a notably imperfect measure of cash income used due to consistent under-reporting of income). 75

Table 8: Preliminary outcomes data for Group Antenatal Care Study cohorts. Primary outcomes were assessed through self-report after delivery. Knowledge changes were assessed at enrollment in group antenatal care or household visits and 1-2 months after delivery and a t-test calculated on the score differences between intervention and cohort groups. Primary Outcomes across Treatment Groups Standard ANC + home visits Group ANC + home visits P-value ANC Visits, median (IQR) 4 (4) 4 (4) 0.59 Institutional Birth, n (%) 42 (82%) 51 (91%) 0.25 Secondary Outcomes across Treatment Groups Quality Assessments, n (%) ANC visits were 'very useful' 48 (94%) 51 (91%) 0.72 ANC providers gave 'excellent care' 46 (90%) 52 (93%) 0.73 ANC visits were 'very enjoyable' 30 (59%) 47 (84%) <0.01 Knowledge Changes, mean score diff (CI) Birth Preparedness 0 (-0.4-0.5) -0.1 (-0.5-0.3) 0.61 Antenatal Risks 1.5 (1.0-2.0) 1.8 (1.0-2.6) 0.61 Labor Risks 1.0 (0.6-1.4) 0.9 (0.5-1.4) 0.77 Postpartum Risks 0.9 (0.4-1.4) 0.9 (0.4-1.3) 0.9 Newborn Risks 1.4 (1.0-1.8) 1.3 (0.8-1.9) 0.79 Achieved Planned Delivery Location, n (%) 45 (88%) 51 (91%) 0.76 76

Table 9: Results of pregnancy screening pilot protocol by pregnancy risk category Not at risk a Known pregnancy b Unexplained amenorrhea c Post-pregnancy amenorrhea d Positive, n (%) - 12 (86%) 2 (14%) 0 Negative, n (%) - 0 9 (90%) 1 (10%) Not Tested, n (%) 172 (93%) 2 (1%) 10 (5%) 2 (1%) a. Not at risk is defined as either regular menses (within last 6 weeks) or use of longacting contraception that typically causes amenorrhea (e.g. depo provera or progesterone implants). b. Known pregnancy is defined as a positive response to Are you currently pregnant? c. Unexplained amenorrhea is defined as amenorrhea lasting greater than 6 weeks, with no identifiable cause (e.g. contraception or recent pregnancy within the past 3 months). d. Post-pregnancy amenorrhea is defined as no return to menses after more than 3 months postpartum or post-abortion. Note, this does not account for possible lactational amenorrhea, as exclusive breastfeeding is not typical in this setting. Table 10: Results of pregnancy screening pilot protocol by (suspected) gestational age. Bolded results indicate potentially missed pregnancies. Known Pregnancy "At-Risk" Unexplained amenorrhea Postpregnancy (Suspected) Gestational age 1-3 months 4-6 months 7+ months 1-3 months 4-6 months 7+ months unknown a Positive, n (%) - 5 (36%) 7 (50%) - 2 (14%) - - Negative, n (%) - - - 3 (30%) 2 (20%) 4 (40%) 1 (10%) Not Tested, n (%) - - 2 (1%) 3 (2%) - 7 (4%) 2 (1%) a. Suspected gestational age was not calculated for women who had no return to menses after delivery or abortion, but all women were greater than 3 months post-pregnancy as per the protocol s definition (see Table 9). 77

Figure 1. Distribution of facility and home births before and after roll-out of Comprehensive Emergency Obstetric Care at Bayalpata Hospital. Abbreviations: CEmOC, Comprehensive Emergency Obstetric Care. Figure 2. A social contextual theory of change based on qualitative analysis of women's birth stories. The diagram shows the interplay between modifying and mediating factors, socio-demographic factors, and the intervention. Abbreviations: BH, Bayalpata Hospital; ANC, antenatal care. 78

Figure 3: Possible s for-impact culture code: ten key identity statements framing Possible s organizational mission and culture, as defined by organizational leadership. Reprinted with permission from Possible. Accessible at http://possiblehealth.org/wp-content/uploads/2014/02/possible-culture-code1.pdf. Figure 4. Virtuous cycles in global health delivery, in which science is depicted as the grease for the enginge of effective service delivery, which drives implementation and scale-up of successful interventions. Reprinted with permission from Maru, D. and Famer, P. Human Rights and Health Systems Development: Confronting the Politics of Exclusion and the Economics of Inequality. Health and Human Rights. December 2012, 2:12. 79

Figure 5: Core components of the CenteringPregnancy model, as defined by Rising, et al. in Redesigning Prenatal Care Through CenteringPregnancy, Journal of Midwifery and Women s Health, 2004, 49: 5. Figure 6: Heatmap of under-two deaths reported during a cross-sectional censusbased survey in 14 village clusters of Achham, Nepal, using a truncated two-year birth history. 80

Figure 7: Comparison of cross-sectional, retrospective data collection methods versus continuous surveillance methods. Continuous surveillance offers increased spatio-temporal precision for real-time data use in quality improvement and research. 81