Data Science Approaches to Improve Maternal and Child Health in Africa

Preamble

This challenge is being launched by Grand Challenges Africa (GC Africa) - a program of the African Academy of Sciences (AAS) supported through the AAS funding and programs implementation platform, the Alliance for Accelerating Excellence in Science in Africa (AESA) - and the South African Medical Research Council (SAMRC), in collaboration with the Bill & Melinda Gates Foundation’s Knowledge Integration (Ki) initiative. Ki seeks to develop a deeper understanding of the risk factors contributing to poor maternal, neonatal, and child health (MNCH) outcomes and how best to address them through new data analysis tools and techniques. GC Africa has also collaborated with the UK Academy of Medical Sciences in hosting agenda-setting workshops for MNH priorities in Africa that have helped to shape this call for applications.

The Challenge

There are key knowledge gaps in our understanding of how nutrition, prenatal and antenatal care, maternal support, and environmental and social factors contribute to an elevated risk of poor maternal and childhood health outcomes – and the degree to which early life health outcomes can have lifelong, or even intergenerational consequences for health and productivity. Such an understanding is required to determine what interventions, including health policies, should be delivered to which specific groups of individuals and at what point in their lifecycle to ensure optimal outcomes through the most effective use of healthcare resources.

The Opportunity

Globally, there is an ongoing burden of 5.4 million deaths annually, including neonatal deaths (2.5 million), stillbirths (2.6 million), and maternal deaths (0.3 million). Africa, with only 17% of the world's population, carries nearly half of this burden with 2.3 million deaths per year. Based on current trends, most sub-Saharan African countries won't meet the United Nations Sustainable Development Goals' target of 12 or fewer neonatal deaths per 1,000 live births and are also at risk of missing targets for maternal mortality reduction.

Key areas that have been identified by which to improve maternal and child health in Africa are: 1) better care during pregnancy, 2) better care at birth, 3) better postnatal care for women and their newborns, and 4) better hospital care of sick newborns (see report from the September 2018 policy workshop organized by the African Academy of Sciences (AAS) and the UK Academy of Medical Sciences). Nevertheless, developing and validating approaches to foster maternal and child health is challenging due to the complex interaction of biological, environmental, and social factors. Furthermore, policy recommendations for such approaches frequently lack sufficient supporting scientific evidence, while clinical trials are expensive, time-consuming, and increasingly difficult to implement. There is now a key opportunity to accelerate research in this area by establishing robust collaborative research networks in Africa (especially those focused on pregnancy and the peripartum period) and analyzing data from multiple African sources to guide public health recommendations that are data-driven and cost effective.

The purpose of this call for proposals is to promote new quantitative approaches to synthesizing and analyzing data and evidence obtained from African public health surveys, longitudinal observational studies, clinical trials, and other relevant data sources – including combined datasets – to produce novel insights which can be used to improve maternal and child health in the African context and around the world. By enabling and fostering access to diverse datasets, we intend to engage a broad spectrum of collaborators – including research and clinical scientists working with data scientists, bioinformaticians, statisticians, epidemiologists, engineers, and computer programmers - to identify how innovative data analytic approaches can be used to develop more effective solutions for the maternal and child health challenges in Africa.

Successful applicants should assume that with the appropriate agreements in place, they will receive not only funding, but also access to anonymized subsets of data that have been obtained through existing partnerships under the Gates Foundation's Ki initiative (see Appendix for more information about data sources and access principles). We also welcome applicants who have access to other relevant data sets, including publicly available data, clinical research, and cohort and survey studies. Note that such investigators are responsible for securing appropriate authorization to use this data, and we encourage them to describe in their proposals the steps required for such access. We encourage applicants who propose using data that is not publicly available to describe partnerships that could contribute data to the Ki database to become a shared resource, since combining data is likely to reveal insights that would not otherwise be obtained. Note that the Ki team will help funded applicants combine different datasets from the Ki repository to prepare "analysis datasets" that are tailored to the proposed area of research.

What we are looking for

We seek proposals designed to answer critical scientific questions related to maternal, neonatal and child health and development outcomes. Proposals should use innovative quantitative analytics and modeling approaches that can be applied to the relevant existing data sets and should yield actionable results with a potential to significantly impact pan-African public health policy or that of specific African countries.

We will give highest priority to proposals that:

• support innovative collaborations between African researchers, healthcare professionals, and data scientists that could contribute to a sustainable African maternal, neonatal, and child health research network
• answer critical scientific questions identified in this call for proposals, while building and strengthening data science capacity for Africa
• consider African genetic diversity by accessing genomics databases, gene sequences, genome-wide association studies, or genetic cohorts that include Africans such as HapMap
• take into account social, environmental and cultural determinants of outcomes and incorporate an understanding of the target community that includes barriers and constraints to delivery of interventions and to implementation of public health programs
• contribute to a portfolio of funded projects that addresses regional diversity and the need to provide health equity for diverse vulnerable populations
• explain how answers will have the highest likelihood of being relevant for implementation broadly in the public health system
• employ innovative and interdisciplinary analytical methods

Examples of what we are looking for include analytical approaches that:

• apply innovative and highly efficient analyses - such as machine learning, Bayesian hierarchical techniques, or extreme value modeling - to identify patterns and forecast from real world data and evidence
• describe mechanistic models for establishing the relationship between interventions and their related outcomes
• end with derived, manipulated, and results data in a format that is findable, accessible, interoperable and reusable (i.e., FAIR principles), enabling further study and making projects more impactful
• make use of or convert data to data structures and protocols found in free and open source, global standard software initiatives such as District Health Information System 2 (DHIS2), Open Mobile Health (Open mHealth), or Observational Health Data Science & Informatics (OHDSI) Common Data Model
• stratify risk of adverse pregnancy outcomes, including preterm birth, stillbirth, and low birth weight, including algorithms based on serial ultrasound imaging
• identify particularly vulnerable subpopulations where geographical, cultural, or other context contributes to health inequity that could be remediated by targeted approaches
• result in implications for drug dosing of existing or new drugs (e.g., for malaria, HIV, and TB)
• incorporate weight gain during pregnancy as a variable, including helping to determine the relative contributions to neonatal health outcomes of maternal diet quantity versus quality
• determine the relative contributions to infant health outcomes of diet quantity versus quality (e.g., protein quantity versus quality)
• target underexplored subsets of data (e.g., rare but important maternal and child health events) that can be studied because of the large size and statistical power of the database, including birth cohort studies
• help convert correlations in data into strong inferences that influence policy (e.g., health outcomes correlated to sex differences, maternal education, birth spacing, or age of first pregnancy, or establishing the causal impact of air pollution on fetal growth)
• build on a "positive deviance" perspective to identify actionable lessons from existing data demonstrating positive maternal and child health outcomes in population subgroups despite a high level of challenges (e.g., low level of resources or high burden of infection)
• identify new ways to aggregate risk factors and identify vulnerable populations for adverse maternal and child health outcomes, including innovative data integration strategies and visualization tools
• specifically incorporate the roles of women – as perceived locally – from adolescence to motherhood (including pregnancy during adolescence)
• specifically incorporate the roles of men – as perceived locally – from adolescence to fatherhood and their influence on maternal and child health generally
• determine the contribution of violence against women and girls, including domestic violence, to the burden of adverse maternal and child health outcomes
• study linkage to care during early pregnancy and retention in care throughout pregnancy and outcomes in the early post-natal period
• evaluate programs for pre-pregnancy intervention for women and the effect of doing so on prenatal, maternal, fetal, and neonatal mortality
• determine the best care for low-birth-weight babies
• help determine the window of opportunity to foster catch-up growth for preterm and low-birth-weight babies, and the most effective interventions for doing so
• help identify critical periods for intervention during pregnancy and early childhood
• stratify the risk of stunting and wasting from birth through two years of age, including identifying data sets where a high risk despite high quality of care suggests unknown variables
• answer specific questions about the window of opportunity to restore healthy child development, including whether there are developmental stages after which impaired neurodevelopment become irreversible if appropriate interventions are not instituted
• target direct and indirect root causes of adverse birth outcomes, including those leading to or necessitating caesarean section, and address the most vulnerable population groups considering age and ethnicity
• investigate the role of co-morbidities in pregnancy and childhood outcomes
• stratify risks for child development aiming to establish national indicators for healthy development from the neonatal period to the child's first two years, preferably addressing the most vulnerable population groups considering age and ethnicity
• help to understand the relationship between social indicators, nutritional conditions, and mortality from the prenatal period to early childhood
• help to understand maternal and child undernutrition as an underlying vulnerability that impacts maternal, neonatal, and child health, survival, and resilience
• incorporate data from novel and effective labor monitoring approaches
• data with plausible link to informing new solutions (e.g., aetiological data for maternal, newborn, stillbirth infections and/or NCD exp)
• reanalyzes to inform new ways to measure coverage and quality of care especially quality of care at birth or for care of small and sick newborns, e.g., validation of core indicators etc. assessing or informing ways to improve large scale data platforms to belter measure MNH outcomes/coverage with particular priority given to DHIS2 and large-scale household surveys such as DHS/MICS or large-scale Health Facility Assessments

Examples of what we are NOT looking for:

• Proposals submitted by applicants from institutions outside of Africa
• Proposals that do not focus on health outcomes in Africa
• Proposals for new studies to generate new data
• Development of new primary data collection tools
• Proposals not related to maternal and child health challenges
• Approaches that do not meaningfully involve data from adolescents, mothers, or infants
• Applications that do not adequately describe how the information in the datasets that will be used has the potential to answer the research question
• Applications proposing data science algorithm development without clear relevance to answering the types of questions described in this call for proposals
• Applications that do not adequately describe the methodology and explain why it is novel
• Ideas without a clearly articulated and testable hypothesis together with metrics of success
• Ideas for which the described indicator of success cannot be demonstrated or significantly advanced within the scope of the grant award (USD $100,000 over 24 months)
• Proposals that do not describe the innovation’s potential effects on health policy making
• Analyses that are only slight improvements over existing approaches (e.g., replication of an approach in a new geography in the absence of added innovation)

Overview of Data Sources Available for This Call

A wide range of data types is required to develop data-driven solutions for guiding cost-effective health policy and programs. These data are typically captured through various independent clinical research trials, health delivery programs, and various surveys, but are stored across different departments, organizations, or at various levels of the health system. Child health and development programs need to bring these data together – and frequently in a harmonized and integrated form – to ensure research and analysis is fully informed by the best available data. These data include already existing primary data in Africa from varied sources, such as those from one or both of the following data repositories:

• Africa-Ki data repository, available for this call for proposals: The Foundation's Ki team and its partners have established this MNCH-focused repository of harmonized, anonymized longitudinal and cross-sectional clinical study data for secondary data analysis. The Africa studies in the Ki repository may be suitable to study and analysis of some of the topics listed in this call. can be used for the development of algorithms and for answering critical questions. Note that the Ki team will help funded applicants combine different datasets from the Ki repository to prepare "analysis datasets" that are tailored to the proposed area of research. To explore the metadata for the Africa-Ki repository, please see https://www.synapse.org/Ki_GCAfrica_MNCH_DataCall
• Other data sources that applicants may propose: This includes existing data repositories, including publicly available population-level surveys (i.e., DHS and UNICEF) and cross-sectional and longitudinal cohort studies, program-level data on disease burden from government agencies, surveillance system data on disease burden, administrative records of health-service encounters, disease registries, and other relevant studies in Africa. These studies include published literature as well as unpublished studies that can be identified and accessed by collaborating with healthcare experts and other stakeholders in Africa. Note that investigators are responsible for securing appropriate authorization to use these other data sources, and we encourage them to describe in their proposals the steps required for such access. GC AFRICA planning partners will work with selected grantees and the access committees of existing data repositories that they have listed in their application to help secure timely data access. For applicants who propose using clinical study data that is not publicly available, we encourage them to describe partnerships that could be established to contribute data to the Ki database to become a shared resource, since combining data sets is likely to reveal insights that would not otherwise be obtained. Examples of potential data sources include:

• DHS Program: https://dhsprogram.com/data/
• H3AfricaBionet: https://h3abionet.org/
• WWARN: https://www.wwarn.org/
• Datasets through Institut Pasteur sites in Africa
• Guttmacher Institute: https://www.guttmacher.org/population-center/datasets
• AHRI/CAPRISA/UKZN
• Rwanda Biomedical Centre (RBC)
• Digital health provision databases that have MNCH parameters
• InDepth Network
• South Africa Provincial Health Data
• KEMRI-WT Programme

GC Africa Maternal, Newborn & Child Health (MNCH) Data Challenge Community

The Gates Foundation's Ki team has established a website to share additional information that may be helpful to applicants as they prepare their proposals. The following link provides access to the website: https://www.synapse.org/Ki_GCAfrica_MNCH_DataCall

Below are two primary ways that this website of additional information may be helpful to applicants. The Ki team will be adding details and updates to this website throughout the open call period, so please check the site regularly to look for additional information.

1. This website can be explored by applicants to learn more about available data sources and to form useful collaborations (for the GC Africa application) with people from a variety of areas of expertise - including research and clinical scientists working with data scientists, bioinformaticians, statisticians, epidemiologists, engineers and computer programmers. The website includes a discussion forum where interested applicants can communicate to form innovative teams to apply for this call
2. This website includes a primer on MNCH for data scientists, a primer on data science for research scientists, a description of the Africa data sets available in the Ki data repository and a discussion forum. The HBGDki team will be adding additional information including sample datasets for download and experimentation, so please check the site regularly to look for additional information.