Gestational Age – Rallies 1, 5, and 9
Accurate estimation of gestational age is important at both the individual level and the population level (World Health Organization, 2015). At the individual level, gestational age informs clinical management during pregnancy and at birth. At the population level, estimates of gestational age are used for global surveillance of preterm birth and epidemiological research (Blencowe et al., 2012; Howson et al., 2013; Perumal et al., 2018).
Each new study contributes to fine-tuning the precision and accuracy of gestational age estimates. When access to ultrasound early in pregnancy is unavailable, using ultrasound later in pregnancy or simpler metrics at birth has the potential to improve screening of preterm births. Improving gestational age estimation will also contribute to better quality of clinical study data and clinical decision making.
1. Improve estimation of gestational age to within two weeks at the individual level using ultrasound(s) and/or non-ultrasound measurements.
2. Improve estimation of the distribution of gestational age at birth using methods other than ultrasound, as a way to obviate the limited access to ultrasound and early antenatal care in some populations.
COUNTRY AND DATASET UTILIZED
Dataset comprises 6 studies representing over 2,500 individuals from at least 15 countries. Data includes ultrasound and non-ultrasound measurements.
Gestational age, preterm birth rate, newborn, lens vascularity, anthropometric, ultrasound, pregnancy
Stunting and Wasting – Rallies, 4, 6, and 7
Child malnutrition kills about 3 million children every year, with 45% of all childhood deaths attributable in part to malnutrition.*1 Many more children suffer long-term health consequences from impaired physical growth. Wasting, or a low weight for a given height, is a measure of acute malnutrition. Wasting is a factor in 1 million child deaths every year.*1,2 Prevalence can vary dramatically over time due to seasonal food insecurity or disease outbreaks. Similarly, stunting, or a low height for a given age, is a measure of chronic malnutrition. Stunting is a major problem in low- and middle-income countries (LMICs); it is associated with a range of poor outcomes later in life, including poor cognitive development.*3 Most studies of wasting use cross-sectional data, and so cannot provide information on the longitudinal patterns in episode incidence, duration, and recovery. Cohorts with frequent measurements are generally small and therefore can’t examine regional or age-specific patterns in wasting. Examining the timing of wasting incidence using longitudinal measurements will help inform interventions to prevent the onset of wasting. Many children in low-resource settings become stunted early, often in the first 2 years of life, because they fail to grow as fast as well nourished infants. Understanding the underlying epidemiology of stunting, its primary risk factors, and the effect modifiers of existing interventions will help inform future intervention strategies.
1. Combine standardized individual-level data from multiple datasets to summarize broad longitudinal and regional patterns in wasting incidence, duration, recovery, and severity.
2. Examine longitudinal and regional patterns in stunting incidence, growth velocity, and catch-up growth.
3. Identify and rank-order the child, parental, and household characteristics associated with wasting and stunting incidence.
COUNTRY AND DATASET UTILIZED
30+ studies in LMICs measuring height of 41 cohorts & ~86,000 children multiple times in the first 24 months
35 studies in LMICs measuring weight and height of 43 cohorts & ~123,000 children multiple times in the first 24 months
Stunting, wasting incidence, wasting recovery, WHZ, LAZ, breastfeeding, low birth weight, undernutrition
Pregnancy Risk Stratification – Rallies 8 and 19
Maternal and neonatal death has declined more slowly than under-5 mortality as a whole. (Liu et al., 2016) A few key conditions drive maternal mortality, stillbirths, and neonatal mortality in low-and-middle-income countries (LMICs), especially pre-eclampsia.
Improving the identification of women who are at higher risk for poor pregnancy outcomes is important to determine the right level of care. Currently, it is difficult to properly identify high-risk pregnancies in low resource settings because there may be a lack of tools for early detection, a lack of clinical specialists, or other logistical barriers.
The goal of these data science rallies was to inform the development of an algorithm that can be used by minimally trained users in LMICs and enable the prediction and prevention, or mitigation, of adverse birth outcomes.
- Identify existing tools, studies, needs, and opportunities related to PRS in LMIC settings.
- Predict the risk of pre-eclampsia throughout pregnancy using longitudinal data.
- Predict the risk of neonatal death or stillbirth and determine the most important contributing factors.
COUNTRY AND DATASET UTILIZED
6 studies from a variety of countries measuring clinical and socioeconomic variables on 25 cohorts & ~137,000 mothers and children multiple times during pregnancy and the first month of life.
Pregnancy, Preeclampsia, Neonatal Death, Stillbirth, Machine Learning, Longitudinal Modeling
CONCLUSION AND RELEVANCE
The results of the first model provide an initial prediction of a woman’s risk of pre-eclampsia. The change in risk over time is reflected in the prediction as more data is collected during pregnancy. We tested several different models. The final model was a univariate longitudinal model for diastolic blood pressure and a time-to-event model.
To model the risk of neonatal death or stillbirth, we used a variety of machine learning algorithms. We compared the results across models and studies to determine risk of neonatal death and stillbirth and the relative importance of predictors. In all cases the area under the ROC curve was approximately 0.7. Global models based on data from all sites generally outperformed site-specific models.
The results from these rallies provide clinical algorithms that combine clinical history, physical examinations, ultrasound, and biomarkers to effectively identify risk factors for key pregnancy adverse events. These predictions could help facilitate the appropriate matching of risk level with clinical care, alongside improved quality of referral and intervention availability. Another important outcome from these rallies is an increasing awareness of the need to collect data longitudinally during pregnancy, and at the optimal timepoints to help facilitate clinical decision-making.
- Liu, L., Oza, S., Hogan, D., Chu, Y., Perin, J., Zhu, J., Black, R. E., et al. (2016). Global, regional, and national causes of under-5 mortality in 2000–15: An updated systematic analysis with implications for the Sustainable Development Goals. The Lancet, 388(10063), 3027–3035. https://doi.org/10.1016/S0140-6736(16)31593-8
CHAMPS Rally Summary – Rallies 12 & 20
Policy makers and other stakeholders need high-quality data on the cause of death for stillbirths and children under five, so they can target interventions toward the leading causes of death in their regions. The Child Health and Mortality Prevention Surveillance (CHAMPS) Network provides robust, standardized, longitudinal mortality data on stillbirths and deaths in children under age 5 in Sub-Saharan Africa and South Asia from many sources. (Salzberg et al., 2019)
CHAMPS is committed to making its data accessible to the scientific, clinical, and public health communities through their website. However, a lack of tools makes getting started with CHAMPS data more daunting for new researchers. The goal of these data science rallies was to develop programmatic tools that researchers need to build reports, compute statistics, and complete custom data analyses from CHAMPS data.
These programmatic tools are owned and made available by the CHAMPS Network at https://champshealth.org.
- Provide tools that help researchers generate reports from the data reliably and efficiently.
- Provide tools that make it easier for new researchers to get acquainted with the data.
COUNTRY AND DATASET UTILIZED
Child Health and Mortality Prevention Surveillance (CHAMPS) Level 2 de-identified data from sites in Bangladesh, Ethiopia, Kenya, Mali, Mozambique, Sierra Leone, South Africa
Child mortality, surveillance, CHAMPS
RESULTS AND RELEVANCE
The primary output of these rallies is an R package with a companion website containing seven articles that guide the user through topics including how to access the data, getting acquainted with the data, creating standard reports, and doing custom analyses. In addition, other resources in the package help analysts learn important nuances in the data. The package provides utilities to read and transform the data into convenient formats, functions to compute several statistics of interest, and some utilities for presenting these statistics in various formats such as plots and HTML tables. It also provides detailed explanations of how to compute statistics of interest and complete custom data analyses.
CHAMPS partners with governments and ministries of health to integrate insights from CHAMPS data into their evidence-based decision-making processes. A major objective for CHAMPS is that the data and its corresponding insights have as broad a reach as possible. The multiple types of data collected in CHAMPS and the often complex way in which it is combined to provide insights can be daunting for new researchers. These data science rallies have produced tools and resources that are being disseminated to help make CHAMPS data more widely accessible.
- Salzberg, N. T., Sivalogan, K., Bassat, Q., Taylor, A. W., Adedini, S., El Arifeen, S., Assefa, N., Blau,M., Chawana, R., Cain, C. J., Cain, K. P., Caneer, J. P., Garel, M., Gurley, E. S., Kaiser, R., Kotloff, K. L., Mandomando, I., Morris, T., Nyamthimba Onyango, P., … Child Health and Mortality Prevention Surveillance (CHAMPS) Methods Consortium. (2019). Mortality Surveillance Methods to Identify and Characterize Deaths in Child Health and Mortality Prevention Surveillance Network Sites. Clinical Infectious Diseases, 69(Suppl 4), S262–S273. https://doi.org/10.1093/cid/ciz599