These are initial findings of rapid ongoing research that haven’t been finalized or validated. Their purpose is to share results of our ongoing work.
Sprint 1A: Accuracy and Precision in Gestational Age Measurement: Physical Assessment Tools
Improve the postnatal estimation of gestational age in order to determine preterm birth rates and small for gestational age rates with the goal of 90% to be ~ 2 weeks across the gestational age spectrum to 28 weeks out to 42 weeks. Evaluate across sites and by sex.
A model using SuperLearner was constructed to better predict gestational age, with the precision of the model nearing ~ 2 weeks precision.
Sprint 1B: Accuracy and Precision in Gestational Age Measurement: Physical Assessment Tools
What combination(s) of physical assessments can accurately predict gestational age, especially in SGA and LGA children? Does that differ by geography?
The SuperLearner model was updated to weight age groups to mitigate bias in both preterm & SGA populations, improving the accuracy of the model.
Sprint 2A: Characterizing Fetal Growth and Relations Between Maternal Covariates, Fetal Growth, and Birth Outcomes
Determine fetal growth patterns and population differences using several data sets and determine if one growth standard can be applied to all geographies.
Nonparametric function estimation methods were used to examine several questions about fetal growth trajetories and birth outcomes, finding effects between maternal characteristics and birth outcomes.
Sprint 2B: Fetal Growth: One Standard Fits all?
Are small for gestational age (SGA) and large for gestational age (LGA) proportions different between countries? What are the results of constructing population fetal growth curves by combining all available data sets and comparing these curves with the INTERGROWTH standard?
Population growth curves were constructed by combining all available datasets and were compated with intergrowth standards.
Sprint 4A: Descriptive Epidemiology of Wasting
Describe the epidemiology of incident wasting (WHZ < −2) from birth to age 24 months.
A broad look at wasting across many populations revealed many interesting insights and led to additional addressed in the following sprints.
Sprint 4B: Pilot of the Analysis of Risk Factors for Wasting Incidence and Recovery
The focus of this ral;y will be to: 1. Refine the descriptive analysis report for each cohort. 2. Explore additional measures of wasting recovery, and the proportion of children wasted who recover within 30, 60, and 90 days. 3. Summarize prevalence, incidence, and duration in figures rather than tables. 4. Extend the descriptive analysis to all 22 cohorts that include a monthly measurement. 5. Begin the risk factor analyses to identify time-invariant characteristics associated with wasting episodes and recovery among children ages 0-6 months.
Created and a single, harmonized data analysis pipeline and then used it to summarize the descriptive wasting epidemiology across 22 contemporary cohorts.
Sprint 4C: Dose-Response Relationship of Breastfeeding on Wasting
Investigate the dose-response relationship of breastfeeding on wasting through exploratory visualization of the MAL-ED data, providing a basis for future modeling efforts. In this rally sprint, we seek to understand the effect of a 0-6 months breastfeeding dose on growth from 0-24 months and also how this effect varies across different groupings of wasted infants.
Exploratory analyses and visualizations uncovered some interesting insights about the relationship of breastfeeding and wasting.
Sprint 4D: Wasting Analysis Clean-Up and Documentation
In this rally, we will: 1. Clean up and document prior analyses. 2. Verify accuracy of incidence calculations against simulated data with a known incidence rate. 3. Combine prior analysis results with a more extensive literature review in order to make plans for further analyses. 4. Investigate context of all the studies that were combined in prior analyses and quality-assurance check each cohort’s anthropometry measurements. 5. Draft an analysis plan formalizing all preliminary analyses completed in the rally and finalizing analyses to be completed and publications to be written as part of the project. 6. Draft a timeline for project completion
Prepared and presented a webinar to GHAP data contributors and other interested parties.
Sprint 5A: Gestational Age Shift Analysis
The focus of this first two-week rally sprint: using US data, if we could shift the gestational age of 26-29 week-olds by 1, 2, 3, 4 weeks, what would be the impact on mortality?
Built and studied a model that investigates the impact on mortality if we could shift the gestational age of 26-29 week olds, by 1, 2, 3, 4 weeks, enabling, for example, a look at a hypothetical population of 140 million, where there is possibility of saving more than 150,000 lives by merely increasing the Gestational Age of those born less than 39 weeks by just one week.
Sprint 6A: Descriptive Epidemiology of Wasting and Stunting in India
Describe the India-specific epidemiology of incident wasting and stunting (WHZ and HAZ < −2) and severe wasting and severe stunting (WHZ and HAZ < −3) from birth to age 24 months.
Wasting methodology from rally 4 was applied to India-specific data, uncovering insights and providing a way to engage with PIs in India.
Sprint 7A: Epidemiology of Stunting Analysis Plan
Prepare an analysis plan to describe the epidemiology of stunting (HAZ < −2) from birth to age 24 months.
An analysis plan for the rally was developed.
Sprint 7B: Stunting Objective 1a. Age-specific incidence and prevalence
Many insights gained from descriptive statistics and visualizations of the same large collection of longitudinal data as created in rally 4, providing an unprecendented in-depth look at the progression, prevalence, and incidence of stunting across many populations, and producing many insights.
Sprint 7C: Catch up Growth, Growth Velocity and Population Attributable Fractions
Continued analysis of the stunting dataset, revealing evidence for a large percentage of children having multiple stunting episodes, where the traditional view of stunting has been that it is a chronic, cumulative process.
Sprint 7D: Risk Factor Analyses
This effort will focus on analyzing risk factors which will inform our understanding of the causes and underlying drivers of stunting and how they differ by region.
Sprint 8B: PRS TPP Phase I – Data, SME Survey and Model Prep
To begin defining the PRS TPP, SME engagement plans were created and SMEs were identified, a draft TPP was created, datasets in GHAP were studied for relevance to PRS-related goals, and a ki scenario modeling tool was created based on the BCG model from 2017 for pre-eclampsia to better elicit input from subject matter experts for determining the TPP.
Pregnancy Risk Stratification Model Outcome Evaluation Prototype >
Sprint 8C: SME Engagement with TPP and Model Refinement
Can we produce a finalized Target Product Profile with SME engagement? What are the performance requirements, format, measures (risk factors), and use case for a PRS tool, and how does that affect our technical approach?
Objectives for 8C were met and opened the doors for future, more specific conversations. Remaining: transferring information into the TPP format
Sprint 8D: Data Prioritization, Prep and Mapping
Do we have mapped data in GHAP that is ready for analyses and future prototyping of a PRS algorithm? Can we acquire (or expedite acquisition of) other data sources to enhance prototyping?
Significantly improved understanding of data available: Promising data sets: JiVitA-3, INTERBIO, St. Johns, GUSTO, BIGCS Ultrasound, ANISA, maybe MatlabD
Sprint 8E: Preliminary Risk Factor Analysis of Preeclampsia
What will initial exploratory analyses into the most relevant PRS data in GHAP tell us? Specifically, is a joint model for longitudinal and time-to-event data (vs. time-to-event alone) a valuable methodological approach for a dynamic PRS algorithm?
A Joint Model is a viable approach to modeling and predicting the dynamic nature of risk in pregnancy and allows for incorporating time varying biomarkers as predictors. While the availability of longitudinal data and PE as an outcome made these datasets the best to use of the ones in GHAP right now, they are not sufficient for representing populations of interest.
Sprint 9A: Determining Characteristics of Population Surveillance Estimates for Pre-Term Birth Rates
Compute covariate means for 15 groups defined by 5 sites x 3 size categories (SGA/AGA/LGA). Weighted logistic regression models were fitted to different population groups for various small subsets of covariates that accurately estimated the pre-term birth rates in LGA and SGA segments of the population.
Sprint 10AB: Childhood Acute Illness and Nutrition (CHAIN) Collaboration: Optimizing Prevention and Treatment of Acute Illness
Identified groupings of children based upon a predictive algorithm allowing for the design of interventions that reduce death after discharge or readmission to the hospital.