Protocol: the International Milk Composition (IMiC) Consortium - a harmonized secondary analysis of human milk from four studies

Abstract

Introduction: Human milk (HM) contains a multitude of nutritive and nonnutritive bioactive compounds that support infant growth, immunity and development, yet its complex composition remains poorly understood. Integrating diverse scientific disciplines from nutrition and global health to data science, the International Milk Composition (IMiC) Consortium was established to undertake a comprehensive harmonized analysis of HM from low, middle and high-resource settings to inform novel strategies for supporting maternal-child nutrition and health.

Methods and analysis: IMiC is a collaboration of HM experts, data scientists and four mother-infant health studies, each contributing a subset of participants: Canada (CHILD Cohort, n = 400), Tanzania (ELICIT Trial, n = 200), Pakistan (VITAL-LW Trial, n = 150), and Burkina Faso (MISAME-3 Trial, n = 290). Altogether IMiC includes 1,946 HM samples across time-points ranging from birth to 5 months. Using HM-validated assays, we are measuring macronutrients, minerals, B-vitamins, fat-soluble vitamins, HM oligosaccharides, selected bioactive proteins, and untargeted metabolites, proteins, and bacteria. Multi-modal machine learning methods (extreme gradient boosting with late fusion and two-layered cross-validation) will be applied to predict infant growth and identify determinants of HM variation. Feature selection and pathway enrichment analyses will identify key HM components and biological pathways, respectively. While participant data (e.g., maternal characteristics, health, household characteristics) will be harmonized across studies to the extent possible, we will also employ a meta-analytic structure approach where HM effects will be estimated separately within each study, and then meta-analyzed across studies.

Ethics and dissemination: IMiC was approved by the human research ethics board at the University of Manitoba. Contributing studies were approved by their respective primary institutions and local study centers, with all participants providing informed consent. Aiming to inform maternal, newborn, and infant nutritional recommendations and interventions, results will be disseminated through Open Access platforms, and data will be available for secondary analysis.

Clinical trial registration: ClinicalTrials.gov, identifier, NCT05119166.

Keywords: breastfeeding; human milk; infant growth; infant nutrition; machine learning.

Figure 1

IMiC consortium research framework (A) and overview of participating studies (B). Only the nutritional interventions provided to breastfeeding mothers are shown (ELICIT also provided a nutritional intervention to the child from 6 to 18 months). Variation in milk sample collection timing is shown as bars extending from points. CHILD had additional anthropometric measurements at 24, 30, 36, and 48 months, and questionnaires at 2, 2.5, 3, 4, and 5 years that are not shown here. Postnatal questionnaire time-points shown here included questions about breastfeeding practices. “X” indicates a primary questionnaire time-point that included additional maternal, environmental, demographic and/or socioeconomic related questions (e.g., subject-level data not expected to change over the course of the study). VITAL-LW has many follow-ups for time-varying factors (breastfeeding and child health), and therefore the questionnaire data is shown as a continuous bar. For example, breastfeeding practices are recorded on about 20–30 days between birth and 6 months.

Figure 2

Availability of harmonized data across IMiC studies. (A) Note that the derivation annotation is just a general rule, there are sometimes exceptions for individual sites (see Supplementary Table S1). Only variables harmonized by the Ki Data Curation team are shown (e.g., excludes derived summary variables generated downstream for specific analyses). Other excluded variables: identifiers, method/descriptors, and analysis/derivation flags (e.g., recall, sampling, imputation flags). Also combines recall and non-recall versions of the same variable. (B) Number of variables available in harmonized data across IMiC studies. Excludes identifiers, method/descriptor, and analysis/derivation flags (e.g., recall, sampling, imputation flags), and combines recall and non-recall versions of the same variable.

Figure 3

Data and sample flow for the International Milk Composition (IMiC) Consortium. AL, Academic Labs; BMGF, Bill & Melinda Gates Foundation; CL, Commercial Labs; DSP, Data Science Partners; FS, Field Sites; GHC, Global Health Collaboratory; Ki, Knowledge Integration; UHN, University Health Network; UM, University of Manitoba.