This protocol was preregistered in Open Science Framework ( https://osf.io/9nq7z) and all individual trials were prospectively registered online ( Table 1). We followed The Preferred Reporting Items for Systematic Reviews and Meta-Analysis-Protocol framework, and the checklist is appended. ¹⁴ The reporting of the IPD meta-analysis will use the PRISMA-IPD reporting checklist. ¹⁵

Four BEP supplementation trials during lactation will be included in this prospective IPD meta-analysis located in: India (IMPRINT), Burkina Faso (MISAME-III), Pakistan (MumtaLW), and Nepal (MINT) (see Table 1 for study information and acronym definitions). All four studies are part of the BEP Initiative. Recruitment is complete for IMPRINT, MumtaLW, and MISAME-III, and sample sizes are as follows: IMPRINT (n=816), MumtaLW (n=957), MISAME-III (n=800), and MINT (n=726). The expected total sample size is 3,299 women. The sample size for some exploratory outcomes will be lower than the total enrollment sample size due to certain information (e.g., blood collection or analysis) being collected on a subset of participants (by design). Also, the sample size for the MISAME-III and MINT trials is lower than the sample size for the full trials because these trials have a factorial design to provide BEP during pregnancy and/or lactation. Groups receiving BEP supplementation during pregnancy are not included in this analysis. Risk of bias assessments will be carried out and reported for each of the individual trials using the Cochrane risk of bias tool (a publically available tool). ¹⁶ There is no risk of duplicate data.

For this prospective meta-analysis, we assume the mean LAZ at six months of age to be -0.57 (standard deviation 1.10) in the control group. ¹⁷ A total sample size of 3,299 (1,809 for BEP package group and 1,490 for control group) yields 88% power to detect a minimum LAZ difference of 0.10 between BEP and control at a significance level of 5% using mixed-effects linear regression, and assuming an intra-class correlation as low as 0.30. For secondary outcomes, we will have 89% power to detect a standardized mean difference of 0.10, assuming an intra-class correlation as low as 0.20. The PASS Software v22.0.4 was used for sample size calculation (NCSS, LLC, Kaysville, Utah).

In the IMPRINT trial, mother-infant dyads were included in the study if a participant-initiated breastfeeding within seven days postpartum. In the MumtaLW trial, lactating women with undernutrition (MUAC <23.0 cm), between 13-49 years of age, and their newborns (captured within seven days from birth) were included in the study. Additionally, lactating women had to intend to exclusively breastfeed the infant for the first six months of age. In the MISAME-III and MINT trials, women between 15-40 and 15-30 years, respectively, were enrolled in the study following a positive urine test if they were found to be missing menstruation in the prior five weeks and following an ultrasound examination that revealed an intrauterine pregnancy <21 completed weeks of gestation. They were in the trial from early gestation through 6 months postpartum and were counseled to exclusively breastfeed for 6 months. Additionally, in all trials, participants indicated that they were not allergic to BEP ingredients (e.g., peanuts).

Table 1 summarizes the basic study information for the included maternal BEP lactation trials. Among them, IMPRINT followed an individual randomization, controlled efficacy trial design. MumtaLW followed a multi-arm, community-based randomized controlled, open-label, assessor-blinded superiority trial design with a treatment allocation ratio of 1:1:1. MISAME-III was an individual randomized 2x2 factorial efficacy design where participants are individually and randomly allocated to a prenatal intervention or control and a postnatal intervention or control group. Similarly, MINT followed a household randomized 2x2 factorial efficacy design where participants in the same household obtained the same prenatal and postnatal intervention or control. The current analyses will focus only on the postpartum intervention.

The intervention tested in each trial was a BEP supplement given to women from birth (or within a week of birth) to six months postpartum. The intervention groups also received what the control group received as the standard of care (see below) in addition to BEP. IMPRINT’s nutritional intervention was five different BEP snacks produced by a local company and a separate multiple micronutrient supplement ( Table 2). ¹⁸ The MumtaLW trial provided a BEP local product called Mumta that is fortified with multiple micronutrients; two sachets per day were provided. ¹⁹ MumtaLW had a third intervention arm in which the women received BEP and the infants also received one dose of azithromycin at 42 days. The MISAME-III and MINT trials provided a micronutrient fortified BEP supplement produced by the Nutriset (Malaunay, France). ²⁰

All four trials had a control arm that was intended to align with the standard of care for postpartum women in each country ( Table 2). In the IMPRINT trial, women were encouraged to use IFA, calcium, and vitamin D supplements for six months postpartum from the national program in India (i.e., not provided by the trial). ¹⁸ The national program also provides a food supplement that contains 600 kcals (18-20 g protein). In the MumtaLW trial, women received IFA from the government program or the trial (i.e., the trial ensured it was provided) for six weeks postpartum. In the MISAME-III trial, women also received IFA from the trial for six weeks postpartum. Finally, in the MINT trial, IFA was not provided as it was not part of the standard of care during lactation in Nepal.

Most trials provided participants (intervention and control) with counseling for nutrition, breastfeeding, and infant care, or referred to services where counseling was available. Trials also encouraged women to use postnatal clinical care for themselves and their infants and referred to clinical services in the case of illness.

The primary outcome for this IPD meta-analysis is infant length-for-age z score (LAZ) at six months of age, which represents linear growth across that timeframe. Secondary outcomes include infant weight and malnutrition at six months of age (weight-for-length z score (WLZ) and weight-for-age z score (WAZ)), infant growth velocity (i.e., change in length and weight), infant stunting (LAZ <-2), wasting (WLZ <-2), and underweight (WAZ <-2)) at six months of age. All infant weight and nutrition status indices (LAZ, WAZ, WLZ) will be calculated based on WHO Growth Standards ²¹ (or INTERGROWTH-21 ^st for infants born preterm, when gestational age is available), which are sex-specific. We will also assess maternal anthropometry including mean BMI and MUAC, underweight (BMI <18.5 kg/m ²) and low MUAC (<23.0 cm) at six months postpartum ( Table 3).

Exploratory outcomes include infant head circumference-for-age and MUAC-for-age z scores (based on WHO Growth Standards, see above) at six months; maternal and infant biomarkers for iron deficiency, anemia, inflammation, and other micronutrients at six months of infant age (ferritin, transferrin receptor, hemoglobin, C-reactive protein (CRP), alpha-1-acid glycoprotein (AGP), other micronutrients (biomarkers to be determined)); and maternal mortality rate (per 100,000 live births) and infant mortality rate (per 1,000 live births; Table 3).

Individual-level variables collected by trials for women include maternal age, education, baseline weight and height, parity, adherence to BEP, health and pregnancy history, diet, breastfeeding type (exclusive, predominant, and partial), and household food insecurity. For infants, variables include sex, gestational age at birth, age at each measurement (calculated from birth date), weight, length, head circumference, and MUAC. Nutritional biomarkers are available for some maternal and infant blood samples (see list in exploratory outcomes). Study-level variables include nutritional content of BEP, geographic setting (Africa or South-East Asia), and prevalence of undernutrition in study population (e.g., infant stunting or maternal underweight).

Data for this meta-analysis will come from the four participating trials. The investigators of each trial have agreed to provide data and will send individual trial datasets to a data repository (hosted by the study sponsor). Trial investigators will remove personal identifiers, and the data repository analysts will further scan for and remove any identifiable information. Then, de-identified data will be shared with the meta-analysis research team at Penn State. After Penn State receives individual data sets, they will clean and merge them to create a pooled dataset for the IPD meta-analysis. While working with these data sets, only the Penn State IPD meta-analysis research team will have access to the individual and pooled datasets, which will be stored in a secure drive, password protected by the university.

Each trial will share their original data to the data repository team, which will prepare de-identified datasets. First, we will review the dataset processed by the data repository team to see if it includes all variables needed for the meta-analysis. Although we will receive cleaned data, we will conduct additional examination of variables quantitatively and visually by checking distributions, frequencies, missingness, and outliers (biological or statistical). For instance, for the LAZ outcome and all other z-scores, we will use the WHO criteria in determining biologically implausible values using the recommended cutoffs (LAZ <-6 or >+6; WAZ <-6 or >+5, WLZ, MUAC z-score, and head circumference z-score <-5 or >+5). ^{22 , 23} Next, we will complete initial transformations to normalize continuous distributions and categorize variables as appropriate. This step will include re-coding or creating new variables to align variable definitions for analysis with the proposed definitions from the prior harmonization work. ¹³ Further, we will query any anomalies and compare the sample sizes and descriptive statistics with prior publications and study protocols. We will resolve any data issues or questions with the investigators for the corresponding trial. Then, we will be able to combine individual trial data sets into an analytic data set along with a variable to indicate which trial the data came from. The final product of this stage will be a merged dataset that is ready for the main analysis.

Data validation and merging will be conducted in Stata (StataCorp, College Station, Texas); data analysis and graphic presentation of the results will be run in Stata and R (R Core Team). To investigate the effects of the BEP intervention package on outcomes, a one-stage IPD approach will be implemented for all analyses. ²⁴ This approach is preferred over the conventional two-stage IPD approach, in which trial-level aggregated summary data are pooled in the first stage for a meta-analysis model in the second stage. ^{25 , 26} Overall, the use of a one-stage IPD approach will allow us to adopt more appropriate likelihood functions, make fewer model assumptions, and incorporate more effective modelling of effect modifiers to accommodate the between-study heterogeneity while we quantify the effects of BEP on maternal and child outcomes. ²⁷

For the main analysis, the BEP intervention package will include all study arms that provided BEP to women (in the MumtaLW trial, BEP and BEP with infant azithromycin are combined; for the other trials, there is only one BEP arm). For continuous outcomes measured at six months of infant age, the effect of BEP on the outcome (e.g., LAZ) will be assessed with mixed-effects linear regression models. For dichotomous outcomes (e.g., stunting), the effect of BEP on the outcome will be assessed using relative risk estimates from mixed-effects log binomial models (or alternatively, Poisson regression if log-binomial models do not converge). We will perform both univariable analysis (BEP intervention and trial in model) and multivariable analysis (adding adjustment for baseline individual- and study-level characteristics) to examine the effect of BEP on outcomes. For multivariable models, we will include variables that are statistically significant predictors of the same outcome in the univariable analysis and variables that are unbalanced at baseline in randomized groups. In all models, intervention arm will be specified as a fixed effect, trial will be specified as a random effect (to account for heterogeneity between trials), and any other variables in multivariable models will be specified as fixed effects. Heterogeneity between trials will be assessed using I ² statistics. Risk of bias assessments will be carried out and reported for each of the individual trials using the Cochrane risk of bias tool.

We plan to examine effect modifiers because understanding which groups may benefit the most from BEP supplementation is an important goal of this initiative. An interaction term between the treatment and a potential effect modifier will be included in separate mixed-effects linear or log binomial (or Poisson) regression models. We will examine the following individual-level subgroups based on biological plausibility and prior literature: maternal age (<20 y, 20-29 y, ≥30 y), education (none vs ≥1 year), parity (1 child vs ≥2 children), maternal BMI (<18.5 kg/m ² vs ≥18.5 kg/m ²), maternal height (<150 cm vs ≥150 cm), MUAC (<23 cm vs ≥23 cm), ²⁸ infant sex (male vs female), and infant malnutrition as defined by low birth weight (<2,500 g vs ≥ 2,500 g), stunting (LAZ <-2 vs ≥ -2), wasting (WLZ <-2 vs ≥ -2), and underweight (WAZ <-2 vs ≥ -2). Study-level variables we plan to examine are: prevalence of infant stunting at baseline, and prevalence of maternal underweight at baseline.

In a set of exploratory analyses, we will examine if breastfeeding types (exclusive, predominant, partial) mediate the effect of BEP on the outcomes at six months. We will also analyze differences in benefits to maternal outcomes by levels of BEP adherence (<80% vs ≥80%), which can be useful in planning future studies or dissemination efforts.

We will set an alpha cutoff of p <0.05 to determine if our results are significantly different from those expected if the null hypothesis was correct. For testing interactions, we will use a cutoff of p <0.10. The method of Benjamini and Hochberg ²⁹ will be applied to control for false discovery in multiple comparisons in the assessment of interaction effects in subgroup analyses.

The IPD meta-analysis will be analyzed based on an intent-to-treat protocol, which assumes that all randomized individuals will be included in the analyses. However, if we observe more than 20% of data missing per treatment arm, it will be flagged as potential bias (differential missingness), and we will consider two possible approaches, either performing multiple imputation assuming that data are missing-at-random, ^{30 , 31} or modeling the missing data by building a Bayesian hierarchical modeling and test different missing data patterns. ³² The robustness of the results will be tested in sensitivity analyses using complete cases. To estimate the missing values, the participant baseline characteristics (maternal age, maternal education, maternal BMI and height at baseline, parity, household food insecurity, baseline value of LAZ, and infant sex) will be summarized using descriptive statistics.

We will test a two-stage approach in sensitivity analyses to assess the robustness of our results to different approaches. Usually, both approaches provide similar results. ^{25 , 26} We will also assess the effects of variations in interventions such as BEP alone vs BEP with azithromycin vs control; BEP given with vs. without an IFA tablet; and differences in nutrient content (e.g., total calories (400 vs 600 vs 800 kcals)) in BEP products.

The results of this work will seek peer-review and publication in fully open-access journals. Further, a centralized data repository will provide pooled and de-identified datasets for data analysis. This is the first prospective IPD meta-analysis that tests the effect of BEP given during lactation on maternal and infant outcomes. The results from this work will aid in expanding the WHO recommendation of BEP given to pregnant to also lactating women, if proven beneficial.

Protocol version 2. The project began on October 15, 2020. It is funded by the BMGF. We preregistered the lactation meta-analysis on Open Science Framework ( https://osf.io/9nq7z) on October 4, 2022. This protocol manuscript mirrors our preregistration but expands and adds important details for the included trials and data analysis methods. We have received data from three out of four participating trials via an external data repository. The data collection is on-going for the fourth trial and will conclude in 2024, when we anticipate having the complete pooled dataset.

For each individual trial, local ethical approval was obtained. This IPD meta-analysis uses de-identified data (deemed exempt from ethical approval by the Institutional Review Board at Penn State). The de-identified and harmonized IPD meta-analysis dataset will be stored in a data repository. The research findings will be disseminated in peer-reviewed, open-access journals.