The effect of milk type and fortification on the growth of low‐birthweight infants: An umbrella review of systematic reviews and meta‐analyses

Abstract Approximately 15% of infants worldwide are born with low birthweight (<2500 g). These children are at risk for growth failure. The aim of this umbrella review is to assess the relationship between infant milk type, fortification and growth in low‐birthweight infants, with particular focus on low‐ and lower middle–income countries. We conducted a systematic review in PubMed, CINAHL, Embase and Web of Science comparing infant milk options and growth, grading the strength of evidence based on standard umbrella review criteria. Twenty‐six systematic reviews qualified for inclusion. They predominantly focused on infants with very low birthweight (<1500 g) in high‐income countries. We found the strongest evidence for (1) the addition of energy and protein fortification to human milk (donor or mother's milk) leading to increased weight gain (mean difference [MD] 1.81 g/kg/day; 95% confidence interval [CI] 1.23, 2.40), linear growth (MD 0.18 cm/week; 95% CI 0.10, 0.26) and head growth (MD 0.08 cm/week; 95% CI 0.04, 0.12) and (2) formula compared with donor human milk leading to increased weight gain (MD 2.51 g/kg/day; 95% CI 1.93, 3.08), linear growth (MD 1.21 mm/week; 95% CI 0.77, 1.65) and head growth (MD 0.85 mm/week; 95% CI 0.47, 1.23). We also found evidence of improved growth when protein is added to both human milk and formula. Fat supplementation did not seem to affect growth. More research is needed for infants with birthweight 1500–2500 g in low‐ and lower middle–income countries.

LBW infants, both during their initial hospitalization after birth and after their discharge to home, is important for survival, growth and normal development.
In 2011, the World Health Organization (WHO) published a broad review of LBW nutrition, Guidelines on Optimal Feeding for Low-Birthweight Infants in Low-and Middle-Income Countries. The guidelines recommended mother's unfortified milk as the initial option for feeding LBW infants, with donor human milk being the next best choice if mother's milk is not available. Fortification of human milk was recommended only in the case of inadequate weight gain. Notably, most of the studies included in this WHO review were judged to be of poor quality, such that 13 of the 18 guidelines (72%) are based on 'weak' or 'weak situational' evidence.
A number of systematic reviews of the feeding of LBW infants have been published since the establishment of the WHO guidelines.
We chose to conduct an umbrella review, an overview of systematic reviews, to coalesce the data on a large number of feeding interventions. Umbrella reviews are used to synthesize evidence on a broad topic and facilitate decision making (Biondi-Zoccai, 2016). The objective of this umbrella review is to summarize the available review literature on the relationship between milk options for LBW infants, including human milk, infant formula and infant milk fortifiers, and growth up to 6 months post-term. We hope that this evidence synthesis may provide guidance for the formation of feeding recommendations, while acknowledging that many other factors, such as morbidities including necrotizing enterocolitis, cost and feasibility, are also important considerations to guide feeding choices.
The prevalence of LBW is disproportionately high in low-and middle-income countries (LMICs). An estimated 91% of LBW infants are born in LMICs (Blencowe et al., 2019). Given the size and vulnerability of this population, we were particularly interested in principles for feeding LBW infants that are tailored to resourcelimited environments. In undertaking this umbrella review, we anticipated that the bulk of research on the feeding of LBW infants has been conducted in high-income settings. Because we were interested in conducting a comprehensive search, we chose not to restrict our inquiry to LMICs but to pay particular attention to the results stemming from this group. Because the preponderance of LBW infants surviving from LMICs fall into the 1500-2500 g weight band, we have a special interest in infants with these birthweights.
As much as possible, we wanted to synthesize the available evidence to formulate recommendations for feeding LBW infants in LMICs, while acknowledging the limitations in extrapolating principles between populations.

| METHODS
We followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines while conducting this review (see Data S1). We registered the protocol for this review with PROS-PERO prior to review initiation and submitted updates for protocol modifications. The full protocol is available in the Data S1.

| Search strategy
We conducted a search of Medline, CINAHL, Embase and Web of Science databases. The initial query was done in Medline, CINAHL and Embase in December 2018 with the addition of studies from the Web of Science in March 2019. The search was updated in January 2020. Search terms included probes for 'low birthweight', 'premature', 'small for gestational age', 'breast milk', 'infant formula', and 'systematic review', in addition to a number of related terms (see Data S1). We limited our selection to articles published in English and only included systematic reviews or meta-analyses. We had no limitations on publication dates. If multiple versions of a systematic review were available, we only included the most recently updated version.
We only included reviews for which full-text articles were available and did not include studies from the grey literature.

| Population, intervention and outcomes
Our population of interest was LBW infants. As such, we limited our umbrella review population to reviews primarily targeted to preterm infants or those with birthweight <2500 g. We calculated a pooled weighted average birthweight among the primary studies included in each review both to see if the population met the inclusion criteria (<2500 g) and to better understand the profile of the population represented by the review.
We considered interventions relating to infant milk options, including types of milk such as formula or human milk and milk fortification with macronutrients including added fat, carbohydrate or protein components. Reviews only addressing vitamin or mineral fortification were excluded. We included both inpatient and outpatient

Key messages
• Energy (fat or carbohydrate) and protein fortification of human milk is associated with increased growth in lowbirthweight infants during birth hospitalization, although not associated with increased growth between discharge and 6 months.
• Formula compared with donor human milk is associated with increased growth in low-birthweight infants.
• Most low birthweight feeding studies have focused on infants with a birthweight <1500 g; only a few focus on infants with birthweight 1500 to <2500 g, a group with a unique nutritional profile.
• Only a small percentage of the studies of nutritional interventions for low-birthweight infants have been conducted in low-and lower middle-income countries.
interventions. Comparison groups varied by review, but all included unfortified mother's own milk and/or other infant milk and fortification options.
Our outcome of interest was growth assessed through measurement of weight, length, head circumference, body composition, skinfold thickness or fat-free mass. Reviews that did not report anthropometrics were excluded. We included growth outcomes from birth up to 6 months post-term.

| Study selection and data extraction
Two reviewers (K. N. and M. M. D.) independently screened eligible articles first by title and abstract, then full text. We determined inclusion based on population, intervention and outcome criteria discussed previously. Any screening conflicts were resolved through discussion between the two reviewers and adjudication by a third reviewer (K. E. A. S.). In accordance with PRISMA guidelines, two reviewers (K. N. and M. M. D.) independently completed data extraction for 50% of the included studies and independently achieved >80% agreement (K.N. and M. M. D.), with a single reviewer extracting data from the remainder of the studies (K. N.). See the Data S1 for a complete list of extracted information.

| Data analysis
We evaluated the risk of bias of selected reviews using the 'A Measurement Tool to Assess Systematic Reviews (AMSTAR 2)' checklist, a quality assessment tool specifically designed for systematic reviews (Shea et al., 2017). The strength of evidence from every unique metaanalysis was graded on the basis of conventions established in other umbrella reviews (Belbasis et al., 2015;Bellou et al., 2017;Fusar-Poli & Radua, 2018). We extracted all data used to grade the evidence from the reviews. We did not calculate a pooled effect size between reviews. The evidence was classified as follows: • Convincing: fixed-or random-effects P-value <0.00001, population size >500, 95% confidence interval excludes null, heterogeneity I 2 value <50%.
• Highly suggestive: does not meet criteria for convincing and fixedor random-effects P-value <0.00001, population size >500, largest study excludes null.
• Suggestive: does not meet criteria for convincing and fixed-or random-effects P-value <0.001, population size >500.
• Weak: does not meet criteria for convincing and fixed-or randomeffects P-value <0.05 • Not significant: fixed-or random-effects P-value >0.05

| RESULTS
We screened the titles and abstracts of 1278 references. Sixty fulltext articles were reviewed, and 26 reviews met eligibility criteria for data extraction (Figure 1). A list of full text articles that were not included as well as reasons for elimination is shown in the Data S1. These 26 review articles included 150 unique studies. Some of these studies were included in more than one systematic review. A list of individual studies included in multiple reviews is shown in the Data S1.
F I G U R E 1 Flow chart of selection of eligible reviews T A B L E 1 Characteristics of the 26 included reviews Author ( (1) Low protein intake (<3.0 g/kg/day) (2) High protein intake of equal to of greater than 3.0 g/kg/day but less than 4.0 g/kg/ day (3) Very high protein intake of equal to or greater than 4.0 g/kg/day Intergroup comparison

Moderate quality
Liu (2015) Infants with birth weight

| Characteristics of included reviews
The included reviews address a number of feeding options for LBW infants (Table 1). Twenty-one of the 26 reviews had a pooled weighted average birthweight of 900-1499 g. Four reviews had a pooled weighted average birthweight between 1500 and <2500 g.
Twenty-three reviews limited their population to studies of premature infants. One review specifically addressed infants who were term but small for gestational age (SGA) (Santiago et al., 2019). Twenty reviews reported the country or region that was the setting of the primary studies. Of these, 98% (147/150) of primary studies were conducted in upper middle-or high-income countries or predominantly highincome regions. Two per cent of primary studies (3/150) occurred in India, a lower middle-income country. No studies occurred in lowincome countries. See the Data S1 for additional study details. Nineteen reviews conducted formal meta-analyses, whereas seven reviews presented outcomes as a narrative summary of included studies. Some reviews included population-based or non-randomized studies (Data S1). Authors were limited in the conclusions that could be drawn from these studies. Studies without a control population were not included in meta-analyses and thus did not impact the effect size.

| Quality assessment of included reviews
None of the included reviews met AMSTAR 2 criteria to be considered high quality. Seventeen of the 26 systematic reviews were of moderate quality. Seven reviews were considered low quality, and two were considered critically low quality. See Figure 2 and Table 1 for AMSTAR 2 results. The effect size and 95% confidence intervals of key meta-analyses are depicted as forest plots in Figure 3. See Data S1 for detailed growth outcomes of individual reviews.

| Donor human milk compared with formula
Three reviews comparing donor human milk with formula found consistently greater growth in the formula group (Boyd et al., 2007;Quigley et al., 2018 ;Yu et al., 2019 Boyd and colleagues did not perform a meta-analysis but presented a narrative summary of studies comparing donor human milk and formula. Most studies in this review found greater weight, length, head circumference and skinfold thickness gains in the formula group when given as either an exclusive diet or as a supplement to mother's own milk (Boyd et al., 2007).

| Exclusive breastfeeding
A single review compared exclusive breastfeeding with a number of other infant feeding options for full-term infants who were small for F I G U R E 2 Outcomes and quality of evidence for key meta-analyses included in this umbrella review are grouped by type of intervention. The direction and significance of the weighted mean difference are indicated by the colour of the circle.

| Carbohydrate only fortification
One review investigated carbohydrate fortification of human milk with a nonhuman short-chain galacto-oligosaccharide/long-chain fructo-oligosaccharide supplement (Amissah et al., 2018a). The weight in the intervention group was higher at 30 days compared with infants fed nonfortified human milk (MD 160.4 g; 95% CI 12.4, 308.4). Other growth metrics were not reported.

| Fat only fortification
Several reviews analysing fat fortification of both human milk and formula found that it makes no difference in growth. Five reviews specifically addressed the effect of long-chain polyunsaturated fatty acids (LC PUFA) (Gibson et al., 2001;Moon et al., 2016;Newberry et al., 2016;Rodríguez et al., 2012;Udell et al., 2005). Three metaanalyses of formula fortified with LC PUFA showed no statistically significant effect on weight, length or head circumference within the first 6 months of life (Moon et al., 2016;Newberry et al., 2016;Udell et al., 2005). Two systematic reviews described mixed results in a qualitative assessment (Gibson et al., 2001;Rodríguez et al., 2012 (Nehra et al., 2002).

| Protein and amino acid fortification
Five reviews examined the impact of protein fortification on growth; most reported increased growth associated with protein supplementation (Amissah et al., 2018c;Fenton et al., 2014;Liu et al., 2015;Pimpin et al., 2019;Tonkin et al., 2014).  et al., 2018c). Similarly, a review of high-versus low-protein fortification of human milk found increased weight gain, linear growth and head growth in the higher-protein group (Liu et al., 2015). Increased weight gain (MD 2.36 g/kg/day; 95% CI 1.3, 3.4) and head growth (MD 0.37 cm/week; 95% CI 0.16, 0.58) were also seen in a comparison of high-versus low-protein formula (Fenton et al., 2014). One

| Strength of evidence for individual metaanalyses
Nineteen systematic reviews presented the data for a combined total of 100 meta-analyses. We graded the quality of evidence of the individual meta-analyses on the basis of established criteria commonly used in umbrella reviews as convincing, highly suggestive, suggestive, weak or nonsignificant (Belbasis et al., 2015;Bellou et al., 2017;Fusar-Poli & Radua, 2018). See Data S1 for a complete list of meta-analyses and associated strength of evidence components. No individual metaanalysis met criteria of convincing or highly suggestive evidence. Seven associations were supported by suggestive evidence. These included greater weight gain, linear growth and head growth associated with (1) energy and protein-fortified human milk compared with unfortified human milk and (2) formula (term or preterm) compared with donor human milk (fortified or unfortified), as well as greater weight gain associated with preterm formula compared with fortified donor human milk.
Fifty meta-analyses met criteria for weak evidence. Thirty-nine metaanalyses were not significant. Four meta-analyses were not adequately assessed because of the absence of data.

| Summary of main results
This umbrella review found 26 reviews composed of 150 unique primary studies evaluating the effect of infant milk options on the growth of LBW infants up to 6 months post-term. We found evidence that energy and protein fortification of human milk is associated with increased weight gain, linear growth and head circumference compared with unfortified human milk (quality: suggestive) . We also found evidence that formula is associated with human milk against necrotizing enterocolitis. We acknowledge that many factors beyond a simple calculation of growth are important to consider in the formation of feeding recommendations, but our findings beg the question of what is the most appropriate feeding strategy for this population, particularly for infants with birthweight 1500 to <2500 g who are at lower risk for necrotizing enterocolitis than infants with birthweight <1500 g.
Multiple systematic reviews supported the use of higher protein content in both human milk and formula to increase growth, although the sample size was too small for this evidence to be considered 'suggestive', as defined by umbrella review criteria. Several interventions did not result in increased growth. These include infant milk with added fat, carbohydrate, LC-PUFA, glutamine or taurine.
Hydrolyzation also seemed to make no difference in growth. These interventions generally occurred during the birth hospitalization.
Several reviews investigated the question of postdischarge fortification of human milk and formula. This was generally not associated with increased anthropometric parameters at 3 to 4 months or 6 months post-term (Young et al., 2013(Young et al., , 2016. This finding raises questions about the continued use of fortification after discharge, a common practice following the initial hospitalization of LBW infants.
The use of energy and protein-fortified formula may be less practical for many families because of limited access and increased cost of fortified formula compared with standard formula. Energy and protein-fortified human milk requires the steps of expressing the milk, mixing it with milk fortifier, and bottle feeding an infant, a process that can be burdensome compared with direct breastfeeding.
It is important to acknowledge that we considered greater growth to be desirable in this population given the high risk for poor growth early in life and associated morbidities. There is increasing evidence that children who were born with LBW are at increased risk for metabolic syndrome later in life, particularly those who were SGA.
The paradigm of 'growth is good' may not be appropriate in an older cohort. We attempted to include more nuanced measures of growth such as skinfold thickness or fat-free mass but found little data in the review literature.

| Quality of the evidence
We

| Completeness and applicability of the evidence
In addition to limitations due to the quality of evidence, the generalizability of the included reviews may be limited because of the population represented within the primary studies. Most reviews were primarily composed of studies of very low birthweight (VLBW) (birthweight <1500 g) preterm infants in high-income countries.
Although we sought to include recommendations for all LBW infants, the study populations in the included reviews were concentrated within the VLBW weight band. Eighty-one per cent of reviews had an average population birthweight between 1000 and 1500 g, whereas only 15% had an average birthweight falling between 1500 and <2500 g (Cao et al., 2018;Fenton et al., 2014;Gibson et al., 2001;Santiago et al., 2019). Most of these were deemed to be low quality on the basis of the AMSTAR 2 rating (Cao et al., 2018;Gibson et al., 2001;Santiago et al., 2019). This is significant because the growth patterns of VLBW infants cannot necessarily be extrapolated to infants within the higher birthweight population. Based on our findings, the unique population of infants with birthweight between 1500 and <2500 g is underrepresented in the current review literature. These infants constitute the majority of LBW infants, but the current WHO recommendations are based on literature with a population that is not truly representative of this contingent of the LBW population.
SGA full-term infants were another LBW group that was underrepresented among these reviews. SGA infants have unique growth patterns and nutritional requirements (Tudehope et al., 2013). Only one low-quality review specifically focused on full-term infants who were SGA (Santiago et al., 2019). Nutritional recommendations for a VLBW preterm population cannot necessarily be extrapolated to the SGA full-term infant. This is particularly important for an LMIC setting, in which the majority of LBW infants will be SGA but term (Lee et al., 2013).
We found that the majority of research on LBW feeding was con-

| Limitations
The umbrella review methodology facilitates the evaluation of a broad research question in a manner difficult to achieve in an individual systematic review. We chose to conduct this type of review given the breadth of literature regarding nutrition of LBW infants. The scope that can be achieved is sweeping in nature; however, the overview methodology has a number of inherent limitations. Primary studies related to the research question but not included in other systematic reviews may be missed, potentially excluding valuable information.
Umbrella reviews are necessarily limited to the most recent literature preceding the search date of the individual systematic reviews. The majority of reviews in our study were published in the past 5 years, but four reviews were >10 years old. We used a strength of evidence classification system that has been well described in the umbrella review literature, but which relies heavily upon P-values. The P-value is a potentially misleading tool for determining the quality of a study, as it does not account for risk of bias or the degree to which a significant result may be clinically meaningful. We have included the AMSTAR 2 results and the forest plans with effect size and 95% confidence intervals to provide a more complete picture of the quality and strength of the evidence for the individual systematic reviews and meta-analyses included in this umbrella review.
Given the broad scope of interventions that we considered in this review, we limited our outcome of interest to infant growth. Necrotizing enterocolitis, neurodevelopment, kidney function and a number of other outcomes in preterm infants have all been correlated with the choice of infant milk and milk components. Choosing to focus on a single outcome, albeit an important one, provides an incomplete picture of the effect of nutrition on the heterogeneous components of health and well-being.
We were interested in all metrics of growth but found the majority of growth outcomes were reported as weight gain, linear growth and head growth. Important measures of body composition, such as fat-free body mass or skinfold thickness, received very little attention in the review literature.

| CONCLUSION
Through this umbrella review, we were able to conduct a high-level survey of the landscape of evidence on various sources of nutrition for LBW infants and their association with growth that led us to a number of conclusions.
(1) Energy and protein fortification of human milk is associated with increased weight gain, linear growth and head growth (quality of evidence: suggestive).
(2) Formula compared with donor human milk is associated with increased weight gain, linear growth and head growth (quality of evidence: suggestive).
(3) Studies of the ideal nutritional interventions for LBW infants in low-and lower middle-income countries are vastly underrepresented in the literature.
(4) Reviews of infant milk interventions are focused primarily on the <1500 g birthweight population, with few studies focused primarily on infants in the 1500 to <2500 g weight band, a group that may have a unique nutritional and growth profile.
Based on the gaps we have identified, we recommend additional research focused on the nutritional needs of infants with a birthweight 1500 to <2500 g and infants born in LMICs because both of these subgroups represent vulnerable populations who are underrepresented in the available review literature. We included all growth metrics in our search but found very few outcomes in the reviews addressing body composition or assessment of lean versus fat mass, important areas of focus for future research.

ACKNOWLEDGMENTS
The Bill & Melinda Gates Foundation funded this review as part of their support for the Low birthweight Infant Feeding Exploration (LIFE) study. The funders did not have input on data collection, management, analysis or interpretation of the data. Further, they did not have any authority over the writing of the reports or decision to