A Systematic Review and Meta-Analysis of Human Milk Feeding and Morbidity in Very Low Birth Weight Infants

This systematic review and meta-analysis synthesised the post-1990 literature examining the effect of human milk on morbidity, specifically necrotising enterocolitis (NEC), late onset sepsis (LOS), retinopathy of prematurity (ROP), bronchopulmonary dysplasia (BPD) and neurodevelopment in infants born ≤28 weeks’ gestation and/or publications with reported infant mean birth weight of ≤1500 g. Online databases including Medline, PubMed, CINAHL, Scopus, and the Cochrane Central Register of Controlled Trials were searched, and comparisons were grouped as follows: exclusive human milk (EHM) versus exclusive preterm formula (EPTF), any human milk (HM) versus EPTF, higher versus lower dose HM, and unpasteurised versus pasteurised HM. Experimental and observational studies were pooled separately in meta-analyses. Risk of bias was assessed for each individual study and the GRADE system used to judge the certainty of the findings. Forty-nine studies (with 56 reports) were included, of which 44 could be included in meta-analyses. HM provided a clear protective effect against NEC, with an approximate 4% reduction in incidence. HM also provided a possible reduction in LOS, severe ROP and severe NEC. Particularly for NEC, any volume of HM is better than EPTF, and the higher the dose the greater the protection. Evidence regarding pasteurisation is inconclusive, but it appears to have no effect on some outcomes. Improving the intake of mother’s own milk (MOM) and/or donor HM results in small improvements in morbidity in this population.


Introduction
Human milk (HM) is the feed of choice for preterm infants [1]. However, not all mothers can provide sufficient milk to meet requirements, and supplementation with either preterm formula (PTF) or donor human milk (DHM) is common practice. Current recommendations are for the use of mother's own milk (MOM), when available, with appropriately screened and pasteurised DHM the Where statistical heterogeneity was low to moderate (I 2 ≤ 50%), a fixed effects model was used and where this changed statistical significance this has been noted in text. Where possible we have explained statistical heterogeneity above 50%.
A "Summary of findings" table was prepared for each comparison using the GRADE system (GRADEpro GDT, 2015) [30]. GRADE is designed to evaluate the quality of evidence and strength of recommendations. RTs with no limitations are considered high quality evidence and observational studies as providing low quality evidence. Studies can then be downgraded by one (for serious concern) or two (for very serious concerns) based on risk of bias, inconsistency, indirectness, imprecision and publication bias. Observational studies with a large effect size have been upgraded by one for a strong association, defined as a RR of ≤0.5 [31]. For each outcome, we report our certainty in the findings as very low, low, moderate or high separately according to study design (RTs, observational).
To interpret the overall evidence for each outcome and comparison, we used the following terminology: 1.
Clear effect/clear evidence of no effect: The certainty of evidence is moderate or above with a clinically important result from RTs, ideally aligning with results from observational studies or moderate certainty evidence from observational studies; and with reasonable numbers of events and/or participants.

2.
Probably an effect/probably no effect: There is moderate certainty from either RTs or observational studies and point estimates may be different between the 2 study types with overlapping CIs but can be explained (e.g., through heterogeneity). There are large numbers of participants and studies. 3.
Possible effect/possibly no effect: There is low/ moderate certainty with CIs which may suggest a difference although not reaching conventional statistical significance; or with a confidence interval which indicates a trivial difference only.

4.
Inconclusive: The certainty of evidence is very low to low, CIs are wide, and number of participants and studies is low.
Where possible the overall effect (absolute risk reduction (ARR), or mean difference (MD), with 95% CI) have been reported. Table 1 (Tables S3-S9) with a collated  summary of findings table presented in the manuscript (Table 2).

Risk of Bias
The six RTs [32][33][34][35][36][37] all had low risk of bias (Table 1). Sequence generation was not reported in three [34,36,37], blinding was not possible in one [37] and blinding of physicians but not nursing staff occurred in another [33]. However, as the outcomes of interest are objective, we thought these unlikely to introduce bias. Of the observational studies, 26 were assessed as low risk of bias, 14 as moderate and 3 as high (Table 1).
Overall: The observational studies show there is a possible reduction in any NEC with EHM compared with EPTF (ARR, 4.3%, from 2.5 to 5 fewer cases/100) ( Table 2). There is inconclusive evidence relating to severe NEC (Table S4).

Comparison 2: Any Human Milk vs. Exclusive Preterm Formula
Randomised trials: There were no RTs identified reporting NEC for this comparison. Observational studies: Nine cohort studies [40,41,47,52,54,61,63,66,68], comparing infants fed any HM with infants fed EPTF (Table 1), were included in the meta-analysis for this comparison (Figure 1) There was a clear effect of any HM in reducing NEC (RR 0.51, 95% CI 0.35, 0.76, n = 3783, I 2 7%; Figure 1; moderate certainty, Table S3). Henderson et al. [77] reported a case controlled study (53 NEC cases, 53 controls) from 10 NICUs in the UK which could not be included in the meta-analysis. Seventy-five percent of NEC cases received any HM compared with 91% of controls (OR 0.32, 95% CI 0.11, 0.98). This finding is consistent with the meta-analysis, however all stages of NEC were included (whereas most other studies defined NEC as Bell's stage 2 or above), and matched controls on GA only, which may not have accounted for other potential confounders [77].
Overall: There is a clear effect of any HM in reducing NEC (any) with an ARR of 3.6% (from 1.8 to 4.8 fewer cases/100); the evidence is inconclusive for severe NEC (Table 2).

Comparison 3: Higher vs. Lower Dose Human Milk Intake
Randomised trials: Four RTs [32,[35][36][37] (Table 1) were included in the meta-analysis for this comparison (Figure 1) and showed a reduction in any NEC (RR 0.59, 95% CI 0.39, 0.89, n = 1116; fixed effects; Figure 1; moderate certainty, Table S3). In all four trials the higher dose of HM was a combination of MOM and DHM, therefore making this an EHM group while HM intake in the low dose group was either not reported [36] or varied between a median proportion of enteral intake of 63% [35] to 85% [32].
Overall: There is a clear reduction in the incidence of any NEC with higher dose HM (ARR ranging from 4.3% (0.2 more to 6.8 fewer cases/100 for RTs to 3.8% (2.6 to 4.6 fewer cases/100) for observational studies) ( Table 2). There is a possible reduction in the incidence of severe NEC (ARR from the observational studies 1.8%, from 0.8 to 2.4 fewer cases/100) ( Table 2).
Overall: The evidence for an effect of pasteurised vs. unpasteurised HM on the incidence of any or severe NEC is inconclusive ( Table 2).

Late Onset Sepsis (LOS)
LOS was reported in 35 studies with the majority (n = 23) defining sepsis by the presence of a positive blood culture at >48 h to >5 days, with the need for supportive laboratory markers, treatment with antibiotics and for multiple positive cultures in the case of coagulase negative Staphylococcus.

Comparison 1: Exclusive Human Milk vs. Exclusive Preterm Formula
Randomised trials: One RT [34] reported the effect of EHM feeding (including a human milk derived fortifier) compared with EPTF feeding on the incidence of LOS (Table 1) and indicating a possible reduction in LOS (RR 0.70, 95% CI 0.47, 1.03; n = 53, Figure 2; low certainty, Table S5).
Overall: Although the RT and meta-analysis of observational studies did not reach significance, the CIs neared 1 and as such, thus we conclude there is a possible reduction in the incidence of LOS with an EHM diet (ARR from RT of 23.8% (from 42 fewer to 2.4 more cases/100) and from observational studies 5% (from 0.9 more to 8.9 fewer cases/100; Table 2).

Comparison 2: Any Human Milk vs. Exclusive Preterm Formula
Randomised trials: There were no RTs identified reporting LOS for this comparison Observational studies: Eight observational studies including seven cohort [40,47,52,54,61,63,68], and one interrupted time series study [21] (Table 1) compared the incidence of LOS in infants fed with any HM compared with those fed exclusively with PTF. On meta-analysis, no difference was detected on LOS (RR 0.95, 95% CI 0.67, 1.34; n = 2497, I 2 59%, Figure 2, very low certainty, Table S5). The source of the heterogeneity is not readily apparent but baseline differences in the population and varying dosage of HM may contribute.
Overall: The evidence to determine if the receipt of any HM compared with EPTF reduces LOS is inconclusive (Table 2).

Comparison 3: Higher vs. Lower Dose Human Milk Intake
Randomised trials: The impact of high dose vs. low dose HM on the incidence of LOS was addressed in five RTs, four of which could be combined in a meta-analysis [32,[35][36][37] (Table 1, Figure 2). No difference in LOS was detected with higher vs. lower dose HM (RR 1.07, 95% CI 0.89, 1.28, n = 1186, I 2 0%, Figure 2; moderate certainty, Table S5). In contrast, Cossey et al. [33] reported the risk of LOS according to quantity of human milk, in increments of 10 mL/kg/day, and showed that the risk of LOS was lower as both the quantity (hazard ratio (HR) 0.89, 95% CI 0.83, 0.95, p = 0.0008) and cumulative quantity of MOM increased over time (HR 0.99, 95% CI 0.98, 0.99, p = 0.0001).
A further prospective case-control study [82] conducted logistic regression and found an independent protective effect of the average daily dose of HM for every 10 mL/kg/day increase from day of life 1-28 (OR 0.98, 95% CI 0.97, 0.99, p = 0.008).
Overall: The evidence for high vs. low dose HM on reducing LOS from RTs and observational studies differs and is inconclusive ( Table 2).

Comparison 4: Unpasteurised vs. Pasteurised Human Milk
Randomised trials: One RT assessed the impact of unpasteurised HM vs. pasteurised HM on the risk of neonatal LOS [33] (Table 1) and showed no difference in the effect of pasteurisation on LOS (RR 0.71, 95% CI 0.43, 1.18, n = 303, Figure 2; moderate certainty, Table S5).
Overall: The use of unpasteurised compared with pasteurised human milk is not likely to have an effect on the incidence of LOS (Table 2).

Comparison 1: Exclusive Human Milk Compared with Exclusive Preterm Formula
Randomised trials: There were no RTs reporting BPD for this comparison. Observational studies: The relationship between an EHM diet and EPTF diet on BPD was reported in two observational studies, an interrupted time series [21] and a cohort study [66] (Table 1). There was no effect of an EHM diet on BPD (RR 0.94, 95% CI 0.26, 3.41; n = 706; I 2 = 79%, Figure 3; very low certainty, Table S6). Heterogeneity is possibly due to differences in study design.

Comparison 2: Any Human Milk Compared with Exclusive Preterm Formula
Randomised Trials: There were no RTs reporting BPD for this comparison.
Overall: The evidence for an effect of any HM compared to EPTF on the incidence of BPD is inconclusive ( Table 2). Forest plot of relative risk for the association between human milk and bronchopulmonary dysplasia.
Overall: The evidence for an effect of EHM vs. EPTF on BPD is inconclusive ( Table 2).

Comparison 2: Any Human Milk Compared with Exclusive Preterm Formula
Randomised Trials: There were no RTs reporting BPD for this comparison.
Overall: The evidence for an effect of any HM compared to EPTF on the incidence of BPD is inconclusive (Table 2).
Data from two case control studies [75,83] were unable to be included in the meta-analysis. Both studies showed a reduction in BPD associated with increasing amounts of human milk. Fonseca et al. [75] reported that a minimum amount of human milk (≥7 mL/kg/day) in the first 42 days was associated with a reduced incidence of BPD and Patel et al. [83] reported that, for every 10% increase in HM intake, the risk of BPD was reduced (RR 9.5%, 95% CI 0.824, 0.995).
Overall: The evidence for an effect of high vs. low dose HM on BPD is inconclusive ( Table 2).
Overall: There is inconclusive evidence for an effect of pasteurisation of HM on BPD (Table 2).
No RTs reported severe ROP for this comparison.
Observational studies: Four studies (one interrupted time series [21] and three cohort [53,58,66], Table 1) reported the association between EHM and EPTF feeding on any ROP. No difference was detected in any ROP with this comparison (RR 0.65, 95% CI 0.31, 1.34; n = 1256, I 2 = 84%, Figure 4; very low certainty, Table S7). The source of the substantial heterogeneity is unclear and likely due to a combination of differences in study design, baseline differences in the population and an imbalance of numbers in each group (Table 1).
Overall: The evidence for an effect of EHM compared with EPTF on ROP is inconclusive. There is a possible reduction in severe ROP with EHM (ARR 7.6%, from 2.7 to 9.1 fewer cases/100; Table 2).

Comparison 2: Any Human Milk Compared with Exclusive Preterm Formula
Randomised trials: No RTs reporting ROP were identified for this comparison Observational studies: Six observational studies including one interrupted time series [21] and five cohort studies [47,50,53,54,66] (Table 1) compared any HM with EPTF. No effect of feeding type on ROP was detected (RR 1.08, 95% CI 0.79, 1.48; n = 3576, I 2 = 75%; Figure 4, very low certainty, Table S7). Overall, there was an imbalance of infants in groups (2897 and 679 in any HM and EPTF groups, respectively). Heterogeneity is likely due to the variation in HM intake (Table 1), and to the larger more mature infants in the EPTF group in three of the studies [21,47,56].
Overall: There is inconclusive evidence for an effect of any HM vs. EPTF on either ROP or severe ROP ( Table 2).
One retrospective case control study [81] reported any ROP and feeding and could not be included in the meta-analysis. Porcelli et al. [81] found that HM intake in Postnatal Week 2 was an independent predictor for ROP surgery (OR = 0.94, CI not reported).
Overall: The evidence regarding high vs. low dose of HM on both ROP and severe ROP is inconclusive ( Table 2).
Observational studies: Three observational studies (one prospective case-control [76], one cohort study [64] and one non randomised arm of RT [36]) compared the effects of pasteurisation on any ROP (Table 1). Similar to the RT, there were no differences between feeding groups (RR 0.89, 95% CI 0.33, 2.38, n = 681, I 2 = 73%, Figure 4, very low certainty, Table S7). Meta-analysis of the two studies reporting severe ROP [36,64] also did not detect a difference (RR 0.81 95% CI 0.13, 5.08, n = 589 infants, I 2 = 86%, Figure S3; very low certainty, Table S8). The source of heterogeneity may be from differences in study design, and the variation in the relative dose of pasteurised and unpasteurised HM used (Table 1).
Overall: The evidence for an effect of pasteurisation of HM on any or severe ROP is inconclusive ( Table 2).
Motor: One study showed better motor development at 12 months of age [40] (assessed using AIMS) in the human milk group vs. The formula group (63 ± 20% vs. 46 ± 15%, respectively, n = 39, p < 0.05). Three studies [48,54,68] could be included in the meta-analysis for the age range 18 to <36 months, with no difference detected in psychomotor development index (PDI) between feeding groups (MD −0.8 points 95% CI −6.02, 4.42, n = 1744, I 2 = 77%, Figure 5; very low certainty, Table S9). Heterogeneity may be explained by the different population with 2 studies examining infants born in the late 1990s [48,68] and one using a cohort of infants born in 2005 [54], as well as differences in the dosage of HM.
The study by Vohr et al. (2007) [57] could not be included in the meta-analysis and showed that both Bayley MDI and PDI in the three highest quintiles of HM intake were significantly higher than the no HM group, p < 0.05 (mean MDI in no HM, 40th-60th, 60th-80th and >80th groups 76.5,82.7,86.4,89.7 and mean PDI 78.4,85.2,87.3,90.2 respectively) at 30 months corrected age (CA).
Overall: The evidence is inconclusive for an effect of any HM vs. EPTF on either cognitive or motor development (Table 2).
Three additional studies could not be included in the meta-analysis but reported on this comparison. Belfort and co-workers' cohort study [39] found that IQ was positively associated with the number of days that the infant received >50% human milk feeds (0.5 points/day, 95% CI 0.2, 0.8). Were and Bwibo [67] assessed a cohort of 120 preterm infants in Kenya and found an association between the use of EHM in the first month of life and functional disability at two years of age (RR 2.04, 95% CI 1.1, 3.78) p = 0.02). Vohr et al. (2007) [57] reported, for every 10 mL/kg/day increase in HM, at 30 months, the MDI increased by an estimated 0.59 points, p = 0.0005 and the PDI by 0.56 points, p = 0.009.
Overall: The evidence for an effect of high vs. low dose HM on both cognitive and motor development is inconclusive.

Comparison 4: Unpasteurised vs. Pasteurised Human Milk
No studies were identified for this comparison.

Summary of Main Results
Six RTs with 1472 infants and 43 observational studies with 14,950 infants were included in this systematic review. Both EHM and any HM, compared with EPTF, reduced NEC. A higher proportion of HM was more effective than lower amounts with a 4% ARR in any NEC and 2% reduction in severe NEC. This supports a policy of moving to 100% human milk for NEC protection when mothers are unable to meet all their infant requirements. An EHM diet was associated with a possible 5% reduction in LOS, however there does not appear to be a dose effect. There is inconclusive evidence for an effect of exclusive or any HM on the incidence of BPD or ROP, except for a possible effect of EHM, compared with formula, on reduction of severe ROP with a 7.6% reduction. We also found insufficient evidence to draw any conclusions regarding the role of HM on neurodevelopment. This outcome was complicated by the variation in the timing of testing, and the different tests used. What is clear is that the mean differences between feeding groups is small and hence large numbers will be required to show an effect. Many individual studies included in this meta-analysis are not sufficiently powered to determine these differences. The overall evidence for the effect of pasteurisation was inconclusive except for possibly no effect on LOS.

Strengths and Limitations
In this review, we have used robust methods to search, synthesise and critique evidence on this topic. We have combined five major morbidities on preterm infants into the one review, providing a comprehensive overview that is relevant to neonatal clinicians and will inform clinical decisions regarding feeding, particularly of DHM. In addition, we have attempted to differentiate the effects of various combinations of HM and PTF by synthesising data in four distinct comparisons, each designed to answer a particular question.
It was beyond the scope of the review to determine the effect of introducing a bovine derived, compared to a human derived, fortifier. We also limited our search to English language which may have failed to retrieve some literature.
For each meta-analysis, we used standard Cochrane methods for presenting pooled results-these methods appropriately give greatest weight to large studies and/or large number of events. For example, in the NEC meta-analysis (Figure 1), two large studies [66,68] provided most of the data, and thus the greatest weight, showing a clear advantage with use of any human milk compared with exclusive preterm formula.
While our inclusion criteria stipulated our population and outcomes of interest, we still encountered heterogeneity with some studies choosing to study only very preterm infants (<1000 g or <1250 g) which were a more vulnerable subset of our population of interest and may limit applicability. The majority of studies included in the meta-analysis were from developed countries, reflecting modern NICU practice, making these results quite generalisable. A large source of variability in the studies arose from the exposure to HM. Most studies measured exposure over the neonatal admission whereas some focussed on early feeding only. There was considerable heterogeneity in the dose of HM within each group and this was particularly so in the "any HM vs. EPTF" and the "high vs. low dose" HM groups which could vary from as high as EHM to the lower 20% of intake, or was not measured at all in many cases. Where heterogeneity was substantial, the certainty of the evidence was downgraded to reflect this, and so for many of the outcomes we are uncertain about the evidence despite quite large numbers of studies included in the meta-analyses. The true effect may be substantially different from the estimate provided from these studies and more studies of robust design are needed to increase our confidence. In addition, the fortifier used for HM was generally bovine derived but sometimes human derived and we did not differentiate between these as this was beyond the scope of this review. Nevertheless, the avoidance of bovine protein in an otherwise EHM diet, may have an impact which we have failed to take into account. Finally, another source of heterogeneity arises from the various definitions of the outcomes used, and in the case of neurodevelopment, the tools used to measure this.
All six RTs were assessed as low risk of bias and the observational studies varied with 26 considered low risk, 14 as moderate and 3 as high risk of bias. Our risk of bias assessment did not take into account poor statistical methods, typical of many of the observational studies, as this is not relevant to a meta-analysis, but makes individual study results unreliable. Additionally, many of the studies had a small sample size or were designed to answer a different question and included the outcomes of interest as secondary outcomes, hence were often not powered to detect small differences.

Findings from Other Reviews
Two recent narrative systematic reviews [85,86] and three meta-analyses [87][88][89] have been published on this topic. Cacho et al. [85] reviewed the evidence for the effect of DHM, EHM and the dose of HM, on NEC and, in line with our results, showed no clear evidence that DHM compared with formula reduces NEC, while an EHM diet may be protective and a higher dose of HM reduces the risk of NEC.
De Silva et al. [87] conducted a narrative review of infection rates in preterm infants. Of the nine studies they included, five were not included in our review due to being published prior to 1990 (n = 3), the study population not meeting our inclusion criteria (n = 1) or not published in English (n = 1). De Silva et al. concluded that the literature overall did not support a benefit of HM in preventing LOS, despite some small studies showing a protective effect and poor study design in many of the included studies. Our review included a larger number of more recent studies and despite this only found possible evidence of a protective effect EHM vs. EPTF. A recent meta-analysis of the effect of DHM (+/− MOM) vs. PTF on BPD by Villamor-Martinez [88] with considerable overlap of studies in our review, found no effect from the seven RTs included in their review but eight observational studies showed reduced BPD with DHM. However, our certainty of this finding, as determined by GRADE, is very low; hence we have given more weight to the RT results. A recent meta-analysis of observational studies on the effect of HM on ROP by Zhou et al. [89] used comparisons which overlapped with ours, and showed a protective effect of HM on ROP and severe ROP for both EHM vs. EPTF and "mainly HM vs. mainly formula" which equates to our high vs. low dose HM group. Similar to our findings, Zhou et al. found no effect in the "any HM vs. EPTF" group. In a narrative review of neurodevelopment, which included many of the same studies as in our review, Lechner and Vohr [85] presented evidence of a small protective effect of HM but also acknowledge the challenges of studying an outcome that has so many confounding variables such as parental IQ and associated socioeconomic differences. They also highlighted the lack of high quality studies in this area and the need to control for confounding variables.
In addition, one study [90], which we were unable to include because the outcomes were reported as a composite, also found an association between HM (during the first 10 days of life) and improved outcomes. In their retrospective review of 349 infants born weighing <1500, any HM in the first five days of life was associated with a lower incidence of NEC, LOS and/or death. During Days 6-10, it was only when HM intakes were >50% of the total intake was a protective effect elicited.

Implications for Practice
We have shown evidence of a clear protective effect of HM against NEC and a possible reduction in LOS, severe ROP and severe NEC. In addition, we have shown that any HM is better than none, that the more HM the preterm infant receives the better the outcome, and that for NEC there is an advantage in topping up infants who are already receiving quite large proportions of their enteral intake as HM, to EHM. From a clinical perspective, it would seem just as important to offer DHM to an infant who is getting nearly all MOM as it is for an infant who is getting none.

Implications for Research
The benefits of HM feeding are difficult to study given that it is not ethical to randomise breast feeding. However, there is a need for large and well conducted studies, designed to answer specific questions, particularly in relation to the effects of DHM and pasteurisation.