Human milk-derived versus bovine milk-derived fortifiers for preterm infants: a systematic review and meta-analysis of randomised trials

Plain language summary

In preterm infants receiving human milk, fortification with a human milk-derived product rather than a bovine milk-derived product has not been shown to reduce necrotising enterocolitis, mortality, sepsis, chronic lung disease or retinopathy. The pooled estimate for NEC favours the human milk-derived product but is compatible with no effect, depends heavily on the earliest and least directly relevant trial, and rests on 48 events in total. The certainty of the evidence is low for both primary outcomes. Adequately powered randomised trials that hold the base diet constant are needed before human milk-derived fortification can be recommended on clinical grounds.

Abstract

Background. Human milk does not by itself meet the protein, mineral and energy requirements of the very preterm infant, and multi-nutrient fortification of maternal or donor milk is standard practice in neonatal intensive care. Conventional fortifiers are manufactured from bovine milk protein. An alternative class of fortifier is manufactured from pooled, pasteurised donor human milk, and is marketed as part of an "exclusive human milk diet" in which every component of enteral nutrition — base milk, fortifier and any supplement used to cover a shortfall in the mother's own milk — is of human origin.

Objective. To determine the effects of human milk-derived multi-nutrient fortifier, compared with bovine milk-derived fortifier, on necrotising enterocolitis, mortality and other clinical outcomes in preterm infants receiving human milk.

Methods. MEDLINE/PubMed, Europe PMC and ClinicalTrials.gov were searched to 22 September 2026, supplemented by reference-list searching of five prior syntheses. Randomised trials published from 2010 onward comparing a human milk-derived multi-nutrient fortifier with a bovine milk-derived fortifier in preterm infants fed human milk were eligible. Two primary outcomes were pre-specified: NEC Bell stage II or greater, and all-cause mortality before discharge. Dichotomous data were pooled as Mantel-Haenszel random-effects risk ratios, continuous data as inverse-variance mean differences. Risk of bias was assessed with RoB 2 and certainty with GRADE.

Results. 7 trials (15 reports; 920 infants) were included. RR 0.63 (95% CI 0.36 to 1.09) for NEC (I² = 0%, τ² = 0.00, Q p = 0.49) and RR 0.77 (95% CI 0.43 to 1.36) for mortality. No secondary outcome showed a significant difference. Certainty was moderate for bronchopulmonary dysplasia, low for both primary outcomes, and very low for surgical NEC and growth.

Conclusions. Human milk-derived fortification has not been shown to reduce NEC or death; the evidence is of low certainty and adequately powered trials are needed.

Background

Human milk does not by itself meet the protein, mineral and energy requirements of the very preterm infant, and multi-nutrient fortification of maternal or donor milk is standard practice in neonatal intensive care. Conventional fortifiers are manufactured from bovine milk protein. An alternative class of fortifier is manufactured from pooled, pasteurised donor human milk, and is marketed as part of an "exclusive human milk diet" in which every component of enteral nutrition — base milk, fortifier and any supplement used to cover a shortfall in the mother's own milk — is of human origin.

The biological rationale for avoiding bovine protein rests on the hypothesis that intact bovine milk proteins provoke an inflammatory response in the immature intestine and thereby contribute to necrotising enterocolitis (NEC), the most feared gastrointestinal complication of prematurity. Observational series and the first randomised trial of an exclusive human milk diet reported substantially lower rates of NEC, and a number of neonatal units have adopted human milk-derived fortification on that basis. The products are, however, an order of magnitude more expensive than bovine fortifiers, and donor milk is a finite resource, so the size and certainty of any clinical benefit matter a great deal for practice and for policy.

Two features of the evidence base complicate its interpretation. First, several trials have compared not fortifiers alone but whole dietary strategies: the human milk-derived fortifier arm also receives donor human milk or human milk-based formula in place of preterm formula, so the contrast confounds the fortifier with the base diet. Second, the outcome that motivates the intervention — NEC — is uncommon, so individual trials have been far too small to detect or exclude clinically important differences.

This review synthesises the randomised evidence published from 2010 onward, separating, where the data allow, trials that compared fortifiers alone from trials that compared an exclusive human milk diet with a mixed diet.

Objective

To determine the effects of human milk-derived multi-nutrient fortifier, compared with bovine milk-derived fortifier, on necrotising enterocolitis, mortality and other clinical outcomes in preterm infants receiving human milk.

Methods

Results

All-cause mortality before discharge (6 trials)

Six trials reported deaths before discharge (25/496 versus 29/415). RR 0.77 (95% CI 0.43 to 1.36); p = 0.37, an absolute difference of -13 per 1000 (-43 to +16). Heterogeneity was low (I² = 10%, τ² = 0.05, Q p = 0.35) and the subgroup test was not significant (p = 0.52). No sensitivity analysis changed the conclusion; the most extreme leave-one-out estimate was RR 0.58 (0.32 to 1.06) on removal of Embleton 2023.

Surgical necrotising enterocolitis (3 trials)

Three trials reported NEC requiring surgery (7/330 versus 13/259). The random-effects estimate was RR 0.42 (95% CI 0.12 to 1.49); p = 0.18, with moderate heterogeneity (I² = 41%, τ² = 0.52, Q p = 0.18). This outcome was not stable across models: the fixed-effect estimate (RR 0.41 (0.17 to 0.96)) and the Peto odds ratio (0.37 (0.15 to 0.93)) both excluded the null, and removal of Jensen 2024 produced RR 0.21 (0.06 to 0.76) while removal of Sullivan 2010 produced RR 0.83 (0.26 to 2.71). With 20 events in total these differences reflect the fragility of the analysis rather than a substantive finding.

Late-onset or culture-proven sepsis (6 trials)

Six trials contributed (86/482 versus 70/401). The pooled estimate was almost exactly null (RR 1.00 (95% CI 0.75 to 1.34); p = 0.98) with no heterogeneity (I² = 0%, τ² = 0.00, Q p = 0.49). Sepsis definitions differed between trials, from culture-proven bloodstream infection to clinically treated late-onset infection.

Bronchopulmonary dysplasia / chronic lung disease (5 trials)

Five trials contributed (218/450 versus 196/365). RR 0.96 (95% CI 0.85 to 1.09); p = 0.54, I² = 4%, τ² = 0.00, Q p = 0.39. This is the only outcome for which the pooled confidence interval excludes a large effect in either direction.

Severe retinopathy of prematurity (5 trials)

Five trials contributed (47/329 versus 47/316). RR 1.02 (95% CI 0.70 to 1.47); p = 0.93, I² = 0%, τ² = 0.00, Q p = 0.44.

Feeding intolerance or major feeding interruption (3 trials)

Three trials reported a feeding-intolerance or feeding-interruption outcome (73/204 versus 75/199), using trial-specific definitions; in O'Connor 2018 this was the pre-specified primary outcome. RR 0.97 (95% CI 0.76 to 1.25); p = 0.82, I² = 0%, τ² = 0.00, Q p = 0.42. Subgroup analysis was not performed for this outcome because all contributing trials fell in the same design subgroup.

In-hospital weight gain velocity (3 trials)

Three trials reported in-hospital weight gain velocity in 404 infants. MD 0.36 (95% CI -1.30 to 2.03) g/kg/day; p = 0.67, with substantial heterogeneity (I² = 80%, τ² = 1.67, Q p = 0.01). The trials point in opposing directions: Mizuno 2026 favoured the human milk-derived fortifier and Sullivan 2010 the bovine comparator. Sullivan 2010 reported medians and interquartile ranges, which were converted to means and standard deviations, adding a further source of uncertainty.

Change in weight z-score (2 trials)

Two small trials reported change in weight z-score. MD -0.37 (95% CI -0.93 to 0.18) z-score units; p = 0.19, I² = 65%, τ² = 0.11, Q p = 0.09. Embleton 2023 reported survivors only and Berrington 2025 measured change from enrolment rather than birth, so the two estimates are not strictly commensurable.

Discussion

Summary of main results. Across seven randomised trials and 920 infants, human milk-derived fortification did not produce a statistically significant reduction in NEC, and the confidence interval remains wide enough to accommodate both a clinically important benefit and a small increase in risk. The same pattern holds for mortality, sepsis, bronchopulmonary dysplasia, retinopathy and feeding intolerance: every pooled estimate is compatible with no effect, and the intervals for the rarer outcomes are wide. Growth outcomes were heterogeneous and gave no consistent signal in either direction.

The point estimate for NEC — a 37% relative reduction — is in the direction the biological hypothesis predicts and is consistent with what earlier syntheses have reported, but it is driven disproportionately by Sullivan 2010: omitting that trial moves the pooled risk ratio from 0.63 to 0.91. Sullivan 2010 is also the trial in which the exposure contrast is widest, because the comparator arm received preterm formula rather than an all-human-milk base diet, and it is the trial whose NEC rate in the control arm (16%) is highest. The surgical NEC result behaves in the same way and is unstable: it reaches conventional significance under a fixed-effect or Peto model and loses it under random effects, which is the expected behaviour of a 20-event analysis with moderate heterogeneity and should not be read as evidence of benefit.

Overall completeness and applicability. The evidence applies to very preterm and very low-birth-weight infants receiving a human milk base diet in well-resourced neonatal units, and almost all of it comes from a single manufacturer's product. It does not address infants fed predominantly formula, nor does it establish whether any benefit — if real — comes from the fortifier itself or from displacement of preterm formula elsewhere in the diet. The subgroup contrast between exclusive-human-milk-diet trials and fortifier-only trials was not significant for any outcome, but with three to four trials per subgroup that test has very little power and the absence of a detected difference should not be taken as evidence that the two designs answer the same question.

Quality of the evidence. Certainty was rated moderate for bronchopulmonary dysplasia, low for NEC, mortality, sepsis, retinopathy and feeding intolerance, and very low for surgical NEC and the two growth outcomes. Imprecision was the dominant reason for downgrading throughout: the entire evidence base contains 48 cases of NEC and 54 deaths, an order of magnitude short of the information size needed to detect a plausible effect on either outcome.

Agreements and disagreements with other studies. The direction and magnitude of the NEC estimate agree with the 2019 Cochrane review and with subsequent pairwise and network meta-analyses, all of which likewise found no statistically significant difference while noting that observational studies report larger benefits. The discrepancy between the randomised and observational literatures is substantial and is the strongest argument for treating the observational estimates with caution: the units that adopted exclusive human milk diets early also changed feeding practice in other ways over the same period.

Implications for practice. The randomised evidence does not currently support a recommendation to replace bovine milk-derived fortifier with human milk-derived fortifier for the prevention of NEC or death. Neither does it exclude a worthwhile benefit. Given the substantial cost difference and the constraints on donor milk supply, a unit adopting human milk-derived fortification is doing so on the basis of evidence that is, at best, of low certainty.

Implications for research. A trial powered on NEC and death is the clear priority; the information size required — several thousand infants — implies an international collaboration. Such a trial should hold the base diet constant between arms so that the fortifier itself is the only variable, should adjudicate NEC blind to allocation, and should report cost and donor-milk utilisation alongside clinical outcomes. The NEOVASC trial, whose main clinical results are not yet published, and any subsequent registration of the trials identified here should be checked at the next update of this review.

Authors’ conclusions