Introduction

Necrotizing enterocolitis (NEC) is a devastating illness in premature infants, with severe forms reaching mortality rates of 15–25%, and survivors often developing short bowel syndrome and neurodevelopmental impairment.1,2,3,4 As treatment of NEC remains mainly supportive in nature, efforts over the last 50 years have focused on prevention. Apart from the protective effect of human milk and antenatal steroids, no interventions have shown a clear benefit in preventing NEC.1,5,6,7 With incidence rates between 5 and 12% in very low birthweight (VLBW) infants, there is a clear need for preventive therapies.4,8 Evidence from clinical and animal studies indicates that aberrant activation of the intestinal immune system by Gram-negative bacteria is central to the pathogenesis of NEC.1,5,9 As such, attempts to modulate the intestinal microbiome using prebiotics and probiotics have been tested for over two decades.10,11,12 The use of probiotics has been shown to decrease rates of stage 2 and 3 NEC in clinical trials and meta-analysis encompassing over 100,000 infants.13,14,15 Strikingly, probiotics have also been shown to decrease rates of late-onset sepsis and all-cause mortality in preterm infants, however, effect sizes have been modest.14,15,16

Despite such evidence, widespread adoption of probiotics in neonatal intensive care units (NICU) has met several challenges. Particularly in the USA, these include the unavailability of a pharmaceutical-grade product and limited efficacy data for extremely low birthweight (ELBW) infants. Recommendations for adoption have been further complicated by the risk of probiotic sepsis (i.e., sepsis caused by an administered probiotic organism). A recent case report of an infant death attributed to probiotic sepsis elicited a strong warning from the Federal Drug Administration (FDA) and has resulted in near cessation of probiotic use in the USA for preterm infants.17 The FDA’s stance aligns with a cautious note struck by the American Academy of Pediatrics (AAP) committee on the fetus and newborn.18 However, the European Society for Paediatric Gastroenterology Hepatology and Nutrition (ESPGHAN) and European Foundation for the Care of Newborn Infants (EFCNI) have endorsed the use of probiotics in premature infants, considering the benefit-to-risk ratio.19,20 A sentiment echoed by the American Gastroenterological Association (AGA).21,22

Yet, as the incidence rates of probiotic sepsis remain unknown, the data underlying such statements remain incomplete. To date, FDA and AAP warnings have been primarily grounded in case reports of probiotic sepsis, failing to capture the magnitude of exposed infants. While support of probiotic efficacy often references studies that indirectly address probiotic sepsis in discussion of adverse events, leading to ambiguity if a lack of data indicates no cases occurred were not directly assessed. Without jointly considering both a number of infants exposed to probiotics and robust event rates, recommendations for or against probiotics cannot be accurately assessed. To directly address this paucity of data, this study presents a meta-analysis of probiotic sepsis incidence in preterm infants. Utilizing over 60 studies for which counts of probiotic sepsis and a denominator of exposed infants were available. We also provide a comprehensive series of sub-analyses within the ELBW and VLBW populations, by formula vs. breastmilk feeding, study design, geographic location, timeframe, and probiotic type. Rates of secondary outcomes of NEC, death, and sepsis within these studies were also extracted as a balancing measure to rates of probiotic sepsis. Finally, a synthesis of published case reports is provided to explore the risk factors and outcomes associated with probiotic sepsis.

Methods

Study selection

To assure a comprehensive review of neonatal probiotic sepsis rates, we focused our search on identifying all forms of published studies that evaluated probiotic use in this population, including randomized trials, observational studies, and case reports. Given the lack of standardization around probiotic products used, identification of research studies through a keyword search is known to be challenging (previous probiotic reviews retained <5% of identified papers from database searches). Rather than replicate low-yield searches, we seeded our pool of candidate studies with all citations from two recent landmark studies: a network meta-analysis15 and systematic review23 on the topic. The search strategy in these works was shown to be highly robust, and in one case included an independent peer review of the search protocol.24 We supplemented this initial pool of studies through a snowball review of citations from included works.

All studies were obtained through online resources or library requests. Each study was screened by two independent reviewers (drawn randomly from the team of neonatologists, nutritionists, and statisticians) to determine if probiotic sepsis was explicitly reported. In the event of a split decision, a third reviewer made a final determination. This preliminary screening removed duplicates, studies involving non-human subjects, non-probiotic interventions (i.e., prebiotics), and those that occurred prior to 2000. Randomized trials were assessed with the Risk-of-Bias 2 criteria25 and observational studies with the Newcastle-Ottawa Scale26 (Supplementary Materials Tables S1 and S2).

Data abstraction

Positively screened studies were then subjected to detailed data abstraction. A standardized data extraction sheet provided reviewers with a uniform guidance for the abstraction process. Again, for reliability, dual-reviewers were used. For each study, several categories of data were captured. These included details around: study design and logistics, probiotic used, cohort demographics, probiotic sepsis rates, and secondary outcomes of mortality and morbidity (NEC, and other sepsis events). Specific data elements of each category can be found in the Supplementary Materials.

Finally, for all case reports and any study for which a description of a probiotic sepsis event was provided, an additional data abstraction was completed by reviewers. This form captured detailed information regarding the event and outcome, and included infant demographics (birthweight, gestational age, and sex), narrative hospital course, antibiotic use, and sepsis outcome.

Data synthesis and statistical analysis

Once processed, pooled estimates were created through a series of meta-analysis models. Given heterogeneity in treatment outcomes and an array of infant conditions represented in the exposed population across the studies, random-effect models were used to avoid the assumption of a common underlying treatment effect across studies made by the alternative fixed-effect approaches.27,28

The primary outcome around the incidence of probiotic sepsis was assessed as a proportion of reported probiotic sepsis cases divided by the total number of infants exposed to probiotics. To pool estimates across each study, a generalized linear mixed model (GLMM) was used. In line with best practices from literature illustrating such models result in lower biased estimates for meta-analysis comprised both of a large number of total studies, and studies reporting 0 events.29,30

For each secondary outcome of NEC, sepsis, and death, a relative risk was calculated between cohorts of probiotic-exposed and non-exposed infants in each study (as available). Again, pooled estimates were obtained using GLMM. To obtain estimates for studies reporting zero events, continuity correction was used. Studies for which an outcome was not directly assessed were excluded. Additionally, in line with standard practice for Cochrane protocols, studies with 0 events in either cohort were excluded.

Data processing and cleaning were performed in Python v.3.8.8. All analyses were conducted using the meta v.7.0,31 and metafor v.4.432 packages in R v.4.2.1.

Results

A total of 160 candidate studies were identified for initial review. After exclusions, 75 underwent complete data abstraction (Fig. 1). Broadly, these included 63 studies that provided a defined cohort from which outcome rates could be estimated (20 observational studies [3 prospective, 17 retrospective], and 43 clinical trials), as well as 12 case reports. An overview of the study cohorts is shown in Table 1, further stratified by study design (observational and randomized trials).

Fig. 1: Study selection flow diagram.
figure 1

This figure presents inclusion and exclusion criteria for studies considered for the two primary analyses (meta-analysis of probiotic sepsis/secondary outcomes and review of case probiotic sepsis case reports).

Full size image
Table 1 Overview of cohorts within included studies.
Full size table

Primary outcome—probiotic sepsis

Looking first to the incidence of probiotic sepsis, we found over 20,000 exposed infants across 63 study cohorts,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66,67,68,69,70,71,72,73,74,75,76,77,78,79,80,81,82,83,84,85,86,87,88,89,90,91,92,93,94,95 where eight infants were reported with sepsis events related to their probiotic organism (<0.04%). Random-effects from generalized linear meta-analysis models result in an estimate of incidence at 0% (95% confidence interval (CI) [0–10%], Fig. 2). Of note, the 8 infants were found in only 2 distinct studies (Esaiassen et al.59 and Sakurai et al.89). These studies presented 3 (of 290 total infants) and 5 (of 272) infants with sepsis, resulting in individual estimates of incidence rates at 1% [0–3%] and 2% [1–4%], respectively. The remaining 61 studies reported 0 incidence of probiotic sepsis.

Fig. 2: Meta-analysis estimate of probiotic sepsis incidence.
figure 2

Figure displays a forest plot summarizing the binary-event meta-analysis. For each study indicating the number of events, total number of infants exposed to probiotics, and estimated confidence interval for the event rate. The lower part of the image provides properties of the random effects meta-analysis model, including the pooled effect and heterogeneity estimates across studies.

Full size image

For studies reporting on ELBW (n = 1901) and VLBW (n = 11,121) populations (directly, or through sub-analysis of a primary cohort), no probiotic sepsis events were identified. Meta-analysis estimates and sample sizes for these studies can be found in the Supplementary Materials (Data Extraction Details).

Secondary outcomes—NEC/sepsis/death

In total, 56, 53, and 47 studies were considered for the NEC, sepsis, and death outcome comparison, respectively. In addition to studies which did not directly assess a given outcome, and those whose results indicated zero events for both cohorts, two studies did not provide control groups (evaluating probiotic sepsis in two-armed probiotic studies) and were excluded from secondary analyses. Nonetheless, sample sizes allowed for robust estimates, as cohorts for the 3 outcomes were each comprised of >18,000 infants.

Crude rates highlighted a reduced incidence of each outcome in the probiotic-exposed vs control cohorts (NEC: 3.25% vs. 5.70%, sepsis: 18.16% vs. 21.78%, and death: 7.04% vs. 8.71%). This was reflected by meta-analysis estimates of relative risk across studies. The largest effects were seen in NEC with a relative risk of 0.60 [95% CI: 0.51–0.71], followed closely by sepsis 0.85 [95% CI: 0.78–0.93], and death 0.84 [95% CI: 0.76–0.93]. Forest plots and effects of individual studies can be found for each outcome in Fig. 3. As in the primary analysis, the relative risk of each outcome in the ELBW and VLBW populations was computed. Results were generally found to be consistent in these sub-groups and are presented in Supplementary Materials (Figs. S1–S8).

Fig. 3: Meta-analysis estimates for secondary outcomes.
figure 3

Figure displays a forest plot summarizing the relative risk analysis between probiotic exposed and unexposed infants. For each study indicating the number of events and total infants experiencing the outcome for each cohort. The lower part of the image provides properties of the random effects meta-analysis model, including the pooled effect and heterogeneity estimates across studies. Three plots are provided, one for each outcome assessed. Panel a considers the rates of NEC. Panel b considers rates of clinical sepsis, and Panel c considers mortality rates of infants across the included studies.

Full size image

Meta-analysis for these outcomes produced high I2 values, suggesting potential subgroup heterogeneity. For robustness, all modeling was replicated across several strata. These included study design (RCT vs. observational study), probiotic details (single vs. multi-strain), and stratification by study year (pre- vs. post-2016) to account for changes in neonatal practice and probiotic use over time. Results of these sub-analyses were stable across strata. Consistently showing reductions in all three secondary outcomes. The large reduction in NEC rates remained most stable, while sepsis and death outcomes showed more variability. Particularly when stratified by study year, with small effects in post-2016 studies, likely in line with improvements in neonatal care. Forest plots and model estimates for each combination can also be found in the Supplementary Materials (Figs. S9–S24).

Finally, given discussion in literature on the effectiveness of probiotics based on the source of enteral nutrition (breastmilk vs. formula), we also independently analyzed studies by type of infant feeding, comparing cohorts exclusively fed on breastmilk (mothers or donor) vs exclusively formula vs. those including infants on both types of feeds (mixed). In total, 40 studies captured mixed populations, 11 studies were found to use exclusive breastmilk (9 Trials and 2 observational), and 3 were exclusively formula-fed (all single-strain probiotic trials). The remaining nine studies did not provide sufficient information to make a determination (unknown) and were excluded from supplementary analysis. Supplementary Figs. S12, S16, S20 and S24 highlight strong consistency of results across feeding types. As expected from the literature, the effect size was strongest for all outcomes in the exclusive breastmilk cohort. Encouragingly, we continue to see significant improvements in outcomes for probiotic exposure in the mixed-use cohort as well. For this cohort, the majority of infants were breastmilk fed. However, we note n = 2020 exclusively formula-fed infants were considered within relative risk calculations of the mixed cohort. Supporting conclusions and risk estimated are presented in the remainder of this work.

Case descriptions

Finally, we turn to the review of reported cases. In total 27 cases of probiotic sepsis with infant details were identified. A summary of each event can be found in Table 2.

Table 2 probiotic case details.
Full size table

All cases reported gestational age, sex, and probiotic organism. Most infants were born preterm (median GA 28 weeks, IQR: 25–32), and 57.7% were male. In addition to prematurity, about 40% of cases had underlying comorbidities such as genetic conditions (e.g., 5p deletion), congenital gastrointestinal conditions (e.g., gastroschisis), or acquired gastrointestinal diseases (e.g., spontaneous intestinal perforation); and about 60% of cases had a central line. Probiotic use was dominated by Bifidobacterium and Lactobacillus, in both single and multiple strain products. Twenty-three cases reported birthweight (median: 970.0 g, IQR: 739.0–1451.5), and 21 reported days of exposure to the probiotic, centered around a median 13 days of probiotic use (IQR: 7–23).

The outcome post-infection was generally good, with 78% of cases reported to recover from probiotic sepsis and were discharged home. There were 2 reported deaths. One in an infant who recovered from probiotic sepsis but died due to an unrelated cardiac condition. The other in an infant who developed right inguinal tenderness and redness that progressed towards multi-organ failure and death the next day.

Discussion

Probiotics are “live microorganisms that, when administered in adequate amounts, confer a health benefit on the host”.96 Over the past 2 decades, the benefits of routine probiotic supplementation in infants have been demonstrated and replicated in literature. These have included elements of morbidity and motility, as well as nutritional elements tied to infants’ growth and outcomes (e.g., improving gastrointestinal motility, enteral feeding tolerance, and shifting microbiota away from dysbiosis).15,20,97,98

Despite evidence to their benefit, we recognize administration of live bacteria carries a risk of sepsis or bacteremia associated with the given probiotic strain. As such, concerns outlined by the AAP and FDA must be given significant consideration in practice decisions. However, neither report offers estimates to the prevalence of such events, highlighting instead adverse event reports and published case series. As a result, such reports can do little to quantify tradeoffs of alternative adverse outcomes for which probiotics may offer benefit. There exist excellent reviews of neonatal probiotic sepsis23 and probiotic outcomes in neonates.15,99 Yet to the best of our knowledge, this is the first attempt to comprehensively estimate probiotic sepsis risk and associated risk-benefit ratios.

Probiotic sepsis

Results of our analyses demonstrate that the overall rate of probiotic sepsis was overwhelmingly low. Across 63 studies (including 40 randomized trials), less than 0.04% of exposed infants reported probiotic sepsis events. Notably, based on the FDA-specific call-out of VLBW infants, our review finds no cases of probiotic sepsis across 32 unique studies and over 11,000 exposed infants in this population.

Interestingly, all 8 events were captured in two large observational studies representing 562 infants. This imbalance in identification may be due in part to the emphasis on testing methods, as both studies highlight unique aspects of their protocol to improve the ability to detect probiotic sepsis events. Esaiassen et al. note their use of matrix-assisted laser desorption/ionization time-of-flight mass spectrometry for improved strain identification.59 While Sakurai et al. note a tailored culturing protocol to match the properties of the probiotic organism administered (B. breve). They note B. breve has differential growth periods in blood cultures under anaerobic (~40 h in literature) or aerobic (133 h) conditions. They credit the detection of events to the hospital practice of growing all cultures for 7 days (covering both aerobic and anaerobic growth conditions). Given the American Society of Microbiology’s Cumitech Blood Culture Test Guidelines recommend a minimum culture period of 5 days, cultures stopped at this time point would have failed to detect B. breve events under aerobic growth conditions used in their study.89

Exploration of study characteristics

In further considering variability across studies, we undertook a review comparing randomized trials and observational study designs, evaluating key risk factors including demographics and probiotic strain.

Cohort demographics

While the sample size of randomized trials was understandably smaller than observational studies (μ: n = 125 and n = 747.8, respectively) the cohorts were generally similar. With no statistical differences in subject’s mean birthweight or percent of male sex (Mann–Whitney U test, α < 0.05). All studies focused on preterm infants; however, observational studies tended to report on younger infants (~1 week earlier GA, from 29 to 28 weeks). Notably, supporting the benefit of additional stratification of supplementary analyses, there was no difference in GA of studies specifically focused on ELBW infants (μ: 26.5 and 26.8 weeks).

Probiotic organism

At a high level, we find a higher rate of multi-strain probiotics in observational studies, likely reflecting products commonly utilized in real-world practice (e.g., Infloran®), as compared to research in randomized trials. There were also notable differences in strains across studies. All studies using a multi-strain product (both observational and trials) included Bifidobacterium. Interestingly, while Lactobacillus was present in all multi-strain probiotic preparations used in all observational studies, it was present in only 86% of multi-strain trials. These trials were found to include higher rates of less common strains, including Streptococcus (22.7% vs. 7.7%) and Enterococcus (4.5% vs. 0%). One exception was Saccharomyces, which was included in 7.6% of multi-strain formulations for observational studies but only 4.5% of trials.

This pattern of more diverse strains evaluated in trials is mirrored for studies considering single-strain probiotics, with usage of Saccharomyces (19.1% vs 0%) and Bacillus (9.5% vs 0%) elevated in randomized trials vs observational studies, respectively. Notably, the use of common probiotic strains differed in single-strain studies as well. Bifidobacterium was significantly elevated in observational studies 42.9% vs. 23.8% in trials, while Lactobacillus was used in 28.6% of observational studies but 47.6% of trials. Encouragingly, our supplementary meta-analyses stratified both by trial vs. observational studies, and single vs. multiple strain probiotics produced consistent results, improving confidence in results.

Risk-benefit tradeoff

Looking broadly at the concept of risk, it is important to note that with nearly all clinical interventions on preterm infants, a potential for harm exists.100,101,102,103 As such, it is valuable not simply to estimate the risk of probiotic sepsis events, but to calculate tradeoffs between events and rates of secondary adverse outcomes for which probiotics have been shown to improve.

Meta-analysis estimates found non-probiotic-exposed infants to be at 1.67×, 1.18×, and 1.19× the risk of NEC, mortality, and sepsis outcomes compared to probiotic administered cohorts. We can further contextualize these values by estimating common tradeoff measures, comparing rates of each secondary adverse outcome (NEC, mortality, sepsis) between infants in probiotic-exposed and unexposed (control) cohorts. Specifically, we calculate the number needed to treat that are expected from withholding probiotics in the control group, finding 1:40 for NEC, 1:63 for all-cause mortality, and 1:28 for sepsis.

Finally, we can extend these tradeoff estimates to jointly consider the risk of probiotic sepsis events by estimating the excess risk of each outcome per expected case of probiotic sepsis. By considering the rate of probiotic sepsis among exposed infants at 1:~2500, the incidence of NEC in control cohort (~5.6%), and baseline NEC incidence in the probiotic treated group (~3.2%), findings suggest an expected increase (excess) of 62 NEC cases in the unexposed cohorts per (one) case of probiotic sepsis. Similar calculations yield 42 more deaths (8.6% vs. 7.0%), and 92 more infants with sepsis in the unexposed group (21.8% vs. 18.2%) per case of probiotic sepsis.

Case descriptions

Finally, turning to the reports of infants with confirmed probiotic sepsis, we note there were only two infant deaths (93% survival), which itself is lower than late-onset sepsis mortality in preterm infants. However, demographics across the reports suggest risk appears to be greatest in the ELBW population.

These findings are in line with AAP recommendations for caution in this population.18 As well as literature that notes variability in efficacy of probiotics in ELBW infants for NEC, mortality, and infection, highlighted in the recent Cochrane review.104 Such findings are mirrored in the wide confidence intervals of the ELBW-focused analyses presented in this work (Supplementary Materials). Expanding on the heterogeneity of product types and strains, this observed variability illustrates that heterogeneity in probiotic benefit may also be also tied to infants’ clinical profiles. Suggesting more targeted guidance by the FDA and others may be needed for infants at the highest risk for adverse outcomes. Additionally, we note the ELBW population remains understudied (the fewest number of unique studies identified by both this work and the Cochrane review, as compared to VLBW and preterm infants). Suggesting a need to consider both the scope and quality of evidence in making a determination of a cohort’s acceptable risk-benefit tradeoff for probiotic use.

In addition to well-recognized elements of extreme prematurity and birthweight <1000 g, case details in Table 2 provide insight into other notable risk factors related to infants’ clinical condition and hospital course. First, probiotic sepsis was more common in infants with pre-existing surgical conditions of the intestine and central lines.105,106 This ties closely to the increased risk of using probiotics for scenarios other than NEC reduction. Such as in infants with short bowel syndrome (previously associated with CLABSIs and probiotic strains107), significant congenital/genetic conditions, and those with small intestinal bacterial overgrowth.108 Next, there appears to be an increased risk for infants with evidence of compromised intestinal barrier function. Including use of probiotics following established cases of NEC, where the risk of bacterial translocation is potentially high.

While it is clear that risk for probiotic sepsis is multifactorial, there exists a potentially high-risk subpopulation of extremely preterm, medically complex infants for whom probiotic risk may exceed the anticipated benefit, and for whom specific guidance may be warranted. Finally, it should be noted that across all reported events, none occurred within the context of an RCT. Suggesting that standardized preparation and feeding protocols may play a part in probiotic safety. This is in line with several publications offering clinical guidance for the safe administration of probiotics.18,109,110,111 As well as prominent calls to action for improving testing standards for probiotic products themselves. Including the framework proposed by Shane et al.,112 emphasizing collaboration with manufacturers to design probiotics with a higher stringency of testing than products manufactured for the healthy population.

Limitations

Although this work has taken steps to provide a robust assessment of probiotic sepsis incidence and secondary mortality and morbidity outcomes in probiotic-exposed premature infants, it is subject to limitations of review-style analyses. The primary consideration being reporting bias of probiotic sepsis rates. Diagnosis of probiotic sepsis requires confirmation that the underlying sepsis organism matches an administered probiotic, which may not be captured by routine blood cultures. In turn, underestimating the degree of total risk or resulting in misclassification of events as clinical sepsis and introducing bias into estimates of risk-benefit tradeoffs.

To assess the impact of these biases on estimates calculated in our analysis, we leveraged a subset of 18 studies identified during review as describing robust and systematic microbiology-based screening protocols to detect probiotic sepsis in their cohorts (i.e., testing all cultures for administered probiotic organism).37,43,47,49,51,54,56,57,60,61,63,64,68,73,75,76,80,94 These studies were primarily high-quality RCTs (15 trials, 3 observational), and reflected the broader set of literature captured in the meta-analysis with mean (SD) birthweight and gestational age of 1158.70 (171.72) and 28.82 (1.40), respectively. In total these studies captured over 4500 infants exposed to probiotics, with no infant reported to develop probiotic sepsis.

Rates of NEC continued to be reduced (3.46% in probiotic-exposed vs. 6.16% in control cohorts) and similar death rates (6.18% and 7.60% for exposed and control, respectively). Notably, rates of clinical sepsis were found to be elevated relative to full set of studies (22.26% for exposed infants vs 28.73% in the control cohorts), suggesting the robust testing protocols were effective. As the potential of underreporting and misclassification was significantly less likely given these testing protocols, results of this sub-analysis should provide confidence that low rates of probiotic sepsis, and overall tradeoff with risk of secondary outcomes presented by our meta-analysis, are reliable.

Nonetheless, we continue to emphasize the necessity of proactive and systematic assessment protocols in institutions utilizing probiotics that allow for the detection of both anerobic and aerobic strains. Beyond improved estimates of risk, reliable testing allows for appropriate antibiotic coverage to be selected and increases the likelihood of favorable outcomes. We also recognize that culture-based assessment of probiotic strains may no longer be the only option for robust screening. In an era of increasing focus on Precision Medicine, the potential for molecular-based diagnostics and novel biomarkers to identify those babies with potential bacterial translocation, bacteremia, and sepsis should be considered.113,114

Finally, we recognize that there likely exists a more extensive utilization of probiotics in real-world neonatal practice than those infants captured by studies considered in this meta-analysis. Specifically, there exists a potential for publication bias in which studies may fail to report the routine use of probiotics in their unit, biasing the exposure rate. Studies may also fail to adequately capture adverse outcomes associated with probiotics, particularly in retrospective studies. Together with improved microbiological testing capabilities, the availability of robust post-market surveillance data for probiotics could significantly improve our understanding of risk, both by identifying missed cases and by better estimating the denominator of exposed infants.112 Nonetheless, we remain reassured as to the general rates reported here, as similar results were found within a stratified sub-analysis of those RCT studies that utilized robust assessment protocols.

Conclusion

Probiotics have been by far the most extensively investigated therapy in preterm infants, and one of the few that consistently have been shown to decrease NEC rates. This meta-analysis highlights that the incidence of serious adverse probiotic sepsis events is extremely rare, particularly in the older preterm and VLBW cohorts. Further, by considering the occurrence of probiotic sepsis and NEC, sepsis and mortality outcomes in the same study population, this work can provide critical quantitative figures in the risk versus benefits discussion for this population.

Together, the discussion across ESPGHAN,20 AGA,21,22 EFCNI,19 and FDA17 statements highlights the need for a balanced approach of offering an efficacious, but not risk-free therapy. Efforts must be made to identify those infants at high risk for probiotic adverse events or for whom therapy should be avoided. Ongoing support from scientific groups also reminds us that therapeutic decisions should be made in partnership with parents, and additional work remains to appropriately address these considerations.19,115 We are in agreement with the AAP recognizing the lack of pharmaceutical-grade products, heterogeneity of data, and potential for harm from probiotic sepsis. Yet, in the light of current evidence, we believe general avoidance of an overwhelming well-tolerated therapy represents its own form of risk to infants. We hope this work spurs improved reporting not only of outcomes but of formulations and administration practices on which to support future use in neonatal units.