Probiotics Prevent Late-Onset Sepsis in Human Milk-Fed, Very Low Birth Weight Preterm Infants: Systematic Review and Meta-Analysis

Growing evidence supports the role of probiotics in reducing the risk of necrotizing enterocolitis, time to achieve full enteral feeding, and late-onset sepsis (LOS) in preterm infants. As reported for several neonatal clinical outcomes, recent data have suggested that nutrition might affect probiotics’ efficacy. Nevertheless, the currently available literature does not explore the relationship between LOS prevention and type of feeding in preterm infants receiving probiotics. Thus, the aim of this systematic review and meta-analysis was to evaluate the effect of probiotics for LOS prevention in preterm infants according to type of feeding (exclusive human milk (HM) vs. exclusive formula or mixed feeding). Randomized-controlled trials involving preterm infants receiving probiotics and reporting on LOS were included in the systematic review. Only trials reporting on outcome according to feeding type were included in the meta-analysis. Fixed-effects models were used and random-effects models were used when significant heterogeneity was found. The results were expressed as risk ratio (RR) with 95% confidence interval (CI). Twenty-five studies were included in the meta-analysis. Overall, probiotic supplementation resulted in a significantly lower incidence of LOS (RR 0.79 (95% CI 0.71–0.88), p < 0.0001). According to feeding type, the beneficial effect of probiotics was confirmed only in exclusively HM-fed preterm infants (RR 0.75 (95% CI 0.65–0.86), p < 0.0001). Among HM-fed infants, only probiotic mixtures, and not single-strain products, were effective in reducing LOS incidence (RR 0.68 (95% CI 0.57–0.80) p < 0.00001). The results of the present meta-analysis show that probiotics reduce LOS incidence in exclusively HM-fed preterm infants. Further efforts are required to clarify the relationship between probiotics supplementation, HM, and feeding practices in preterm infants.


Introduction
Late onset sepsis (LOS) is one of the most common causes of morbidity and mortality in preterm infants [1,2]. It occurs in approximately 20% of very low birth weight (VLBW) infants, has a significant overall mortality [3], and a high risk of long-term neurodevelopmental sequelae [4].
Beyond an immature skin-mucosal barrier and immune response, other well-recognized risk factors for LOS include long-term use of invasive interventions, failure of early enteral feeding with breast milk, prolonged duration of parenteral nutrition, hospitalization, surgery, and underlying respiratory and cardiovascular diseases [2].
Growing evidence supports the key role of a healthy gut microbiota in promoting and maintaining a balanced immune response and in the establishment of the gut barrier in the immediate postnatal life [5]. However, in preterm infants, the development of the microbial community is disrupted by events related to prematurity: Mode of delivery, antenatal and postnatal use of antibiotics, minimal exposure to maternal flora, and low intake of breast milk [6]. Such disruption, called dysbiosis, results in an altered barrier and immune function and an imbalance between pro-and anti-inflammatory responses, and has been associated with necrotizing enterocolitis (NEC) and LOS [7,8].
Probiotics, defined as live micro-organisms that confer health benefits to the host through an interaction with gut microbiota and immune function when administered at adequate doses [9], have been proposed as potential tools to prevent NEC and LOS [10].
Updated meta-analyses confirm the benefits of probiotics in reducing the risk of NEC [11,12], the time to achieve full enteral feeding [13,14], and the risk of LOS [15,16] in preterm infants. However, most of these meta-analyses fail to explore the role of probiotics in deeper detail, and do not provide specific recommendations regarding which probiotic strain or mixture of strains should be used, and which population would benefit most from the use of probiotics.
Gut colonization in human milk (HM)-fed preterm infants is different from that of formula-fed infants [17]. HM provides nutrients, prebiotic carbohydrates, endogenous probiotics, and a variety of bioactive factors that exert beneficial effects directly and indirectly on host-gut microbiota interactions [18]. Recent data suggest that probiotic efficacy might be dependent upon the type of feeding; specifically, only preterm infants receiving HM would benefit from probiotic use in terms of a lower risk of NEC [19] and a reduction in the time needed to achieve full enteral feeding [13]. Furthermore, in vitro studies have shown that the growth of some probiotic species is enhanced in the presence of HM oligosaccharides (HMOs) [20,21]. Despite these suggestions, however, only a few randomized controlled trials (RCTs) report the type of feeding in infants given probiotics; and also for this reason, meta-analyses are unable to make any consideration about the influence of type of feeding in reducing adverse outcomes, such as NEC or LOS, in preterm infants receiving probiotics [13,16].
The aim of this systematic review and meta-analysis is thus to evaluate the effect of probiotics for the prevention of LOS in preterm infants according to type of feeding (exclusive HM vs. exclusive formula or mixed feeding).

Literature Search
The study protocol was designed by the members of the Task Force on Probiotics of the Italian Society of Neonatology. A systematic review of published studies reporting the use of probiotics for the prevention of LOS in preterm infants, according to type of feeding, was performed in accordance with PRISMA guidelines [22].
The characteristics of the studies included in the systematic review were the following: Randomized and quasi-randomized controlled trials involving preterm infants (gestational age (GA) <37 weeks) who had received, within one month of age, any probiotic compared to placebo or no treatment. The outcome of interest was culture-proven LOS, defined as the presence of a positive blood or cerebrospinal fluid culture taken >72 h after birth.
PubMed (http://www.ncbi.nlm.nih.gov/pubmed/), the Cochrane Library (http://www. cochranelibrary.com/) and Embase (http://www.embase.com/) were interrogated for studies published before 28 October 2016. The following string was used to perform the PubMed search: ((infant OR infants) OR (neonate OR neonates) OR (newborn OR newborns) AND (septi* OR sepsi* OR sepsis) OR (bacterial infect* OR bacterial infections (MH)) AND (probiotic OR probiotics OR pro-biotic OR pro-biotics)) NOT (animals (MH) NOT humans (MH)). The string was built up by combining all the terms related to LOS and probiotics, using PubMed MeSH terms, free-text words, and their combinations through the most proper Boolean operators, in order to be as comprehensive as possible. Similar criteria were used for searching the Cochrane Library and Embase. The review was restricted to English-written studies involving human subjects.
Luca Maggio (LM), Giovanni Barone (GB), Arianna Aceti (AA), and Isadora Beghetti (IB) performed the literature search. Potentially eligible studies were identified from the abstracts; the full texts of relevant studies were assessed for inclusion and their reference lists were searched for additional studies.

Data Extraction and Meta-Analysis
Study details (population, characteristics of probiotic and placebo, type of feeding, and outcome assessment) were evaluated independently by LM, GB, AA, and IB, and checked by Davide Gori (DG). Study quality was evaluated independently by AA, IB, and DG using the risk of bias tool as proposed by the Cochrane collaboration (Chapter 8 of the Cochrane Handbook of Systematic Reviews) [23]. In addition, an assessment of the quality of evidence using the GRADE working group approach was performed [23]. The evaluation was carried out by DG following Chapter 12 of the Cochrane Handbook [23] and classifying the evidence as high, moderate, low, and very low (as suggested by the GRADE Working Group) [24].
When outcome data were not reported according to type of feeding, the corresponding authors of the papers were contacted by email and were asked to provide separate data for LOS incidence in infants receiving probiotics vs. placebo according to type of feeding (exclusive HM vs. exclusive formula or mixed feeding). If the corresponding author was unable to provide these data or did not reply to the email, the paper was excluded from the meta-analysis.
The association between probiotic use and LOS was evaluated by a meta-analysis conducted by AA, IB, and DG using the RevMan software (version 5.3, downloaded on 1 November 2016 from the Cochrane website: http://tech.cochrane.org/revman/download). Risk ratio (RR) was calculated using the Mantel-Haenszel method and reported with a 95% confidence interval (CI). A fixed-effect model was used for the analyses. Heterogeneity was assessed using the χ 2 test and I 2 statistic: If significant heterogeneity was found (p < 0.05 from the χ 2 test) or the number of studies was lower than five, a random-effects model was used instead [23].
The results of the meta-analysis were presented using forest plots, while a funnel plot was used for investigating publication bias.

Literature Search
The number of potentially relevant papers identified through the literature search was 2713 (1401 in PubMed, 83 in the Cochrane Library, and 1229 in Embase).
As shown in Figure 1, 68 papers met the inclusion criteria (35 in PubMed, 13 in the Cochrane Library search, and 20 in Embase). Four additional studies were identified by a manual search of the reference lists of included studies. Among these 72 studies, 32 were excluded as they were duplicates retrieved by at least two search engines. Three studies were excluded after examining the full texts: One study included both term and preterm infants [25], one study reported supplementation with probiotic plus bovine lactoferrin [26], and one study was not written in English [27]. Finally, 37 studies were eligible for the systematic review . Details of the included studies are reported in Table 1; excluded studies are described in Table 2.
For each included study, the LOS rate in the probiotic and in the placebo/control group is reported in Table 3. The study by Dutta et al. [37] was reported three times, as it included three groups of patients supplemented with a probiotic given at three different doses. Data from the study of Hays et al. [39] were reported three times because three different interventions (Bifidobacterium lactis alone, Bifidobacterium longum alone, and B. lactis plus B. longum) were evaluated. The study by Romeo et al. [53] was reported twice, as it compared two different probiotics to placebo (Lactobacillus reuteri ATCC 55730 and Lactobacillus rhamnosus ATCC 53103), and the one by Tewari et al. [62] was reported twice because its participants were stratified as very preterm and extremely preterm.

Probiotic and LOS: Overall Population
Overall, data from 5868 infants (2934 in the probiotic group and 2934 in the control group) were evaluated. Regardless of type of feeding, fewer infants in the probiotic group developed LOS compared to infants in the control group (399 (13.60%) vs. 506 (17.24%), respectively). Probiotic Finally, 37 studies were eligible for the systematic review . Details of the included studies are reported in Table 1; excluded studies are described in Table 2.  For each included study, the LOS rate in the probiotic and in the placebo/control group is reported in Table 3. The study by Dutta et al. [37] was reported three times, as it included three groups of patients supplemented with a probiotic given at three different doses. Data from the study of Hays et al. [39] were reported three times because three different interventions (Bifidobacterium lactis alone, Bifidobacterium longum alone, and B. lactis plus B. longum) were evaluated. The study by Romeo et al. [53] was reported twice, as it compared two different probiotics to placebo (Lactobacillus reuteri ATCC 55730 and Lactobacillus rhamnosus ATCC 53103), and the one by Tewari et al. [62] was reported twice because its participants were stratified as very preterm and extremely preterm. Table 3. Incidence of late-onset sepsis (LOS) in infants treated with probiotics and in control.

Author, Year
Previous LOS Rate

Probiotic and LOS: Overall Population
Overall, data from 5868 infants (2934 in the probiotic group and 2934 in the control group) were evaluated. Regardless of type of feeding, fewer infants in the probiotic group developed LOS compared to infants in the control group (399 (13.60%) vs. 506 (17.24%), respectively). Probiotic supplementation resulted in a significantly lower incidence of LOS (RR 0.79 (0.71-0.88), p < 0.0001; Figure 2a). Number needed to treat was 28. In other words, 28 infants would need to receive probiotic supplementation in order to prevent one additional case of LOS. The funnel plot did not show any clear asymmetry (Figure 2b).

Probiotic and LOS According to Type of Feeding
The data were then analyzed according to type of feeding (exclusive HM, exclusive formula, or mixed feeding).

Probiotic and LOS According to Type of Feeding
The data were then analyzed according to type of feeding (exclusive HM, exclusive formula, or mixed feeding).
Only two studies reported specific data on LOS in ELBW infants. One study [28] included only ELBW infants, who received exclusive HM or mixed feeding; the other one [62] recruited both VLBW and ELBW infants, who were exclusively HM-fed. In these studies, probiotic supplementation did not show any significant benefit in terms of LOS compared to a placebo.
Four studies [46,49,50,52] reported data for 175 HM-fed infants who received a single-strain Lactobacillus probiotic (91 infants) vs. placebo/no treatment (84 infants). No difference between groups in the incidence of LOS was documented (RR 0.87 (95% CI 0.58-1.32); p = 0.63; I 2 = 0%; random effects model). Lactobacillus strains differed among studies: Lactobacillus rhamnosus was used in two studies [46,50] and Lactobacillus reuteri in two studies [49,52]. Lactobacillus sporogenes was used in one study [58], showing no differences between groups in LOS incidence; this latter study was not included in the pooled analysis, as L. sporogenes is a species which has not found international recognition, shows characteristics of both genera Lactobacillus and Bacillus, and its strain should be better classified as Bacillus coagulans [65].
Three studies [36,43,51] reported data for 334 HM-fed infants who received a single-strain Bifidobacterium probiotic (174 infants) vs. placebo/no treatment (160 infants). No difference between groups in the incidence of LOS was documented (RR 1.23 (95% CI 0.70-2.18); p = 0.47; I 2 = 0%; random effects model). Bifidobacterium strains differed among studies: Bifidobacterium breve was used in two studies [43,51] and Bifidobacterium lactis in one study [36]. In order to examine in deeper detail the effect of probiotics in HM-fed infants, sub-meta-analyses restricted according to population and probiotic characteristics, as well as study quality, were performed.
Only two studies reported specific data on LOS in ELBW infants. One study [28] included only ELBW infants, who received exclusive HM or mixed feeding; the other one [62] recruited both VLBW and ELBW infants, who were exclusively HM-fed. In these studies, probiotic supplementation did not show any significant benefit in terms of LOS compared to a placebo.
Four studies [46,49,50,52] reported data for 175 HM-fed infants who received a single-strain Lactobacillus probiotic (91 infants) vs. placebo/no treatment (84 infants). No difference between groups in the incidence of LOS was documented (RR 0.87 (95% CI 0.58-1.32); p = 0.63; I 2 = 0%; random effects model). Lactobacillus strains differed among studies: Lactobacillus rhamnosus was used in two studies [46,50] and Lactobacillus reuteri in two studies [49,52]. Lactobacillus sporogenes was used in one study [58], showing no differences between groups in LOS incidence; this latter study was not included in the pooled analysis, as L. sporogenes is a species which has not found international recognition, shows characteristics of both genera Lactobacillus and Bacillus, and its strain should be better classified as Bacillus coagulans [65].
Saccharomyces boulardii was used in one study [59], as well as Bacillus clausii [62]: None of these studies showed a significant difference between infants treated with probiotics and controls in the incidence of LOS.

Methodological Study Quality
The quality assessment of the studies included in the meta-analysis according to the risk of bias tool as proposed by the Cochrane collaboration is shown in Figure 6. The last column of the Figure  also shows the assessment of the body of evidence using the GRADE working group approach.
Nutrients 2017, 9, 904 8 of 14 Saccharomyces boulardii was used in one study [59], as well as Bacillus clausii [62]: None of these studies showed a significant difference between infants treated with probiotics and controls in the incidence of LOS.

Methodological Study Quality
The quality assessment of the studies included in the meta-analysis according to the risk of bias tool as proposed by the Cochrane collaboration is shown in Figure 6. The last column of the Figure  also shows the assessment of the body of evidence using the GRADE working group approach.
Following a methodology similar to that used in the meta-analysis by Rao et al. [15], we conducted a sensitivity analysis including only studies which had a low risk of bias in both random sequence generation and allocation concealment.
Sixteen studies [30][31][32]36,38,39,44,46,[50][51][52][57][58][59][60]62] were included and reported data for 4628 infants (2306 in the probiotic and 2322 in the control group). The results were similar to those of the overall meta-analysis: LOS occurred less frequently in infants receiving probiotics than in controls (309 (13.4%) infants vs. 366 (15.76%)) with an RR of 0.85 (95% CI 0.75-0.97; p = 0.02; I 2 = 0%; fixed effect model). Figure 6. Evaluation of the quality of the studies included in the meta-analysis according to the risk of bias tool as proposed by the Cochrane collaboration (red represents a high risk of bias, yellow an unclear risk of bias and green a low risk of bias). In addition, the last column shows the assessment an assessment of the body of evidence using the GRADE working group approach. Evaluation of the quality of the studies included in the meta-analysis according to the risk of bias tool as proposed by the Cochrane collaboration (red represents a high risk of bias, yellow an unclear risk of bias and green a low risk of bias). In addition, the last column shows the assessment an assessment of the body of evidence using the GRADE working group approach.

Discussion
In line with the results of previous papers [15,16], the present meta-analysis showed an overall benefit of probiotic supplementation for the prevention of LOS in preterm infants. However, when data were analyzed according to type of feeding, the beneficial effect of probiotics in reducing LOS was confirmed only in exclusively HM-fed preterm and VLBW infants, but not in infants receiving formula. Statistical heterogeneity among studies was almost absent and a low risk of publication bias was documented.
Two recent meta-analyses investigating the effect of probiotic supplementation on LOS in preterm infants reported an overall decrease in the risk of LOS in infants receiving probiotics compared to controls [15,16]. The studies included in the meta-analyses by Rao [15] and Zhang [16] are almost the same as those included in our updated systematic review; in the majority of the studies, both HM-and formula-fed infants were recruited, but no detailed data on the relationship between type of feeding and outcome were published.
Several data suggest that the impact of the type of feeding on clinical outcome in preterm infants is likely to be relevant [66]: It has been previously shown that HM feeding, per se, is associated with a reduction of the risk of developing LOS [67] and with a shorter time to achieve full enteral feeding in VLBW infants [68]. In addition, the use of probiotics in HM-fed, but not in formula-fed, infants appears to be related to a lower risk of NEC [19] and an earlier achievement of full enteral feeding [13].
It is plausible that the effect of probiotics on clinical outcomes could be mediated by HM properties [69]; actually, several HM components, including prebiotic HMOs, growth factors, immunological factors, and probiotic bacteria, can drive the establishment of a beneficial gut microbiota. In addition, HM can constitute the ideal soil for exogenous probiotics and promote a more effective crosstalk among probiotics, gut microbiota, and the developing immune system.
According to the latest recommendations, all preterm infants should receive exclusive HM; OMM is the best nutritional choice, and pasteurized DHM should be preferred to formula when OMM is not available or is contraindicated [70]. However, providing an exclusive HM diet to preterm infants presents a variety of challenges related to the prematurity itself and to hospitalization [71]. The term "exclusive HM feeding" may cover a range of feeding practices beyond direct breastfeeding, such as the use of fresh vs. frozen expressed breast milk given by bottle or tube feeding, the addition of HM fortifiers, and a variable duration of exclusive HM feeding. As described for pasteurization [72], some of these interventions might affect the nutritional and non-nutritional components of HM. In this perspective, the beneficial effect of probiotic supplementation in exclusively HM-fed infants might be related to a synergic action exerted by exogenous probiotics together with the prebiotic components of HM, which could partially restore the symbiotic potential of breast milk.
The data about exclusively HM-fed infants were analyzed according to population and probiotic characteristics in order to evaluate which preterm infants would benefit more from probiotic use and which probiotic strain or mixture of strains would be more beneficial. While there is evidence that probiotics are effective in reducing LOS in VLBW infants, no definite conclusion could be drawn for ELBW infants, as only two studies reported specific data on LOS in these infants, who remain the highest-risk and most vulnerable population.
The currently available literature does not provide a definite recommendation on which probiotic strain or mixture would be more effective in reducing LOS. In the 25 included studies, different probiotic strains and mixtures were used. Consistently with previous papers [12,16,73], our meta-analysis indicated that a mixture of different probiotic strains might be more effective in reducing LOS in exclusively HM-fed preterm infants. A possible explanation for this finding is that a probiotic mixture would provide a better ecological barrier and a more diverse immunological stimulation than a single strain.
The possible limitations of the present meta-analysis should be taken into consideration. Thirty-seven studies were potentially eligible for the meta-analysis, but only 25 studies provided separated data according to feeding type. In addition, infants' classification according to feeding type was not homogeneous across studies, and the meta-analysis had to rely on unpublished information provided by the authors themselves. Finally, although no statistical heterogeneity was found, the characteristics of probiotic administration (dose, duration, time of initiation, and probiotic micro-organisms) differed among the included studies.
More importantly, no separate data for OMM-fed and DHM-fed infants were available; as a result, it was not possible to clarify whether the "synergic" effect of HM and probiotics applies to both OMM and DHM. It remains also unclear whether HM feeding, either OMM or DHM, has a "dose and time-dependent" effect on probiotic supplementation, as reported for outcomes such as NEC [66].
Probiotics appear to be generally safe, but it has to be acknowledged that there are some reports about the occurrence of sepsis in preterm newborns potentially linked to probiotic supplementation [74]. None of the studies included in the systematic review reported any side effect related to probiotic administration.

Conclusions
According to the results of the present meta-analysis, probiotic supplementation reduces the risk of LOS in exclusively HM-fed preterm infants. An exclusive HM diet should be the gold standard for all preterm, VLBW infants. Since direct breastfeeding is almost impossible in this population, it is likely that manipulations of HM, including pasteurization, refrigeration, and administration by tube or bottle, could affect HM bioactive properties; in this context, the administration of exogenous probiotics could help in restoring, at least partially, HM symbiotic properties.
Future research should be aimed at clarifying the relationship between feeding practices and probiotic supplementation, and at addressing the choice of the most effective probiotic products to be used in exclusively HM-fed infants.
Author Contributions: All the authors approved the submission of this version of the manuscript and take full responsibility for the manuscript. Specifically, all the authors, as part of the Task Force on Probiotics of the Italian Society of Neonatology, conceived and designed the study protocol. L.M., G.B., A.A., and I.B. performed the literature search and assessed study details, which were checked by D.G., A.A., and I.B., and D.G. evaluated study quality and performed the meta-analysis. A.A. and I.B. wrote the first draft of the paper, which was critically revised by all the other authors.

Conflicts of Interest:
The authors declare no conflict of interest.