Association of Anemia and Necrotizing Enterocolitis in Neonates: Systematic review and meta-analysis of transfusion thresholds, erythropoietin trials and observational studies

Abstract

Background:

Meta-analyses of observational studies evaluating association between blood transfusion and necrotizing enterocolitis (NEC) have shown mixed results and an independent association between anemia and NEC requires further evaluation.

Objective:

To investigate the association of NEC with anemia in preterm infants via meta-analysis of (1) randomized controlled trials (RCTs) evaluating transfusion thresholds and recombinant erythropoietin (rEpo)/darbepoetin and (2) case-control studies comparing hematocrit levels between NEC cases and controls

Data sources:

PubMed, Embase, Scopus, and the Cochrane Database were searched in August 2025, and the results were supplemented by other sources.

Study selection:

(1) Transfusion threshold and rEpo RCTs evaluating NEC rates in preterm infants with lower hemoglobin (Hb) levels compared with higher Hb levels. (2) Case-control studies comparing hematocrit levels between NEC cases and controls.

Primary outcome:

The incidence of any stage NEC in infants with low versus high Hb group.

Results:

In RCTs evaluating transfusion threshold (n=4, 3307 participants), showed no difference in NEC incidence between low versus high Hb group [OR 1.14 (CI 0.89, 1.46), p 0.647]. RCTs evaluating rEPO vs placebo (n= 11, 4173 participants), showed higher incidence of NEC in low versus high Hb group [OR 1.43 (CI 1.13, 1.80), p 0.003]. Meta-analysis of case-control studies (n =3, 1016 participants) showed association of lower hematocrit levels in NEC cases compared with controls (Standardized Mean Difference −0.28; CI −0.56, −0.002; p 0.048).

Conclusion:

Our analysis suggests an association between NEC and lower hemoglobin levels. Well-designed and adequately powered prospective studies are required to confirm our findings.

Introduction

Necrotizing enterocolitis (NEC) affects 5-10% of preterm infants and remains a leading cause for surgical intervention, mortality, and increased health care costs [1, 2]. The association between blood transfusions and the subsequent occurrence of NEC has been noted in several retrospective studies, though the results of meta-analyses are mixed [3, 4]. Observational studies show anemia, rather than transfusion, may be an independent risk factor for NEC [5–7]. Given the frequency of packed red blood cell transfusions and prevalence of anemia in preterm infants [8], understanding severe anemia as an independent risk factor for NEC is essential.

Blood transfusion and the use of erythropoietin-stimulating agents play a significant role in the management of anemia in preterm infants. The primary mechanism of action of recombinant human erythropoietin (rEpo) or its derivative, darbepoetin, is by stimulating erythropoiesis and increasing hemoglobin (Hb) and hematocrit (Hct) levels. Trials reviewing rEpo vs placebo or various transfusion threshold allows comparison between two arms with high and low Hb levels in preterm infants. These studies have shown a reduced rate of NEC in groups receiving rEpo or darbepoetin [9]. Trials evaluating transfusion thresholds have shown a trend towards higher NEC rates in the restrictive group [10, 11]. Additionally, few case-control and observational studies have directly compared Hct levels prior to onset of NEC with matched control with mixed results in terms of difference of Hct levels predisposing to NEC [5, 7].

In our previous meta-analysis, we reported no association between PRBC transfusions and increased risk of NEC (OR: 0.96; 95% CI: 0.53–1.71; P = 0.88) [12]. However, the study did not assess the impact of anemia on NEC. The varied results on the relationship between anemia, RBC transfusion, and NEC suggest the need for a comprehensive analysis of all available evidence. Hence, we conducted a systematic review to assess the association between anemia and the development of NEC in preterm infants.

Objective

To investigate the association of NEC with anemia in preterm infants via meta-analysis of randomized controlled trials (RCTs) evaluating transfusion thresholds and recombinant erythropoietin (rEpo)/darbepoetin usage and case-control studies comparing hematocrit levels before onset of NEC between NEC cases and controls.

Material and Methods

This systematic review was conducted using data from published studies where ethical approval and informed consent were obtained by the original investigators. Therefore, no separate ethical approval was required. The study followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) reporting guideline.

Selection Criteria

Participants:

Preterm infants with birth weight <1500 g or born at 34 weeks of gestational age or earlier were included.

Type of studies:

RCTs and case-control studies were eligible for inclusion. Reviews and commentaries were excluded. Owing to differences in methodologies between RCTs and case-control studies different comparison groups and outcomes were evaluated for both.

RCT:

Intervention and comparison:

1) Liberal versus restrictive transfusion thresholds; and 2) the administration of early darbepoetin or rEpo vs placebo.

Outcomes:

The incidence of NEC between the two groups: “High Hb” (liberal transfusion threshold or rEpo arm) and “Low Hb” (restrictive transfusion threshold or placebo arm)

Case-control studies:

Studies evaluating NEC cases compared with controls were included if the specified outcomes of 1) lowest Hb or Hct levels prior to onset of NEC; or 2) Hb and Hct levels within 96 hours prior to onset of NEC were reported.

Intervention and comparison:

Comparison of Hct levels (lowest level or within 96 hours of onset) before onset of NEC cases compared to matched controls.

Few studies have evaluated all NEC cases, whereas others have evaluated only cases of transfusion-associated NEC (TANEC). Additionally, the timing of Hct measurements relative to the onset of NEC varied across studies. Hence, to avoid pooling studies with heterogeneous intervention and comparison groups, we grouped case-control studies into the following three groups for pooled analyses:

1) Comparison of Hct levels among all NEC cases versus matched controls irrespective of blood transfusion status; 2) Comparison of Hct levels among NEC cases versus matched controls for infants who received packed red blood cell transfusion (or TANEC); and 3) Comparison of the lowest Hct levels between NEC cases versus matched controls.

Outcomes:

Standardized mean difference (SMD) in Hct levels between NEC cases versus controls.

Information sources and Search strategy:

PubMed, Medline, Embase, SCOPUS, Cochrane Library databases and www.clinicaltrials.gov, were systematically searched. We used following search terms in various combinations: 1) Population: “preterm infant”, “premature”, “preterm”, “neonate(s)”, “newborn(s)”; 2) Intervention: “blood transfusion”, “anemia”, “liberal”, and “erythropoietin”; 3) Outcome: “necrotizing enterocolitis”, and 4) Publication type: “random”, “case-control”. Detailed search strategy is summarized in supplemental file 1. A search was conducted on August 1, 2025, to confirm adequate inclusion of all studies. The reference lists of identified studies and review articles were searched to identify additional eligible studies. Owing to resource limitation and to avoid misinterpretation of results in setting of absence of professional translators, only studies published in English as primary language were included.

Data collection process

All potential RCTs and case control studies were reviewed by three independent investigators (M.R., C.M., D.V.) who selected studies and extracted data using predefined and itemized data collection sheets that included information on study design, population, exposure, outcome, and results. The results were combined with cross-referencing of previous systematic reviews and meta-analyses [13–16]. Data verification was performed by two independent investigators after collection (D.V & P.G)

Quality assessment of included studies

RCTs:

Risk of bias assessment of the selected studies was performed according to Cochrane’s guidelines. The risks of selection bias (quality of randomization, allocation concealment), performance bias (blinding of intervention), detection bias (blinding of outcome assessment), attrition bias (completeness of follow-up), selective reporting (reporting bias), and other potential biases were independently examined by 2 study authors (D.V & P.G) and categorized as low risk, high risk, or unclear risk.

Case-control studies:

The quality of each study was assessed using the well-established Newcastle-Ottawa Scale. Three main criteria were assessed, including participant selection and representativeness, comparability of study groups, and assessment of outcome or exposure. The quality score was based on a “star” system. The maximum score for each domain is 4, 2, and 3 points, respectively. A total score ≤3 indicates low methodological quality.

Statistical analysis

For pooling of data from RCT, number of cases with NEC and total number of patients in each arm were recorded. An odds ratio (OR) with corresponding 95% CIs were derived from this data. For case-control studies, a mean and standard deviation (SD) of hematocrit levels were recorded and standardized mean differences (SMD) with 95% confidence intervals (CIs) calculated. For uniformity between studies, when indicated a conversion factor of 3 was used for conversion of Hct and Hb levels [17]. Discrepancies during data extraction were resolved by joint discussion. All risk estimates, including risk ratios and hazard ratios, when applicable, were log-transformed to determine the pooled effect estimates using the generic inverse variance method. Heterogeneity was evaluated using the DerSimonian-Laird test and the I² statistic, with I² >50% representing substantial heterogeneity. Between-study variance (τ²) was also reported to describe variability in true effect sizes across studies. Forest plots were generated for each comparison to display individual and pooled effect sizes with 95% CIs. The pooled effect sizes for subgroup analyses were computed independently, with p-values <0.05 considered statistically significant. The presence of publication bias or small study bias was explored using a funnel plot and Egger’s test. All statistical analyses were performed using SPSS software for Windows (Version 29. IBM Corp, Armonk, NY, United States).

Results

A total of 1038 records were identified through database search. Of these, 299 were excluded for duplicate records and 653 were excluded as they were not relevant to our study question. On screening, the 86 total reports were sought and assessed for eligibility. An additional 5 studies were identified based on citations from other sources [13, 14, 16] (Figure 1).

Figure 1.

PRISMA flow diagram

Included studies

A total of 15 RCTs with statistical difference in Hb or Hct levels or RBC transfusion exposure were included. Four RCTs comparing blood transfusion thresholds [10, 11, 18, 19], and eleven RCTs comparing rEpo versus placebo [20–30] were included (Table 1). The differences in transfusion thresholds between these four studies are summarized in Supplementary Table 1.

Table 1:

Characteristics of included Randomized controlled trials

Trial	Year	Setting	N.	Participants	intervention	Hb/HCT difference	Number of transfusions	NEC stage
RCT comparing RBC transfusion threshold.
Kirplani [19]	2020	41 NICUs in the USA	1824	BW <1000 g [n, High Hb =911, Low Hb 913, Mean GA High Hb 25.9 w (1.5), low Hb group 25.9w (1.5)]	liberal vs restrictive transfusion	Liberal vs restrictive Hb: Week 2: 13 g/dL (1.4) vs 11.5 g/dL (1.4) Week 3: 12.3 g/dL (1.4) vs 10.9 g/dL (1.5) Week 4: 11.5 g/dL (1.3) vs 10.2 g/dL (1.3)	Different	Stage 2 or higher
Franz [10]	2020	36 level III/IV NICUs in Europe	1013	BW 400 – 999 g [n, High Hb =491, Low Hb =518, Mean GA High Hb 26.1 w (2.0), low Hb group 26.4 w (2.0)]	liberal vs restrictive transfusion	Liberal vs restrictive Hct: Week 2: 41.9 % (5.2) vs 39.5 % (5.5) Week 3: 39.5 % (4.8) vs 36.2 % (5.0) Week 4: 37.6 % (4.6) vs 34.3 % (4.8)	Different	Stage 2a or higher
Kirplani [11]	2006	10 NICUs in Canada, USA and Australia	451	BW <1000 g, [n, High Hb = 228, Low Hb = 223, Mean GA High Hb 26 w (2.0), low Hb group 26 w (2.0)]	Low or High Hb threshold	Liberal vs restrictive Hb: Week 2: 13.1 g/dL (1.3) vs 11.9 g/dL (1.5) Week 3: 12.0 g/dL (1.3) vs 10.9 g/dL (1.5) Week 4: 11.2 g/dL (1.3) vs 10.1 g/dL (1.2)	Different	Stage 2 or higher
Chen [18]	2009	Single center in Taiwan	36	BW <1500g, [n, High Hb = 17, Low Hb = 19, Mean GA High Hb 29.1 w (2.7), low Hb group 29.1 w (3.0)]	Liberal vs restrictive transfusion	No difference pre-transfusion	Volume of transfusion was different	unknown
RCT for comparing rEPO vs placebo
Juul [21]	2020	30 hospitals across USA	741	<28 w, [n, High Hb = 476, Low Hb = 460, Mean GA High Hb 26 w (1.2), low Hb group 25.8 w (1.1)]	rEpo 1000 U/kg every 48 hours for 6 doses then 400 U/kg 3/w sc until 32 w PMA	Not available	Cumulative volume of transfusion, and number of transfusions was different	Stage 2b or higher
Ohls [25]	2001	Multisite USA	290	401 – 1250 g, <32 w [n, High Hb = 146, Low Hb = 144]	rEpo 400 U/kg 3 times/w until 35 w PMA	Weekly Hct was different	No difference in transfusion volume	Stage 2 or higher
Ohls [24]	2014	3 sites in USA	66	500 – 1250 g bw [n, High Hb = 33, Low Hb = 33, Mean GA High Hb 27.8 w (1.9), low Hb group 27.3w (1.8)]	rEpo 400 U/kg 3 times/w until 35 w PMA	Hct was different between rEpo and placebo group	Transfusion per patient and volume of transfusion was different	Stage 2 or higher
Song [26]	2016	Single center NICU in China	613	<32 w GA [n, High Hb = 366, Low Hb = 377, Mean GA High Hb 30.39 w (1.38), low Hb group 30.40 w (1.46)]	rEpo 500 IU/kg Iv every other day for 2w	Hb and Hct different	Number of transfusions per patient was different	All stages including I
Wang [28]	2020	4 NICU in China	1285	< 32 w GA [n, High Hb = 641, Low Hb = 644, Mean GA High Hb 29.7w, low Hb group 30.0w]	rEpo 500 IU/kg IV every other day for 2w	Number of infants with HB <9 g/dl was lower in rEpo compared to control	Number of transfusions per patient was different	Stage 2 and higher
Yeo [30]	2001	1 NICU in Singapore	100 (28)	<1500 g, < 33 w [n, High Hb = 14/50, Low Hb = 14/50, Mean GA High Hb 28.2 w(1.9), low Hb group 28.3 w(2.1)] *n=28 subgroup included in analysis	rEpo 250 U/kg/dose sc 3 times a week 5 – 40 days	unknown	no difference, Reduced only in <1000 g subgroup	Radiologic, operative or postmortem evidence
Turker [27]	2005	Single center Level III NICU in Turkey	93 (30)	<1500 g, < 33 w [n, High Hb = 15/42, Low Hb = 15/51, Mean GA High HB 28w, low HB group 27w] *n=30 subgroup included in analysis	250 U/kg/dose sc 3 times / w 5-40 days	No difference at 3 months	No difference, reduced in <1000 g subgroup	Stage >2
Maier [23]	1994	12 centers in 6 European countries	241	750 - 1500 g, <34 w [n, High Hb = 120, Low Hb = 121, Mean GA High Hb 29w, low Hb group 29w]	250 U/kg/dose IV/sc 3 times/w 3-42 days	Not statistically different. But Hct never fell below 32% was higher in rEpo compared to Control	Volume of transfusion, and mean transfusions per patient was lower	unknown
Maier [22]	2002	14 centers 4 European countries	145	<1000 g infants, <30 w [n, High HB = 74, Low Hb = 71, Mean GA High Hb 26w, low Hb group 27w]	250 U/kg/dose IV/sc 3 times/w 3-68 days	Lower but not statistically significant	Volume of transfusion was lower	Walsh and Kliegman criteria
Wellmann [29]	2022	8 Swiss and Austrian NICU	121	<32 w, <1500 g [n, High Hb = 60, Low Hb = 61, Mean GA High Hb 27w, low Hb group 26.1w]	2000 U/kg every day for 3 days, after 10 days and after 17 days. Total of 5 doses	Unknown	No. of transfusions decreased	Stage 2 or higher
Arif [20]	2005	Single center in Turkey	292	<33 w, <1500 g [n, High Hb = 142, Low Hb = 150, Mean GA High Hb 29.7 w(1.3), low Hb group 30.1 w(1.1)]	200 U/kg sc 2/w from 7 – 42 days	Different at 4 w and 7 w	unknown	unknown

BW birth weight, d days, GA gestational age, Hb hemoglobin, Hct hematocrit, IV intravenous, NEC necrotizing enterocolitis, NICU neonatal intensive care unit, rEpo erythropoietin, sc subcutaneous, TANEC transfusion-associated NEC, w weeks.

A total of five case-control studies with available Hb and Hct levels between the NEC group and the control group were included in this review (Table 2) [5, 7, 31–33]. Three studies, Singh et al. [5], Demirel et al. [31], and Song et al. [7] compared all NEC cases versus controls irrespective of blood transfusion status. Demirel et al. [31], Sood et al. [32], and Wan-huen et al. [33] compared NEC cases versus matched controls following packed red blood cell transfusion (or TANEC vs control). Song et al. [7], Demirel et al. [31], and Sood et al. [32], compared the lowest Hct levels between NEC cases and controls. Singh et al. was only study comparing Hct levels within 96 hours of onset of NEC [5].

Table 2:

Characteristics of included Case-Control studies

Study	Year	Country	Type	Patient population	Type of NEC	Timing of Hct and Outcome	Newcastle Ottawa
Singh [5]	2011	USA	Single center Case – control (matched)	Cases (n = 111; mean GA: 26.9 w; mean BW: 970 g) Controls (n = 222; mean GA: 27.3 w; mean BW: 1023 g)	Nec stage >2a	Within 96 hours of onset of NEC and matched controls based on day of diagnosis in index case *Difference was statistically significant	★ ★ ★ ☆ Selection ★ ☆ Comparability ★ ★ ☆ Exposure
Song [7]	2021	China	Single center- case -control (matched)	Cases (n = 166; mean GA: 30.3 w; mean BW: 1166 g) Controls (n = 166; mean GA: 30.3 w; mean BW: 1187 g)	NEC stage > 2a	Lowest Hct *NEC group associated with higher incidence of severe anemia [ cases: 28/166 (16.9%), control:13/166 (7.9%), p 0.012] and need for transfusion [cases 24/166 (14.5%), controls 9/166 (5.4%), p 0.006] *Hct difference was not statistically significant. Conversion factor of 3 was used to calculation of Hct from Hb	★ ★ ★ ☆ Selection ★ ☆ Comparability ★ ☆ ☆Exposure
Demirel [31]	2012	Turkey	Single center Case control (matched)	Total of 647 infants divided into 5 groups. Following subgroups were used for analysis 1) No transfusion groups Cases (n = 50; mean GA: 29.1 w; mean BW: 1100 g) Controls (n = 301; mean GA: 29.3 w; mean BW: 1188 g 2) TANEC vs transfusion but never developed NEC  Cases (n=15, mean GA=28.4 w, BW 1078 g) Controls (n=250, mean GA =28.8 w, Mean BW =1092 g)	Nec stage >2	1) lowest Hct 2) lowest pre transfusion Hct *Difference was not statistically significant	★ ★ ★ ☆ Selection ★ ☆ Comparability ★ ☆ ☆ Exposure
Sood [32]	2016	USA	Single center case-control study	Cases (n = 39; mean GA: 26.5 w; mean BW: 947 g) Controls (n = 588; mean GA: 28.6 w; mean BW: 1222 g)	NEC stage > 2a	Lowest Hct in first 15 days of life or before NEC *Difference was statistically significant	★ ★ ★ ☆ Selection ★ ☆ Comparability ★ ★ ☆ Exposure
Wan-Huen [33]	2013	USA	Single center Case-control (matched)	Cases (n = 49; mean GA: 26 w; mean BW: 821 g) Controls (n = 97; mean GA: 26 w; mean BW: 840 g)	Nec stage >2	pretransfusion Hct levels *Difference was not statistically significant	★ ★ ★ ☆ Selection ★ ☆ Comparability ★ ★ ☆ Exposure

BW birth weight, GA gestational age, Hct hematocrit, NEC necrotizing enterocolitis, TANEC transfusion-associated NEC, w weeks.

Excluded studies

For RCTs, 27 studies lacked data on NEC, and 9 studies were either follow-up studies or post-hoc analyses of included trials. Out of thirty-four case-control studies identified; 20 were excluded as they lacked data on Hb or Hct levels. Summary of other excluded studies with reason of exclusion is provided in Table 3.

Table 3:

Excluded studies and reason for exclusion

Studies	Reason for exclusion
Patel et al.	Case – control study: Hct level available only from the time of birth [Hct NEC case: 46.9% (4.1), control 49.2% (3.6)]
Bak et al.	Case – control study: Hct level available only from the time of birth [Hct NEC case: 44.6% (7.2), control 45% (7.7)]
Josephson et al.	Case – control study: Hct level available only from the time of birth [Hct NEC case: 43.5% (6.6), control 45.6% (8.1)]
Garg, Faraday et al.	Case- control study: Hct level available only for NEC group. Not for comparison group.
Cunningham et al., Janjindamai et al.	Case – Control study: Comparison between NEC cases with TANEC cases
Bell et al., Blank et al.	RCT: No incidence of NEC not available.
Fauchère et al. Natalucci et al.	RCT: focused on the use of early high-dose rEpo and its impact on neurodevelopmental outcomes, no statistical difference in Hct level documented (incidence of NEC: Fauchere: rEpo = 6/229, control 10/214; Natalucci: rEpo = 4/191, control 5/174)
Peltoniemi et al.	small RCT focusing on iron metabolism and outcomes, using rEpo in the first six days only. No statistical difference in Hct level documented (NEC: rEpo = 1/21, control 1/18)
Qiao et al.	RCT: No difference in Hct level or number of transfusions (reported zero NEC cases in either group)
Obladen et al.	RCT: No difference in Hct level or number of transfusions (NEC cases: rEpo 1/43, control 3/50)
El-Ganzoury et al.	RCT: was excluded because it used enteral EPO, and No difference in Hct level or number of transfusion (rEpo = 0/20, control 3/30)
Haiden et al.	RCT: No difference in Hct level or number of transfusions (rEpo 3/21, control 0/19)

References for these studies is summarized in supplementary table 2.

Risk of bias for RCTs.

Figure 2 summarizes the quality of evidence for all the RCTs. Three studies did not report allocation concealment (unclear risk) [20, 27, 30]. Liberal versus restrictive transfusion studies were blinded, except for the caregivers (unclear risk). One study did not report blinding (unclear risk) [18], and five studies lacked blinding (high risk) [20, 26–28, 30]. Seven studies were not registered (unclear risk) [18, 20, 22, 23, 25, 27, 30] and two studies were registered after enrollment was complete (high risk) [26, 28]. All other biases in the studies were of low risk.

Figure 2.

Risk of bias summary: Review the author’s judgments about each risk of bias item for each included study

Quality for included case-control studies

Table 1 describes the quality of studies, as evaluated by two investigators using the Newcastle-Ottawa scale. All studies were of moderate quality, with an intermediate risk of bias.

Publication bias and sensitivity analysis

Visual inspection of the funnel plots did not reveal any significant asymmetries in the distribution of effect sizes around the mean. Egger’s test also did not detect any statistical evidence of asymmetry, representing there is no evidence of small study effects or publication bias among the studies included in this meta-analysis (Figure 3)

Figure 3.

Funnel plot and Egger’s test for assessment of small study bias and publication bias.

3a) RCT rEpo vs placebo; 3b) RCT liberal (high Hb) and restrictive (low Hb) transfusion threshold; 3c) Case-control: Cases of NEC and matched control studies. 3d) Cases of TANEC and no transfusion studies.

P value for Egger’s test <0.1 was considered positive for publication bias.

Effects of Interventions:

Randomized control trials:

1). Liberal vs Restrictive packed red cell transfusion

Pooled analysis of RCT comparing blood transfusion thresholds found no difference in incidence of NEC between liberal/high Hb (8.5%) versus restrictive/low Hb (5.3%) groups [OR 1.14 (95% CI 0.89, 1.46; p = 0.297) (Figure 4a). There was no significant heterogeneity between the studies (τ² = 0, I² = 0 %). All four trials showed a trend toward a higher incidence of NEC in the restrictive group compared to the liberal group, but this difference was not statistically significant. Summary of differences in Hct/Hb levels at 2 – 4 weeks of age is described in table 1.

Figure 4.

Figure 4a. Forest plot of comparison: the incidence of NEC between restrictive (Low Hb) vs liberal (High Hb) transfusion threshold

Figure 4b. Forest plot of comparison: the incidence of NEC between placebo (Low Hb) vs rEpo (High Hb)

CI confidence interval, p <0.05 is statistically significant

2). Recombinant Erythropoietin vs placebo

Pooled analysis of RCT comparing rEpo versus placebo found 43% higher odds of developing NEC among the infants in placebo group (low Hb) compared to the rEpo (high Hb) group [placebo 200/2087 (9.6%), rEpo 145/2086 (6.9%), OR 1.43 (CI 1.13, 1.80), p 0.0027]. Additionally statistical methods indicated low- heterogeneity between studies (τ² = 0, I² = 1.2%) (Figure 4b).

Case control studies:

1). NEC cases versus matched controls irrespective of blood transfusion status

Analysis showed Hct levels prior to the onset of NEC, were significantly lower in NEC cases compared to matched controls [NEC cases: mean Hct 31.2 (SD ±5.5), control: mean Hct 33.0 (SD ± 7.2)], with a Standardized mean difference of Hct showing a small reduction of −0.281 (95% CI −0.560 to −0.002; p = 0.048). Statistical measures indicated considerable heterogeneity, however with a low τ² suggesting that the true underlying effect sizes are very close to one other (τ² = 0.045, I² = 75.9%) (Figure 5a).

Figure 5.

Forest plot of comparison: Hct levels in case-control

5a: Cases of NEC compared to matched controls, irrespective of blood transfusion status.

5b: Cases of TANEC compared to infants who received a transfusion, but never developed NEC

5c: Cases of NEC or TANEC compared to matched controls, comparing values of lowest Hct levels prior to onset of NEC or transfusion.

SMD = standard mean difference, CI confidence interval, p <0.05 is statistically significant

2). NEC cases versus matched controls of infants who received packed red blood cell transfusion.

A comparison of the pre-transfusion Hct levels between TANEC cases and control showed no significant difference between Hct levels [TANEC mean Hct 31.4 (SD ± 4.9), Control mean Hct 31.8 (SD ± 5.4), Standard Mean Difference 0.115 (95% CI −0.115 to 0.346); p = 0.326] (Figure 5b).

2). Comparison of lowest Hct levels between NEC cases versus matched controls.

Pooled standardized mean difference of Hct levels between NEC cases and control was −0.13 (95% CI −0.28, 0.02), p 0.0965). Moderate heterogeneity was noted with I² of 27.5%, but low τ² of <0.0001 (Figure 5c). Egger’s test for this pooled analysis showed no publication bias or small-study effects (intercept −0.318; 95% CI −2.007, 1.372; p 0.504). Further stratification of these studies into two group based on transfusion status, showed no difference between the group. Standardized mean difference of lowest Hct levels in prior to NEC cases vs. controls irrespective of transfusion status was −0.13 (95% CI −0.30, 0.05), p 0.151) (Supplementary Figure 1a). And standardized mean difference of lowest Hct levels between TANEC cases vs control was −0.07 (95% CI −0.68, 0.53), p 0.810) (Supplementary figure 1b)

Discussion:

Our systematic review provided evidence of association between NEC and lower hemoglobin or hematocrit levels in preterm infants. A higher incidence of NEC was reported in infants with lower Hb levels in RCTs evaluating rEPO use. Evaluation of RCTs of packed red blood cell transfusion thresholds did not show a difference in NEC incidence between different thresholds. A review of case-control studies showed that hematocrit levels prior to onset of NEC were lower in NEC cases compared to matched controls. However, this difference was not observed in cases of TANEC or in review of the lowest hematocrit levels before onset of NEC.

Our results were comparable to those reported by Wang et al. Their meta-analysis included RCTs comparing liberal versus restrictive blood transfusion thresholds and showed no significant difference in NEC incidence (RR 0.99, 95% CI 0.84, 1.16) [16]. Two previous meta-analyses evaluating early erythropoietin and placebo, by Ohlsson et al. and Ananthan et al. [13, 34] found a lower incidence of NEC in infants receiving rEPO compared with those receiving placebo [Ohlsson et al. [13] RR 0.69, 95% CI 0.52-0.91; Ananthan et al. [34] RR 0.75, 95% CI 0.64 – 0.88]. Our analysis of rEpo trials showed similar significant results after including three additional studies [16, 21, 29]. Additionally, few RCTs included in these prior meta-analyses had no statistical difference in Hct or Hb levels between rEpo and placebo arms. Despite including only those RCTs with significant difference of Hct or Hb levels after administration of rEpo.

In another systematic review of case-control studies, Zhao et al. reviewed risk factors associated with NEC [35]. Their analysis showed, receiving blood transfusion (OR 2.41; 95% CI 1.97, 2.95) and presence of severe anemia (OR 2.86; 95% CI 2.06, 3.99)] were associated with NEC cases compared to control. Limitation of this review is temporal relation between timing of NEC, severe anemia and blood transfusion. We attempted to address this factor by only including studies reviewing hematocrit levels prior to onset of NEC. And found a similar significant association in studies comparing all cases of NEC and control, however reviewing studies comparing TANEC cases or only lowest Hct level failed to show similar association.

Severe anemia in neonates is associated with impaired gastrointestinal perfusion in very low birth weight infants [36, 37]. This compromised perfusion and oxygenation predispose the intestinal mucosa to increased permeability and inflammation, which are linked to association of NEC [38]. In a mouse model, Mohankumar et al. demonstrated severe anemia caused a low-grade inflammatory state in the intestinal mucosal with macrophage infiltration, and subsequent RBC transfusions activated cells via TLR4 mediated mechanism to cause bowel injury [39]. These studies highlight plausible explanation of association between anemia and NEC.

The mechanisms of rEpo effects primarily involve stimulation of erythropoiesis raising Hct or Hb level. However, animal models suggest rEpo may have additional mechanism of action including reducing inflammation in intestinal mucosa, maintaining the intestinal barrier integrity by preventing loss of tight junction proteins and decreasing autophagy and apoptosis of intestinal mucosa [28]. Given these data, it is difficult to conclude that the benefits of rEpo in reducing the risk of NEC are solely mediated via anemia.

The strengths of our systematic review and meta-analysis include a large sample size, robust methodology, and minimal statistical heterogeneity among RCTs. Our review includes new RCTs that were not included previous reviews. The inclusion of case-control studies adds a vital dimension, knowing the benefits of assessing the evidence in totality. A separate stratified analyses are performed for case-control studies to address differences between the study population and the timing of Hct levels. The limitations of our study need to be acknowledged. The lack of data on the timing of Hct levels and development of NEC in case-control studies limits the generalizability of our findings. However, the availability of that data would not establish causality. Furthermore, ruling out reverse causality is difficult, given that NEC itself may lead to anemia [40, 41]. Additionally, our finding supports only an association, lacking specificity and dose-response criteria to support a causal relationship (Bradford Hill’s criteria). The rEpo dose and the liberal versus restrictive transfusion thresholds differed across the studies. It is unclear how rEpo dosing and differences in Hct levels between the groups may impact our findings. The heterogeneity of the severity of anemia across studies makes it difficult to comment on the effect of the severity of anemia on NEC. Heterogeneity in NEC staging criteria makes the association less clear. Lastly, grouping transfusion threshold and rEpo/darbepoetin trials combines studies “treating” anemia with those aimed at “preventing” anemia – two mechanistically distinct processes. Although we provide separate analyses, we cannot exclude the possibility that unique physiologic events are responsible for the associations.

Conclusion:

In summary, our analysis indicates that preterm infants with lower Hct or Hb levels in rEpo trials are associated with higher NEC incidence; however, similar differences in NEC incidence are not found across different transfusion thresholds. Pooled analysis of case-control studies showed that NEC cases were associated with lower hematocrit levels; however, a similar difference was not observed in transfusion-associated NEC. Overall quality of evidence was low, given multiple studies with high or unclear risk of bias. Further research is required to investigate this association and to clearly define transfusion thresholds and anemia, to identify preventive strategies that may reduce the incidence of NEC.

Abbreviations:

BW

birth weight

CI

confidence interval

d

days

GA

gestational age

Hb

hemoglobin

Hct

hematocrit

IV

intravenous

NEC

necrotizing enterocolitis

NICU

neonatal intensive care unit

NS

not significant

OR

odds ratio

RBC

red blood cell

RCT

randomized controlled trial

rEpo

recombinant erythropoietin

Sc

subcutaneous

SD

standard deviation

TANEC

transfusion-associated NEC

w

weeks