Background
Necrotizing enterocolitis (NEC) is the most common gastrointestinal emergency among premature infants [
1,
2], occurring in approximately 11% of those born < 29 weeks’ gestation [
3]. Case-fatality rates are as high as 50% for extremely low birth weight (ELBW) infants (≤ 1000 g at birth) who develop NEC [
4]. Survivors are at risk for substantial long-term complications including neurodevelopmental delay, nutritional deficit and failure to thrive [
3,
5]. Costs associated with NEC in the United States are estimated at $1 billion annually [
2]. Transfusion-related necrotizing enterocolitis (TR-NEC) refers to an observed phenomenon that specifically describes a premature infant who develops NEC within 48 h after receiving a red blood cell (RBC) transfusion. Several reports have identified RBC transfusions as a significant and independent risk factor for NEC [
6‐
14]; however, others have not found an association [
15‐
21], but rather an association with degree of anemia prior to NEC development, which has led to considerable controversy [
18].
No current biomarkers reliably predict NEC, limiting efforts to prevent this disease. The range of symptoms are highly variable, from subtle signs such as feeding intolerance and abdominal distention, to complete cardiovascular collapse and shock. Because NEC can progress to extensive bowel necrosis within hours, therapies are often ineffective [
22]. Multiple factors are related to NEC etiology including prematurity, enteral feeding, pro-inflammatory propensity of the immature intestine, and impaired mesenteric blood flow [
23]. The majority of premature infants receive transfusions for anemia of prematurity, and RBC transfusions precede approximately 25–38% of NEC cases [
7,
14,
24]. Transfusion of different storage aged RBCs to premature infants has not been shown to contribute to the risk of NEC [
7,
24]. However, the chronological storage age of RBCs may not be an accurate gauge of donor RBC function and the storage lesion may be exacerbated by gamma irradiation [
25], which is performed to prevent transfusion-associated graft-vs-host disease. Although the Age of Red Blood Cells in Premature Infants (ARIPI) trial investigated the effects of total storage duration of RBCs in preterm infants [
26], the study did not investigate the effects of irradiation [
27]. Currently, the “safe” duration of RBC storage following irradiation (post-irradiation storage time, pIST) is unclear. Given the multifactorial etiology of NEC, preventative efforts will be more successful if clinicians understand the underlying pathophysiologic mechanisms and modifiable risk factors influencing the disease.
Although premature infants weighting ≤1250 g at birth are frequently transfused for anemia of prematurity, optimal transfusion guidelines are ill-defined [
28]. The Premature Infants in Need of Transfusion (PINT) trial [
29] and the smaller Iowa trial [
30] investigated the effects of transfusion practices on morbidity, although neither trial included NEC as the primary outcome. Because the PINT trial suggested lower hemoglobin thresholds decreased the number of RBC transfusions with no adverse effect on mortality, retinopathy of prematurity, or neurologic injury [
29], many centers shifted to conservative transfusion practices. Concurrently, multiple published reports have described an association between RBC transfusion and NEC [
6,
7,
10,
24,
31‐
33], although meta-analyses have shown conflciting findings regarding any association [
11,
34]. The lack of adequately-powered randomized trials evaluating the effect of transfusion thresholds on NEC limit determination of whether increased tolerance of neonatal anemia by use of conservative transfusion thresholds may actually increase the risk of NEC.
A causal link between RBC transfusion and NEC has been proposed, but not proven. Our previous research described a matched case-control study of 184 very low birth weight (VLBW) infants weighing ≤1500 g with NEC, and found a higher risk of late-onset NEC (after 4 weeks of age) in transfused infants (OR 6.7; 95% CI: 1.5–31.2) [
8]. An initial meta-analysis of observational studies also showed increased risk of NEC in VLBW transfused infants [
11], although a more-recent meta-analysis found no association [
34] with findings consistent from our recent multicenter, prospective cohort study [
18]. Many of the studies included in the meta-analyses [
34] were observational and limited in causal inference; no studies have provided data regarding the potential underlying pathophysiologic mechanisms. While these studies identified risk factors for NEC, including severity of anemia and a developmental window at which NEC occurs, few studies have focused on characteristics of the donor RBC transfusion, such as pIST and metabolic/functional abnormalities. Therefore, the critical scientific gap that remains to be addressed is whether transfused RBC characteristics, such as irradiation and metabolism, impair intestinal function and/or microvascular circulation. Our current investigation aims to prospectively evaluate the relationship between pIST, RBC metabolomic profiles, and anemia on mesenteric oxygenation, as measured by near-infrared spectroscopy (NIRS), and NEC.
Candidate biological mechanisms of NEC
A number of potential mechanisms and clinical factors with biologic plausibility support a potential causal connection between RBC transfusion in response to anemia and NEC, despite the limitations described previously. Underlying this association is a common central component of insufficient oxygen delivery to intestinal tissue from a combination of decreased oxygen carrying capacity (anemia) and/or decreased blood flow (cardiac output, vascular tone). Oxygen consumption and extraction in intestinal tissue beds can be continuously and non-invasively monitored by near-infrared spectroscopy (NIRS) through measurement of oxygenated versus deoxygenated hemoglobin in venous (75%) and capillary (25%) blood [
35].
Severity of Anemia and oxygen delivery to intestines
Severe anemia, leading to decreased oxygen delivery, may cause intestinal injury that predisposes an infant to NEC. Alkalay and colleagues [
36] demonstrated that infants who appeared clinically “stable” with either significant anemia (hematocrit < 21%) or milder anemia (hematocrit 22–26%) had high cardiac output and restricted intestinal blood flow. Singh calculated that each percent decrease in nadir hematocrit led to a 10% increase in odds for NEC (OR 1.10; 95% CI: 1.02–1.18;
P = 0.01) [
10]. In our retrospective study, infants who developed NEC after transfusion had lower hematocrits 1 week prior than those without NEC [
8]. We also found lower mesenteric oxygen saturation (MES-rSO
2) measured by NIRS during and after transfusions in infants who developed NEC, and enteral feedings given during RBC transfusion worsened this effect [
9]. Furthermore, a recent prospective study from our group found the rate of NEC was significantly increased among VLBW infants with severe anemia (≤ 8 g/dL) in a given week compared with those who did not have severe anemia (adjusted cause-specific hazard ratio, 5.99 [95% CI, 2.00–18.0];
p = .001) [
18]. However, no study to date has prospectively compared longitudinal hemoglobin/hematocrit measures, MES-rSO
2, and development of NEC in this population.
The RBC storage lesion, irradiation, nitric oxide (NO), and NEC
RBCs mediate local blood flow to preferentially perfuse the most hypoxic tissues, a process termed hypoxic vasodilation [
36]. Nitric oxide (NO) released by RBCs is a potential mediator [
37]. However, RBCs can also scavenge NO, a vasoconstrictive activity that may be enhanced in transfused RBCs with longer storage [
38,
39]. Therefore, transfusion of stored RBCs (“storage-aged RBCs”, saRBCs) or pre-storage irradiated saRBCs (which worsens storage lesion) [
40,
41] may disrupt vascular tone and blood flow. In animal studies, blood vessels of the immature intestine vasoconstrict when NO is depleted [
42‐
45]. Thus, in a preterm infant with anemia, NEC could result from two mechanisms: transfusion of saRBCs (including those with extended pIST) interacting with immature intestinal endothelium, which together synergistically reduce blood flow, causing tissue hypoxia and, in some cases, NEC.
Methods
All RBC transfusions given to infants during hospitalization will be studied. All RBC transfused units are stored in citrate-phosphate-dextrose-adenine (CPDA-1) preservative solution. pIST and storage days will be recorded for each RBC transfusion. All infants will be monitored with NIRS prior to, during and up to 48 h following each transfusion. Consistent with epidemiologic reports of transfusion-related NEC and prior studies at the 3 centers, we anticipate approximately 20 (10%) infants will develop NEC while on study. For our 2:1 case-controlled metabolomic analysis, we will prospectively analyze 40 infants who do not develop NEC and compare to 20 infants with NEC. Within this sub-cohort, we will compare alterations in metabolic pathways from saRBC unit (in vitro) and infant blood sample (in vivo). We will then examine a third sub-cohort of 120 infants without NEC within the NEC “window” (29–34 postmenstrual weeks’). These infants will also be monitored weekly for 24–48 h with mesenteric NIRS to evaluate the relationship between anemic (hemoglobin < 8 g/dL) and non-anemic infants. Reports suggest that this specific population of infants are more likely to experience paradoxical reductions in MES-rSO
2 substantially increasing the risk for NEC when transfusions are given [
10,
36]. Analysis will also include assessment of hemoglobin as a continuous variable.
Data management and quality control
To ensure data quality and procedural adherence of our statistical analysis approach, we will implement a detailed data management plan. Quality control will be applied to each phase of data handling to safeguard data collection and process reliability.
Birth cohort data
Case report form data, as defined and dictated by our study protocol, will be collected and managed using iDataFax, an electronic data capture application with extensive management features including a data query system to help ensure study credibility. The iDataFax system will generate regular reports that summarize and track routine data collection. These reports will help the investigative team monitor and maintain data completeness during follow-up and achieve high data capture performance by minimizing missed scheduled clinical assessments, preventing or reducing missing data, and maintaining high cohort retention rates over the three months of regular infant assessment at our three participating centers.
NIRS data
NIRS data will be downloaded daily and uploaded to a secure server within 24 h of monitoring completion. The data coordinating center will download and process all NIRS files on a weekly basis. During data processing, quality control reports will be generated to summarize the expected and actual duration of NIRS monitoring, percent of missing data, and identify when 30 min or more of consecutive data are missing. This approach will ensure proper data collection for future analysis for the entire duration of the monitoring period and confirm that NIRS machines are working properly. If one of our checks fails, we will notify the study nurses who will flag the machine, assess and correct the issue. If issues continue, we will notify the NIRS machine manufacturer, Medtronic, Inc. (Boulder, CO) for technical support. These steps will ensure consistency of data collection across our three study sites. In addition to individual NIRS monitoring checks, quarterly reports summarizing the total number of patients with NIRS monitoring, patient characteristics, and summary statistics for measurements collected during NIRS monitoring will be generated and reviewed by study investigators and biostatisticians.
Primary outcome
All infants enrolled who receive RBC transfusion will have MES-rSO
2 measured by NIRS as the primary study end-point [INVOS 5100C Cerebral/Somatic Oximeter (Covidien, Boulder, CO)], a Food and Drug Administration approved device for use on premature infants. NIRS noninvasively measures regional tissue saturation (rSO
2) in real time because it calculates the difference between oxyhemoglobin (HbO
2) and deoxyhemoglobin (HHb) expressed as: rSO
2 = HbO
2/HbO
2 + HHb [
47]. WE will obtain a baseline measurement by placing the NIRS probes on the infant at least 30 min prior to transfusion (triggered by the decision to transfuse made by the clinical team). Probes will remain in place to collect data for 48 h following transfusion completion. Two-probe site monitoring on mesenteric and renal beds will be used to evaluate differential tissue bed oxygenation. Adhesive sensor probes are vertically applied to left periumbilical area for mesenteric monitoring and horizontally to right flank for renal monitoring.
Secondary outcomes
We will examine the association between metabolic features of transfused RBC units and pre-transfusion pIST, alterations in MES-rSO
2, and the development of NEC. Our analysis will include methods previously used [
25,
48]. Our preliminary data examining distinct effects of gamma irradiation on saRBC identified four metabolite pathways that were significantly altered by storage (> 7 days) and irradiation: arachidonic acid, linoleic acid, steroid biosynthesis, and alpha-linoleic acid. Alterations in these pathways may worsen RBC function, and we propose this may be involved with adverse intestinal oxygenation following RBC transfusion that, in some infants, could lead to NEC. However, we will not pre-select pathways for the current analysis. Therefore, we will pursue analytic approaches previously described [
25] to identify metabolites that discriminate those infants with paradoxical MES-rSO
2 responses with NEC to unaffected infants, and we will also conduct an additional secondary analyses focused on biochemical pathways previously identified in storage saRBCs generated from metabolomics analyses.
Statistical methods
We will use a novel statistical approach [
51] implemented by the
NIRStat R package for analyzing the NIRS data. Specifically, the
NIRStat method models the observed MES-rSO
2 time series with a nonparametric smooth function via penalized regression splines [
49,
50]. It then provides accurate and robust statistical measures for characterizing the important features in rSO
2 series. We will use the mean area under the fitted spline curves (MAUC) measure generated from the
NIRStat package to measure the MES-rSO
2 levels at baseline and then at post-transfusion. The MAUC changes from baseline to post-transfusion will be used to quantify the changes in MES-rSO
2 due to transfusion. Two sample t-tests will be used to compare the changes in MAUC between saRBCs +/− pIST. Multivariate linear regression models will also be applied to model the changes in AUC in terms of pIST status (with pIST and without pIST) and other potential confounding factors. The multivariate analysis will allow us to assess whether the changes in MES-rSO
2 differs significantly between saRBCs with pIST and those without pIST, controlling for other confounding factors.
The incidence of NEC and death will be estimated by the cumulative incidence function appropriate for competing risks. Gray’s method (modified log-rank test) will be used to compare NEC cumulative incidence according to baseline clinical characteristics. Cause-specific hazard ratios will be calculated to measure the degree of association between baseline characteristics and NEC, and between baseline characteristics and death by fitting a stratified Cox proportional-hazards regression model for competing risks. The competing risks model will be implemented using SAS PHREG using robust sandwich covariance matrix estimates to account for within-mother correlation that may occur in outcomes of multiple-birth infants.
To guard against model overfitting, we will employ both clinical and statistical criteria in making decisions about which independent variables to include; and we will limit the number of candidate variables. In general, the results of models having fewer than 10 outcome events per independent variable are thought to have questionable accuracy and the tests of statistical significance may be invalid. The use of “machine-learning” covariate selection methods, such as bootstrap bagging, will be utilized to improve the reliability of identifying risk factors for NEC and death. The hazard ratio and its 95% confidence interval (CI) will be calculated for each factor in the presence of others in the final model for NEC and mortality.
For metabolomics analysis, we will examine associations between metabolic features of transfused RBC units and pre-transfusion pIST, alterations in MES-rSO
2, and the development of NEC using an approach previously described [
25]. Correlative analyses without pre-selecting specific metabolic pathways will be performed. Methods of analysis will include a number of gene set analysis, such as MSEA and MetaboAnalyst. We can also borrow from the gene expression packages to conduct more complex analysis of metabolite-set differential expression analysis as well as metabolite-set differential coordination analyses. The pathway level analysis will be followed by the detection of metabolites that contribute the most to the changes of metabolic pattern using the built-in scoring system of the packages.
Discussion
There is an urgent need for a large, hypothesis-driven, prospective study to examine the effect of both RBC unit and recipient factors on the physiologic perturbations that cause NEC [
11]. Given that 75–90% of low birth weight infants receive one or more RBC transfusions [
52,
53], it is reasonable to predict that NEC may result from a combination of pre-existing anemia and reduced intestinal oxygenation exacerbated by metabolic/functional changes in transfused RBCs, due to irradiation and pIST. This investigation may allow us to identify new modifiable factors that can be used to test targeted prevention strategies and mitigate this devastating disease.
We propose to investigate intestinal oxygenation changes that precede the development of NEC. Our overarching hypothesis is that irradiation of RBC units followed by longer storage times perturbs donor RBC metabolism and function, and these derangements are associated with paradoxical microvascular vasoconstriction, intestinal tissue hypoxia and injury and/or NEC in transfused premature infants with already impaired intestinal oxygenation due to significant anemia. Specifically, our primary goal is to characterize the association between metabolic changes in transfused RBCs relative to pIST, and adverse effects in the recipient by in vivo NIRS trend monitoring of intestinal oxygenation, and in vitro RBC functional studies. Our primary endpoint is changes in MES-rSO2 trends in response to transfusion of saRBC+/-pIST as measured by NIRS. The secondary endpoints are to determine the hazard ratio of NEC for low birth weight infants transfused with saRBCs +/-pIST, examine metabolomic fingerprints of transfused RBC in vitro, and examine the impact of anemia severity on MES-rSO2 trends.