Serum biomarkers in the early detection of necrotizing enterocolitis: a systematic review

Search

A systematic review was conducted according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [19].

MEDLINE (PubMed), SCOPUS, and Web of Science databases were systematically searched on November 22nd of 2024 using the following terms: marker, biomarker, necrotizing enterocolitis, necrotising enterocolitis, NEC (Table S1).

Study selection

The inclusion criteria were human studies assessing the diagnostic accuracy of serum biomarkers for detecting NEC in preterm and full-term neonates with a confirmed or suspected diagnosis of NEC (Bell stage I, II, and III), published between January 2014 and November 2024. Temporal restriction was applied to the literature search to ensure the most relevant and up-to-date studies were included.

Articles were excluded if they lacked sufficient data on diagnostic accuracy, did not provide any criteria for diagnosing NEC, or were written in languages other than English. Only studies published in English were included due to resource limitations and the predominance of relevant literature in this language. Duplicate articles, comments, literature reviews, systematic reviews, meeting abstracts, and case-reports were also excluded.

The reference lists of relevant systematic reviews and meta-analyses were reviewed to identify potentially eligible studies. A manual search for additional relevant studies was also conducted using references from the included articles.

After removing duplicates, two independent authors (S.Prediction and J.N.) conducted the eligibility assessment. Discrepancies between the reviewers were addressed through discussion. In case of persistent disagreement, the third reviewer (I.A.) was consulted, and a final decision was reached by consensus. The initial phase of the analysis involved a comprehensive review of all article titles and abstracts to identify relevant studies, using the Rayyan platform for this purpose. The next phase consisted of a thorough evaluation of the full texts of the selected articles.

Data collection process and data items

As in the selection phase, data extraction was performed by two independent reviewers (S.P. and J.N.). The following information was collected from each included study, when available: first author’s surname, publication year, country, study design, study population (i.e. Total number of participants, including number of cases and controls, demographic and/or clinical characteristics), characteristics of controls, target Bell stage(s), timing of blood sample collection, cut-off value(s) and marker performance, including sensitivity, specificity, and area under the receiver operating characteristic curve (AUC-ROC). Studies that did not specify the included Bell stages but reported the use of the Modified Bell’s Staging Criteria for diagnosing NEC were assumed to incorporate Bell’s stages I, II, and III.

Risk of bias assessment

The risk of bias and quality of the studies included in this review was assessed using the Quality Assessment of Studies of Diagnostic Accuracy-Revised (QUADAS-2) [20]. Two authors (S.P. and J.N.) independently evaluated the risk of bias across four domains: patient selection, index test, reference standard, and flow and timing. Concerns regarding applicability were also assessed for the first three domains. Each domain was classified as having a low, high, or unclear risk of bias or applicability concern. Traffic-light and summary plots were further developed with robvis (visualization tool) [21].

Study selection

A total of 3,532 articles were identified through database searches: 1,116 in MEDLINE, 1,382 in Scopus, and 1,034 in Web of Science. After removing 1761 duplicates, 1771 articles remained for initial screening. After title and abstract reviewing of these records, 82 articles were selected for full-text screening. Ultimately, a total of 40 articles that met the inclusion criteria were included in this systematic review (Figure 1). Some of the studies excluded focused only on evaluating the predictive value of serum markers for surgical intervention and/or death instead of the outcome of this study: assessing biomarkers for early diagnosis of NEC [22], [23], [24], [25].

Figure 1:

PRISMA 2020 flow diagram for new systematic reviews which included searches of databases and registers only. Source: page MJ, et al. BMJ 2021;372:n71. Doi: 10.1136/bmj.n7.

Study characteristics

The characteristics of the included studies are presented in Tables 1–5. These studies were conducted across several countries, were published between January 2014 and November 2024, and included 5,610 neonates. All 40 studies were observational, including 15 cohort studies, 22 case-control studies, 2 cross-sectional and 1 case-series. Control groups included asymptomatic neonates, neonates with differential diagnosis (e.g. Sepsis, feeding intolerance) or neonates with a different Bell stage. Respectively, Bell stage I was analyzed in 16, Bell stage II in 36 and Bell stage III in 39 articles. Blood samples were collected at the time of symptom presentation in 23 studies [26], [27], [28], [29], [30], [31], [32], [33], [34], [35], [36], [37], [38], [39], [40], [41], [42], [43], [44], [45], [46], [47], [48], while in two studies samples were performed before symptom onset [49], 50], three after diagnosis [51], [52], [53], two at the time of diagnosis [54], 55], three at hospital admission [56], [57], [58], one at the time of admission for Bell stage I and at the time of surgery for Bell stage III [59], two in the first 24 h of life and at the time of diagnosis [60], 61], and one study either before or after diagnosis [62]. Three studies did not specify the timing of blood sample collection [63], [64], [65].

The studies evaluated a range of serum biomarkers, which were categorized according to their biological functions or proposed roles in the pathogenesis of NEC to provide a clearer understanding of their diagnostic relevance. Specifically, markers were grouped into hematological indices, acute phase reactants, immunological markers, tissue damage and tissue repair markers, and metabolic markers.

Risk of bias assessment

In the patient selection domain, 67.5 % of studies exhibited a high risk of bias. Three studies were classified as having a high risk of bias due to their exclusion of patients with clinical or radiologic features suggestive of NEC at various stages, while the remaining studies faced limitations due to their case-control design. Specifically, one study excluded patients with Bell stage I [45], another excluded those with recent infectious diseases within a week prior to the onset of NEC symptoms [35], and a third excluded patients with Bell stage IIIB [54]. Additionally, the inclusion of infants already diagnosed with NEC in any stage compromised the applicability of this domain in 62.5 % of studies.

The interpretation of index tests without knowledge of the reference standard was rarely reported in the analyzed studies. As a result, 87.5 % of studies had an unclear risk of bias. The same was observed for the independent interpretation of the reference standard, leading to an unclear risk of bias in 85 % of the studies. All articles presented a low risk of applicability concerns regarding the index test and reference standard domains. A notable source of bias identified was the timing of blood sample collection, which occurred either after or at the moment of diagnosis of NEC, which led to a high risk of bias in 15 % of the studies reviewed, compromising the assessment of biomarkers as potential tools for early detection. Notably, three studies had an unclear risk of bias in the flow and timing domain due to the absence of explicit information regarding the timing of blood sample collection [63], [64], [65].

Overall, 29 out of 40 articles presented a high risk of bias in at least one domain. Additionally, 25 out of 40 studies exhibited a high risk of applicability concern related to the patient selection domain (Figure S1).

Hematological indices

Hematological parameters were evaluated in 11 studies (Table 1) [29], 30], 32], [35], [36], [37], [38, 48], 50], 51], 58]. Among these, three studies targeted NEC Bell stages≥I [36], 51], 58], six targeted Bell stages≥II [29], 30], 35], 38], 48], 50] and two targeted Bell stage III [32], 37]. Five studies included asymptomatic neonates as controls, three used symptomatic neonates, one included neonates with sepsis, and another involved neonates with highly suspected early-onset food protein-induced enterocolitis syndrome (HSEO-FPIES) as controls [29], 35], 36], 38], 48], 50], 51], 58]. In addition, two studies used NEC Bell stages≥II as the target stage and NEC Bell stage I as the control group [36], 51]. One study compared NEC stage II with NEC stage I [51], while three studies examined NEC stage III in comparison to NEC stage II [29], 35], 51]. Two studies focused on NEC stage III and NEC stages≤II as the target stage and control group, respectively [32], 37].

For detecting any stage of NEC, white blood cells (WBC) presented high sensitivity (96.5 %), but with a very low specificity of 26.8 % [51]. WBC also exhibited high specificity (100 %) for detecting Bell stage I with a sensitivity of 46.67 % [58].

Regarding the detection of Bell stages≥II, absolute monocyte count (AMC) revealed the highest specificity (93 %) in Desiraju et al.’s study, where a 50 % reduction in AMC at symptom onset differentiated symptomatic neonates from NEC, though sensitivity was only 51 % [29]. However, Moroze et al. found that the same 50 % reduction in AMC had lower sensitivity and specificity for NEC stages≥I [36]. In addition, a neutrophil count of 58.85 % at symptom onset showed the highest sensitivity (96.6 %) in distinguishing NEC from asymptomatic neonates, but with a specificity of 60 % [35].

Concerning the distinction of NEC stages≥II from sepsis, a WBC count of 15.25 × 10 ⁹/L and a platelet count of 123.5 × 10 ⁹/L presented high specificity (93.3 and 90 %, respectively), although with low sensitivity (31 and 48.3 %, respectively) [35]. A similar trend was observed when comparing NEC to controls with HSEO-FPIES, where WBC showed high specificity (98.36 %) but low sensitivity (39.02 %) [30].

Neutrophil/lymphocyte ratio and platelet count exhibited high sensitivity in discriminating between NEC stage I and stage II (97.6 %) and in distinguishing NEC severity (91.7 %), respectively, but with lower specificity [35], 51]. For distinguishing NEC stage III from NEC stages≤II, Cai et al. Reported that the combination of mean platelet volume (MVP) of 10.80 fL with a procalcitonin (PCT) level of 18.57 ng/mL at symptom onset improved sensitivity to 90 % but specificity remained moderate (78 %), compared to MVP alone [32]. In contrast, Meng et al., who evaluated a larger sample and used a lower threshold of 9.395 fL, observed a higher sensitivity (96.2 %), but lower specificity (32.6 %) [37]. Specificity improved to 93 %, with a sensitivity of 84.8 %, when MPV was combined with PCT (AUC=0.935) [37].

Acute phase reactants

Acute phase reactants were assessed in 14 studies (Table 2) [28], 30], 32], 35], 37], [40], [41], [42, 47], 51], 52], 58], 62], 63]. Of these, five targeted NEC stages≥I [51], 52], 58], 62], 63], seven focused on NEC stages ≥II [28], 30], 35], [40], [41], [42, 47], and two analyzed NEC stage III [32], 37]. The control groups varied among the studies: seven included asymptomatic neonates, two used symptomatic neonates, two had sepsis as the control group, one included HSEO-FPIES cases, and one combined asymptomatic and sepsis cases as controls [28], 30], 35], [40], [41], [42, 51], 52], 58], 62], 63]. One study used NEC stages ≥II as the target group and Bell stage I as the control [58], while another considered Bell stage I or cases of spontaneous intestinal perforation (SIP) as controls [47]. Additionally, one study compared NEC stage II to NEC stage I [51], whereas two studies examined NEC stage III vs. NEC stage II [35], 51]. Finally, two studies focused on NEC stage III as the target group and NEC stages ≤II as the control [32], 37].

In El-farargy et al.’s study, both C-Reactive Protein (CRP) levels of 2.89 mg/L and PCT of 9.35 ng/mL showed high sensitivity (90 %) and specificity (90 and 95 %, respectively) for detecting NEC stages ≥I, achieving excellent AUC values of 0.954 and 0.976, respectively [63]. However, Zhang et al. [52] and Yang et al. [51] evaluated larger sample sizes with blood samples obtained after diagnosis. Although both studies exhibited a comparable ability of CRP to differentiate NEC from asymptomatic infants, with AUC values of 0.655 and 0.650, respectively, they reported discrepant sensitivity and specificity results. While Zhang et al. [52] reported a low sensitivity of 39.1 % but a high specificity of 92.4 % for CRP levels ≥7.38 mg/L, Yang et al. [51] observed a high sensitivity of 93.0 % but a very low specificity of 31.7 %, without specifying a cut-off value. In addition, a PCT threshold>0.831 ng/mL had a specificity of 100 %, but a low sensitivity (55.56 %) for detecting Bell stage I [58].

Regarding the diagnosis of NEC stages ≥II, Dong et al. Reported an AUC of 0.917 for CRP in distinguishing NEC from asymptomatic neonates [35]. However, a notable discrepancy was observed in the study by Shen et al., which analyzed a larger population sample and reported more limited discriminative capability, with an AUC of 0.690 [41]. Both osteoprotegerin (OPG) and tumor necrosis factor-related apoptosis-inducing ligand (TRAIL) demonstrated high sensitivity (100 and 96.6 %, respectively), although with low specificity (66.7 %) [35]. In contrast, fibrinogen gamma-dimer revealed high specificity (91.67 %), but low sensitivity [42].

Although Dong et al. [35] found that CRP had a weak diagnostic performance in distinguishing between NEC and sepsis (AUC=0.564), Ng et al. [28] reported that a CRP level of 10 mg/L exhibited high sensitivity (92 %), but low specificity (55 %), when compared to asymptomatic or sepsis cases. In contrast, FGG-dimer showed high specificity (100 %), but low sensitivity when compared to sepsis [42]. CRP also showed high specificity (94.64 %) but a lower sensitivity of 64.10 % in discriminating NEC from HSEO-FPIES neonates [30]. When compared with NEC stage I or SIP, a 100 % sensitivity was observed for CRP levels<4 mg/L and inter-alpha inhibitor protein levels<207 mg/L [47]. However, CRP had a relatively low specificity of 64.7 %, while inter-alpha inhibitor protein showed a specificity of 88.2 % and achieved an AUC of 0.98 [47].

Regarding the discrimination between moderate to severe NEC from mild NEC, CRP levels>27.56 mg/L and PCT values>1.42 ng/mL were associated with high specificity (100 and 95.56 %, respectively) and moderate sensitivity (81.48 %) [58]. PCT demonstrated excellent diagnostic performance at symptom onset with an AUC of 0.919 in differentiating severe NEC from stages I and II [37]. Furthermore, OPG showed high specificity (94.1 %) in assessing NEC severity, but with a relatively low sensitivity of 61.5 % [35].

Immunological markers

Immunological markers were analyzed in 10 articles (Table 3) [26], 30], [33], [34], [35, 46], 49], 52], 55], 63]. Six of these studies assessed Bell stages ≥I [26], 33], 34], 52], 55], 63], while four targeted Bell stages ≥II [30], 35], 46], 49]. As controls, eight studies included asymptomatic infants, one had sepsis, one used HSEO-FPIES neonates, and another combined asymptomatic or feeding intolerance [26], 30], [33], [34], [35, 46], 49], 52], 55], 63]. One study compared NEC stage III with NEC stage II [35].

For the detection of NEC stages ≥I, epithelial neutrophil-activating peptide-78 (ENA-78) demonstrated high sensitivity (90 %) and specificity (95 %) at a concentration of 159.67 pg/mL (AUC=0.978) [63]. Interleukin (IL)-8 and IL-18 also presented excellent diagnostic performance, with an AUC value of 0.99 and 0.935, respectively [26], 63].

Regarding the diagnosis of Bell stages ≥II, IL-8 demonstrated high sensitivity (96.6 %) and a specificity of 86.7 % (AUC=0.907) [35]. The combination of IL-8 with OPG, TRAIL, C–X–C motif chemokine 1 (CXCL1), thymic stromal lymphopoietin (TSLP), monocyte chemotactic protein-4 (MCP-4), tumor necrosis factor ligand superfamily member 14 (TNFSF14), IL-24, matrix metalloproteinase-10 (MMP-10), leukemia inhibitory factor (LIF), and C–C motif chemokine 20 (CCL20) showed strong diagnostic performance, achieving the same sensitivity of 96.6 % as IL-8 alone but with enhanced specificity of 90 % (AUC=0.972) [35]. In addition, CXCL1, TSLP, and CCL20 also exhibited high sensitivity values (93.1 , 96.6, and 96.6 %, respectively), while MCP-4 demonstrated the highest specificity (90 %) [35]. Anti-myosin autoantibodies achieved an excellent AUC value of 0.9457 within 48 h of symptom onset for NEC stage I [34]. Tumor necrosis factor-alpha showed a sensitivity of 100 % for discriminating between preterm neonates with severe NEC and asymptomatic newborns or those with feeding intolerance. However, this was accompanied by a specificity of only 8.7 % [46]. In contrast, IL-8 had high specificity (97 %) but a low sensitivity (60 %) [46].

In Qi et al.’s study, an IL-6 level of 201.70 pg/mL was associated with a high specificity of 98.04 %, but a relatively low sensitivity of 44.64 % for discriminating between NEC stages≥II from HSEO-FPIES [30]. In addition, the combination of IL-27 with CRP demonstrated excellent discriminatory ability with an AUC of 0.903 [30]. For the differentiation between NEC and sepsis, TNFSF14 showed high sensitivity (93.1 %), but low specificity [35].

Concerning the distinction of NEC severity, the combination of IL-8, OPG, MCP-4, IL-24, LIF, and CCL20 reported higher accuracy than IL-8 alone, with a sensitivity of 92.3 % and a specificity of 100 % [35]. Additionally, MCP-4 showed a high sensitivity of 92.3 %, and CCL20 exhibited a specificity of 100 %, although with low complementary specificity or sensitivity [35].

Tissue damage and tissue repair markers

Tissue damage and tissue repair markers were assessed in 21 studies (Table 4) [25], [26], [27, 32], [34], [35], [36, 40], 43], 47], [52], [53], [54, 56], [58], [59], [60, 63], 64]. Of these studies, seven included Bell stages ≥I [26], 27], 31], 55], 58], 60], 61], 11 Bell stages ≥II [28], 35], 40], 41], [43], [44], [45, 48], 57], 59], 65], one Bell stage IIIA [54], one Bell stage III [53], and one Bell stages ≤II as the target stage [64]. 15 studies included asymptomatic neonates as controls, two had symptomatic infants, two used sepsis cases, one combined asymptomatic and Bell stage I, and two combined asymptomatic and sepsis cases [25], [26], [27, 31], 32], [34], [35], [36, 40], 43], 47], 52], 54], [56], [57], [58], [59], [60, 63], 64]. Two studies focused on patients with NEC Bell stages≥II as the target group, using Bell stage I as the control group [58], 60]. Another study compared patients with NEC Bell stage III to those with Bell stage II [35]. Additionally, a separate study examined NEC stage IIIA as the target stage, using NEC stage II as the control group [54].

Sensitivity and specificity of 100 % for intestinal fatty acid-binding protein (I-FABP) were achieved in the study conducted by Abdle-Haie et al. with a threshold ≥37.95 ng/mL at the time of diagnosis [60]. A lower threshold ≥9.2 ng/mL had a specificity of 100 % and a sensitivity of 95.8 %, as reported by Al-banna et al. [61]. In the first 24 h of life, I-FABP also showed high sensitivity and specificity for predicting NEC at a threshold ≥3.2 ng/mL (both of 95.8 %) and ≥7.75 ng/mL (94.4 and 100 %, respectively) [60], 61]. Regarding the distinction between NEC stages ≥II and NEC stage I, Abdel-Haie et al. also reported that a threshold ≥131.8 ng/mL yielded a sensitivity of 100 % and a specificity of 90 % [60].

Ischemia-modified albumin (IMA) was assessed in a case-control study by El-Meneza et al. and excelled at detecting NEC stages I and II in full-term infants, demonstrating a sensitivity of 90 % and specificity of 95 % at a threshold>12.66 U/mL (AUC=0.975) [64].

Regarding the detection of NEC stages ≥I, liver fatty acid-binding protein and lysosomal α-glucosidase exhibited an excellent discriminatory ability at symptom onset, with an AUC value of 0.95 and 0.91, respectively [26], 27].

Regarding the diagnosis of NEC stages ≥II, the combination of receptor-interacting protein kinase 3, CRP, and lactic acid improved the diagnostic performance compared to receptor-interacting protein kinase 3 alone, achieving an AUC value of 0.925 [41]. Smooth muscle actin, MMP-10, and resistin-like molecule β (RELM-β) showed high specificity (100 , 90, and 91.7 %, respectively), but lower sensitivity (33 , 62.1, and 71.4 %, respectively) [35], 48], 59]. A higher sensitivity and specificity were observed for RELMβ when combined with a platelet count inferior to 157 × 10⁹/L (82.89 and 93.21 %, respectively) [48].

An I-FABP level of 4.1 ng/mL and MMP-10 at symptom onset demonstrated high specificity (92 and 90 %, respectively) and low sensitivity (38 and 55.2 %, respectively) in distinguishing NEC from sepsis [35], 44]. For the distinction between NEC and asymptomatic neonates or sepsis, in Ng et al.’s study [28], micro RNA-1290 (miR-1290) demonstrated the best diagnostic performance at symptom onset, with an AUC value of 0.917. Notably, miR-1290 levels between 220 and 650 copies/µL, in combination with a CRP>15.8 mg/L one day after symptom onset, showed a sensitivity of 83 % and improved specificity to 96 % [28]. In Ahmed et al.’s study, epidermal growth factor also exhibited excellent diagnostic performance, achieving an AUC value of 0.92 [65].

In the study by Huo et al., which analyzed the largest population sample, an I-FABP level of 12.10 pg/mL and a high mobility group box 1 level of 50.65 pg/mL were associated with high sensitivity (92.5 and 95.3 %, respectively) and lower specificity (72.6 and 71.2 %, respectively) in discriminating NEC from asymptomatic or Bell stage I infants [57]. In Schurink et al.’s study, I-FABP presented high specificity (90 %), but low sensitivity (53 %) at symptom onset when compared to symptomatic controls [43].

For distinguishing NEC stages ≥II from Bell stage I, cytosolic β-glucosidase levels>167.70 nmol/L and galectin 3 values>5.15 ng/mL presented high specificity (100 and 93.33 %, respectively) and moderate sensitivity (81.48 and 88.89 %, respectively) [58]. I-FABP showed high sensitivity (90 %) and a specificity of 72 % at a level>3.24 ng/mL for differentiating NEC stage IIIA from NEC stage II [54].

Metabolic markers

Metabolic markers were assessed in five studies (Table 5) [39], 41], 51], 55], 56]. Three of these five studies analyzed Bell stages ≥I [51], 55], 56], while the remaining focused on Bell stages ≥II [39], 41]. All studies included asymptomatic infants as controls. Additionally, one study compared Bell stage II with Bell stage I, as well as Bell stage III with Bell stage II [51].

Regarding the detection of NEC stages ≥I, in Han et al.’s study [52], the best diagnostic performance was observed with the combination of gamma-glutamyl transferase, calcium, and total bilirubin, achieving an AUC of 0.908. Alpha-fetoprotein also achieved an AUC value of 0.926, demonstrating excellent discriminatory ability between NEC stages ≥I and asymptomatic infants [55]. Pre-albumin showed high sensitivity (93 %) but low specificity (51.2 %), although no cut-off value was available. Pre-albumin also demonstrated high sensitivity in distinguishing NEC stage I from NEC stage II, as well as NEC stage II from NEC stage III (97.6 and 94.7 %, respectively) [51]. However, these values were associated with low specificity (47.5 and 42.5 %, respectively) [51].

In relation to the detection of Bell stages ≥II, a primary/secondary bile acid ratio had an excellent diagnostic performance, with an AUC value of 0.90, rising to 0.95 when clinical characteristics were considered [39].