Sepsis and necrotizing enterocolitis (NEC) lead to significant morbidity and mortality in premature infants (1). Inflammation due to sepsis and NEC can cause end-organ damage (2, 3), including brain injury, which leads to long-term impairment (4–6). Early recognition and treatment may improve outcomes, but signs and symptoms often overlap with non-infectious conditions (7). Therefore, a diagnosis is often made once the infection is advanced, and on the other hand, empiric antibiotics are given when there is no infection (8). Clinicians have limited tests and data to guide decisions on the likelihood of infection.

Inflammatory molecules, upregulated as part of the immune response to infection, hold promise as biomarkers of neonatal sepsis. A biomarker with sufficient diagnostic accuracy could be a useful screening tool to aid in deciding whether to start antibiotics or continue monitoring for further signs of sepsis. Many studies have evaluated biomarkers for neonatal sepsis in premature infants (9–12). But, few have been translated into clinical care (13). Decisions to start or stop antibiotics in the NICU largely rely on the clinician's assessment of clinical signs and non-specific laboratory tests, such as C-reactive protein (CRP) (14) and complete blood count (CBC) components (15).

Despite extensive research, no single blood, stool, or urine biomarker has been adopted in clinical practice for its predictive accuracy to diagnose neonatal sepsis and start antibiotic therapy (16). Physiologic sepsis biomarkers, such as abnormal vital sign patterns, can be detected using analysis of continuous vital sign data from standard bedside monitors. We previously used predictive analytics and a multicenter cohort of very low birth weight (VLBW, <1,500 g) preterm infants to develop and externally validate a cardiorespiratory sepsis risk model (17). The model, called Pulse Oximetry Warning Score (POWS), detects physiologic biomarkers of abnormal heart rate and SpO2 patterns that occur near the clinical diagnosis of late-onset sepsis and calculates the relative risk of blood culture-positive sepsis in the subsequent 24 h (17). POWS has yet to be implemented in clinical practice, so we sought to understand how inflammatory biomarkers will correlate and add to cardiorespiratory markers of sepsis. Combining information from hematologic and physiologic inflammatory biomarkers could improve the accuracy of sepsis diagnosis over using either alone.

While many prior studies have analyzed inflammatory biomarkers in premature infants at the time of suspected sepsis (10, 18–20), few have analyzed baseline levels or changes over time in patients who develop sepsis (21, 22). Inflammatory biomarkers may be less useful in premature infants with non-infectious conditions that lead to inflammation, including respiratory distress syndrome and chronic lung disease. To understand the potential utility of biomarkers in VLBW infants, more data are needed. In this study, we aim to identify inflammatory biomarkers that discriminate sepsis from baseline levels and non-sepsis conditions in VLBW infants. To do this, we measured multiple biomarkers in samples collected from remnant plasma, weekly when available and at the time of evaluation for late-onset sepsis or NEC.

We assayed inflammatory biomarkers at the time of blood culture for suspected sepsis and routine laboratory testing in VLBW infants and found differences that distinguish late-onset Gram-negative sepsis and NEC from other diagnoses, but similar biomarker distributions obtained from samples collected at baseline and at the time of suspected sepsis or Gram-positive sepsis. We also found correlations between inflammatory biomarkers and POWS, an algorithm that detects abnormal patterns of heart rate and oxygen saturation that we previously developed as a physiomarker of sepsis (17).

Our results confirm the findings in a previous study of VLBW infants that multiple biomarkers can discriminate between GN sepsis and GP sepsis at the time of positive blood culture (12). Since Gram-negative infections and NEC are known to carry higher morbidity and mortality, this finding might be a useful adjunct clinical decision support tool in helping pick the appropriate empiric antibiotic therapy and institute treatment early (30–32). The current study adds to our prior work (12) because we analyzed samples remote from suspected sepsis. We found no significant difference in biomarker levels in these samples compared to those obtained at the time of clinical sepsis or Gram-positive sepsis. Some VLBW preterm infants may have a chronic or subacute systemic inflammatory response associated with lung disease or intracranial hemorrhage, which could confound sepsis prediction using hematologic biomarkers or cardiorespiratory physiomarkers (33–35).

Plasma cytokines and chemokines are early markers of immune activation in response to infection, and have been shown to rise in septic premature infants (22, 36, 37). Several studies have evaluated the utility of biomarkers for diagnosing sepsis by comparing biomarker levels in cases vs. controls (11). Others have compared cytokine profiles obtained at the time of suspected infection in VLBW infants when the blood culture returns positive vs. negative (38, 39). Instead, we assessed differences in inflammatory biomarkers measured at times when there was clinical suspicion for infection or sepsis and at times remote from infection. Kuster and colleagues also took the approach of measuring cytokines at baseline and at the time of suspected sepsis in VLBW infants and found IL-1ra and IL-6 to have high sensitivity and specificity in 21 cases of late-onset sepsis, even measured from samples taken the day prior to blood culture (37). A longitudinal study of extremely low birth weight (<1,000 g) measured cytokines at days 1, 3, 14, and 21 after birth and found that IFN-γ, IL-10, IL-18, TGF-β, and TNF-α levels differed among infants who developed fungal or bacterial LOS compared with those who never developed sepsis (22). This study enrolled a large, multicenter cohort but did not evaluate differences relative to the timing of blood cultures. They found that infants with higher levels of immune regulatory cytokines relative to pro-inflammatory cytokines were associated with an increased risk of LOS at any time during the NICU course (21).

Of the biomarkers assayed in our study, IL-6 has been most consistently identified as a promising sepsis biomarker in previous studies, though with variable sensitivity and specificity (39). We found that IL-6 had a high specificity at a threshold set to detect 100% of Gram-negative sepsis and NEC cases but low predictive accuracy for Gram-positive cases, which were mostly due to CONS bacteremia. IL-6 has been shown to predict fatal or severe sepsis in premature infants (40), which occurs more often in Gram-negative than Gram-positive infections due to pathogen virulence factors (41). However, we also found elevated levels of IL-6 measured from samples remote from sepsis with blood draws for routine laboratory tests.

The inclusion of sST-2 as a sepsis biomarker was novel for this population as its utility has mainly been studied in adult populations. It is an IL-1 family receptor that binds IL-33 and has been implicated in diseases involving intestinal inflammation (42). Recent studies in adult patients demonstrate an association between elevated sST-2 and the severity of illness in Clostridium difficile colitis (43, 44). Our results indicate that this may also be a useful biomarker for diagnosing NEC and Gram-negative sepsis, where gastrointestinal dysfunction and bacterial translocation cause systemic inflammation.

Cytokines and chemokines such as IL-6, IL-1ra, and IL-8 have been demonstrated to have diagnostic utility as early sepsis markers (37, 45), while acute phase reactants such as CRP and PCT rise during the later phases of systemic inflammation (46). CRP and PCT are also examples of the few inflammatory biomarkers available in U.S. clinical laboratories, which may drive their clinical use despite evidence of low clinical utility (14). A recent meta-analysis showed overall low sensitivity and specificity of CRP for LOS diagnosis (47).

We note several limitations of the study. First, the analysis was limited by the small sample size, where most late-onset sepsis events were due to CONS bacteremia, some of which may have represented contamination and not a true infection. Second, events diagnosed as clinical sepsis are heterogeneous in severity with subjective diagnostic criteria. We collected data on clinical characteristics, but not to the level of detail to account for concurrent inflammatory processes, such as lung disease, invasive mechanical ventilation, and minor procedures. Finally, the use of remnant plasma allowed us to enroll patients and collect samples without additional blood draws, but the volume of plasma available was small and therefore did not allow assays to be run in duplicate.

Analysis of plasma biomarkers and physiomarkers of LOS using POWS, a cardiorespiratory sepsis risk score, resulted in promising correlations that warrant further confirmation in larger studies. Continuous cardiorespiratory predictive monitoring used in conjunction with biomarker testing could prove useful, both for early initiation of antibiotics when the likelihood of sepsis is high and for sparing antibiotics when biomarkers and physiomarkers indicate low sepsis risk.

In conclusion, in a prospective cohort of VLBW infants, inflammatory biomarkers discriminated between late-onset sepsis due to gram-negative bacteremia or NEC and all other samples, including those collected at times with or without suspected sepsis. Biomarkers measured at the time of sepsis due to gram-positive bacteremia did not differ from those measured at times without sepsis. Several inflammatory biomarkers, measured at baseline and at the time of suspected or confirmed sepsis, correlated with cardiorespiratory physiomarkers of sepsis.