Diagnostic biomarkers for late-onset sepsis in pediatric intensive care: a retrospective cohort study

Abstract

Aim

This study explores the potential of various biomarkers to facilitate the differential diagnosis of late-onset sepsis (LOS) from non-LOS infections in hospitalized pediatric patients.

Methods

We conducted a retrospective cohort study using electronic medical records from our hospital from January 2022 to December 2023, and divided the patients into LOS (n = 178) and non-LOS (n = 159) groups. Data collected included demographic information, levels of inflammatory and metabolic biomarkers. Descriptive statistics were used for demographic data, and multivariable logistic regression followed by ROC curve analysis was used to assess the diagnostic value of these biomarkers.

Results

Significant differences were observed in the levels of PCT, CRP, Lac, HBP, TNF-α, IL-6, IL-1β, IL-10, and IL-12 between the LOS and non-LOS groups (all p < 0.001). Multivariate logistic regression identified PCT, CRP, IL-6, IL-1β, IL-12, and Lac as independent predictors of LOS. ROC curve analysis showed high diagnostic values for PCT, Lac, and IL-1β. A combined diagnostic model of CRP, Lac, and IL-1β achieved the highest performance with an AUC of 0.958, sensitivity of 97.8%, and specificity of 91.8%. Additionally, Gram-negative LOS was associated with higher levels of PCT, CRP, and IL-6 compared to Gram-positive LOS. PCT levels demonstrated moderate diagnostic performance in differentiating LOS caused by Gram-positive vs. Gram-negative bacteria (AUC = 0.626).

Conclusion

The combination of CRP, Lac, and IL-1β serves as a robust set of biomarkers for the differential diagnosis of LOS in pediatric ICU settings. Furthermore, PCT also serves as a critical biomarker for differentiating between Gram-negative and Gram-positive bacterial causes, aiding in more targeted clinical management.

Introduction

Sepsis is a life-threatening organ dysfunction caused by a dysregulated host response to infection and remains a major cause of morbidity and mortality in pediatric intensive care units (PICU) worldwide [1, 2]. According to the 2019 Global Burden of Disease study, there were 6.31 million incident cases of neonatal sepsis and 0.23 million deaths due to neonatal sepsis in 2019 [3]. Sepsis is classified into early-onset sepsis (EOS) and late-onset sepsis (LOS) according to the onset time. Neonatal LOS, typically occurring after the first 72 h of birth, particularly in preterm or low birthweight infants, continues to pose a significant challenge in neonatal intensive care units [4]. The vulnerability of neonates due to their immature immune systems, prolonged hospitalization, and frequent use of invasive devices makes LOS particularly threatening [5]. However, the clinical manifestations of LOS are non-specific, including temperature instability, tachycardia, dyspnoea and feeding difficulties, which makes accurate diagnosis particularly challenging [6]. This can lead to misdiagnoses or treatment delays by clinicians facing similar symptoms, further exacerbating the condition. Thus, the development and validation of reliable diagnostic biomarkers are particularly crucial for enhancing the diagnostic accuracy and therapeutic outcomes in LOS.

Biomarkers have increasingly become important tools in the diagnosis and management of sepsis. In recent years, several biomarkers have been studied for their potential roles in diagnosing sepsis. C-reactive protein (CRP) and procalcitonin (PCT) have emerged as promising candidates [7]. CRP is an acute phase protein synthesized by the liver in response to inflammation, which is significantly elevated in sepsis patients [8]. PCT, a precursor to calcitonin, is produced during bacterial infections and is considered a more specific marker for sepsis compared to CRP [9]. Lactate (Lac), a byproduct of anaerobic metabolism, indicates insufficient tissue perfusion and is associated with the severity of sepsis and the risk of mortality [10]. Inflammatory cytokines, including tumor necrosis factor alpha (TNFα), interleukin-6 (IL-6) and interleukin-1 beta (IL-1β), play a crucial role in the immune response to infection and are linked to the pathogenesis of sepsis [11, 12]. Elevated levels of these cytokines are often observed in septic patients and correlate with disease severity and outcomes [12]. Despite their potential, the clinical application of cytokines as standalone biomarkers for sepsis diagnosis remains limited due to their variability and the influence of multiple factors on their levels.

In addition, the type of pathogens involved in LOS also significantly affects the clinical presentation, management, and prognosis of the disease. Gram-negative bacteria are common pathogens in neonatal sepsis [13]. Due to their production of endotoxins, Gram-negative bacteria are typically associated with more severe disease and higher mortality rates, as endotoxins trigger severe systemic inflammatory responses [14].

Therefore, our retrospective cohort study assessed the diagnostic performance of a set of biomarkers (including CRP, PCT, Lac, and inflammatory cytokines) in distinguishing neonatal LOS from other infectious diseases. Additionally, we categorized the LOS group into Gram-positive and Gram-negative categories to analyze the differences in biomarker levels and their diagnostic utility, aiming to provide compelling evidence on the practicality of specific biomarkers for the accurate diagnosis of LOS.

Methods

Study design and participants

Newborns with infection within 3–28 days of birth from January 2022 to December 2023 in the hospital database were included. Neonates presenting with at least two or more of the following symptoms were considered to have infection: feeding intolerance, lethargy, tachypnea, abnormal body temperature (fever > 38 °C or hypothermia < 36 °C), heart rate > 90 beats per minute, respiratory rate > 20 breaths per minute, arterial partial pressure of carbon dioxide (PaCO₂) < 32 mmHg, peripheral blood leukocyte count > 12 × 10⁹/L or < 4 × 10⁹/L, rash, abdominal distension, or diarrhea. Then, these newborns were divided into the LOS group and the non-LOS group according to the LOS diagnostic criteria [15]. Specifically, newborns were classified into the LOS group if they were older than 3 days and met the following criteria for confirmed diagnosis: (1) presence of clinical signs suggestive of sepsis, including temperature instability, apnea, feeding intolerance, or poor perfusion; (2) positive culture results from blood, cerebrospinal fluid, or other sterile body fluids. The mothers of these newborns had no abnormal blood routine tests before delivery. The exclusion criteria for newborns are as follows: (1) newborns with chromosomal abnormalities, congenital malformations, and genetic metabolic diseases; (2) already suffering from major diseases such as malignant tumors and acute severe hepatitis; (3) incomplete data.

Data collection

Demographic, clinical characteristics, and laboratory results data were extracted from electronic medical records. Demographic data of neonates were collected, including sex, weight, gestational age, date of birth, and use of invasive devices. Biochemical indicators included PCT, CRP, Lac, heparin-binding protein (HBP), TNFα, IL-6, IL-1β, interleukin-10 (IL-10), and interleukin-12 (IL-12). Blood samples for these biochemical indicators were collected at the time of initial clinical suspicion of infection and before antibiotic treatment. LOS patients were further stratified into Gram-negative and Gram-positive groups based on Gram-stain results.

Statistical analysis

Statistical analyses were performed using SPSS 20. Continuous variables were expressed as medians (interquartile ranges) and compared using the Mann-Whitney U test. Categorical variables were expressed as frequencies (percentages) and compared using the Chi-square test. Multivariate logistic regression analysis was conducted to identify independent predictors of sepsis. Odds ratios (ORs) with 95% confidence intervals (CIs) were calculated. The diagnostic performance of each biomarker was evaluated using Receiver Operating Characteristic (ROC) curve analysis, with Area Under the Curve (AUC) values calculated to assess the sensitivity and specificity of each marker. The combined ROC was based on the predicted probability from the multivariable logistic regression model. The sensitivity and specificity values were used to assess the accuracy of the biomarkers in diagnosing LOS and in differentiating the causative bacteria. Results were considered statistically significant at a p-value < 0.05.

Results

General Characteristics

A total of 337 pediatric patients were included in the study, divided into two groups: the non-LOS group (n = 159) and the LOS group (n = 178). The demographic and clinical characteristics of the patients were presented in Table 1. There were no significant differences between the two groups in terms of sex distribution, birth weight categories, gestational age categories, and the use of invasive devices.

Table 1.

Characteristics of the enrolled patients

Index	Non-LOS group (n = 159)	LOS group (n = 178)	P
Males, n (%)	82 (51.57%)	108 (60.67%)	0.093
Weight, g, median (IQR)	1539.3 (1173.9, 1881.4)	1493.2 (510.2, 1768.7)	0.368
Weight, n (%)			0.704
< 1500 g	74 (46.54%)	90 (50.56%)
1500–2500 g	80 (50.32%)	84 (47.19%)
> 2500 g	5 (3.14%)	4 (2.25%)
Gestational age, weeks, median (IQR)	32.1 (29.2, 34.8)	31.69 (28.95, 33.99)	0.392
Gestational age, n (%)			0.134
< 32 weeks	74 (46.54%)	101 (56.74%)
32–37 weeks	66 (41.51%)	56 (31.46%)
> 37 weeks	19 (11.95%)	21 (11.80%)
Invasive devices, n (%)	79 (49.69%)	97 (54.49%)	0.462
Type of infection, n (%)			-
Gram-negative	-	71 (39.89%)
Gram-positive	-	107 (60.11%)
Neonatal Pneumonia	86 (54.09%)	-
Neonatal Urinary Tract Infection	54 (33.96%)	-
Mild Fungal Infections	19 (11.95%)	-

IQR Interquartile Range

-: Indicated that there was no relevant data for this column

Difference of biochemical indicators in two groups

Significant differences were observed in the levels of various biochemical markers between the non-LOS and LOS groups (Table 2). Levels of PCT, CRP, Lac and HBP were significantly higher in the LOS group compared to the non-LOS group (all p < 0.001). Inflammatory cytokines such as TNF-α, IL-6, IL-1β, IL-10 and IL-12 also showed significant increases in the LOS group compared to the non- LOS group (all p < 0.001).

Table 2.

Comparison of biochemical indicators

Index	Non-LOS group (n = 159)	LOS group (n = 178)	P
PCT (ng/mL), median (IQR)	0.29 (0.17, 0.41)	2.07 (1.09, 9.76)	< 0.001
CRP (mg/L), median (IQR)	2.42 (1.28, 3.92)	18.60 (5.51, 44.32)	< 0.001
TNF-α (pg/mL), median (IQR)	5.22 (3.12, 7.48)	6.43 (4.25, 8.75)	< 0.001
IL-6 (pg/mL), median (IQR)	5.85 (3.98, 8.70)	20.46 (5.50, 63.58)	< 0.001
IL-1β (pg/mL), median (IQR)	2.68 (1.58, 3.86)	26.90 (11.94, 44.98)	< 0.001
IL-10 (pg/mL), median (IQR)	6.60 (3.63, 8.88)	103.94 (14.13, 129.28)	< 0.001
IL-12 (pg/mL), median (IQR)	3.36 (2.05, 4.40)	64.48 (44.20, 71.03)	< 0.001
Lac (mmol/L), median (IQR)	1.65 (1.12, 2.21)	3.97 (3.40, 4.46)	< 0.001
HBP (ng/mL), median (IQR)	45.38 (37.47, 53.65)	64.46 (55.05, 76.52)	< 0.001

IQR Interquartile Range, PCT Procalcitonin, CRP C-reactive Protein, HBP Heparin-binding Protein, Lac Lactate

Diagnostic performance of biomarkers in LOS

To determine the effectiveness of the above biomarkers in diagnosing LOS, logistic regression was conducted. As shown in Table 3, PCT, CRP, IL-6, IL-1β, IL-12, and Lac were all significantly associated with a higher risk of LOS occurrence, while TNF-α, IL-10, and HBP were not significant predictors in the multivariate model (Fig. 1). Then, the diagnostic performance of these independent predictors of LOS was assessed using ROC curve analysis. PCT, Lac, and IL-1β all had high diagnostic values, with AUC greater than 0.9 (P < 0.001). A combined diagnostic model using CRP, Lac, and IL-1β also achieved the highest performance, with an AUC of 0.958 (95%CI 0.934–0.983), sensitivity of 97.8%, and specificity of 91.8% (Table 4; Fig. 2).

Table 3.

Multivariate logistic regression analysis

Index	OR (95%CI)	P
PCT (ng/mL)	1.255 (1.022, 1.542)	0.030
CRP (mg/L)	1.055 (1.010, 1.102)	0.017
TNF-α (pg/mL)	/	0.734
IL-6 (pg/mL):	0.890 (0.850, 0.931)	< 0.001
IL-1β (pg/mL)	1.153 (1.085, 1.226)	< 0.001
IL-10 (pg/mL)	/	0.732
IL-12 (pg/mL)	1.029 (1.001, 1.058)	0.043
Lac (mmol/L)	4.108 (2.552, 6.611)	< 0.001
HBP (ng/mL)	/	0.408

PCT Procalcitonin, CRP C-reactive Protein, HBP Heparin-binding Protein, Lac Lactate

Fig. 1.

ROC analysis of the diagnostic efficacy of biomarkers in patients with LOS

Table 4.

Effects of various metabolic indicators in the diagnosis of LOS

Index	Cut-off	AUC (95%CI)	Sensitivity	Specificity	PPV	NPV	Accuracy	P
PCT	0.505	0.918 (0.882, 0.954)	93.8%	91.8%	92.8%	93.0%	0.93	< 0.001
CRP	4.970	0.853 (0.808, 0.898)	78.7%	94.3%	93.9%	79.8%	0.86	< 0.001
Lac	2.495	0.933 (0.901, 0.965)	93.8%	92.5%	93.3%	93.0%	0.92	< 0.001
IL-6	10.030	0.697 (0.639, 0.755)	56.2%	91.8%	88.5%	65.2%	0.73	< 0.001
IL-1β	5.365	0.934 (0.904, 0.963)	91.0%	92.5%	93.1%	90.2%	0.92	< 0.001
IL-12	41.810	0.810 (0.758, 0.862)	79.2%	93.1%	92.8%	79.9%	0.86	< 0.001
CRP + Lac + IL-1β	0.264	0.958 (0.934, 0.983)	97.8%	91.8%	93.0%	97.4%	0.95	< 0.001

Metabolic indicators with AUC > 0.8 were included for screening to obtain indicators for combined diagnosis, and ROC analysis was performed

PCT Procalcitonin, CRP C-reactive Protein, Lac Lactate, PPV Positive Predictive Value, NPV Negative Predictive Value

Fig. 2.

ROC analysis of the diagnostic performance of the combined model of C-reactive protein, lactate, and interleukin-1β

Diagnostic performance of biomarkers in differentiating LOS caused by Gram-negative vs. Gram-positive bacteria

In addition, LOS patients were stratified into Gram-negative and Gram-positive groups based on Gram-stain results. Gram-negative group was associated with higher levels of PCT, CRP and IL-6 compared to Gram-positive group, while Lac levels were lower in the Gram-negative group (all P < 0.05). According to the multivariate logistic regression, low PCT levels were predictive factors for LOS caused by Gram-positive bacteria (Table 5). The ROC curve analysis showed moderate diagnostic performance of PCT levels in differentiating LOS caused by Gram–positive vs. Gram–negative bacteria, with an AUC of 0.626 (95%CI 0.541–0.711; P = 0.004) (Fig. 3). The mortality rate was slightly higher in the Gram-negative group (14.08% vs. 5.61%, p = 0.054).

Table 5.

Effects of various metabolic indicators in differentiating LOS caused by Gram-negative vs. Gram-positive bacteria

Index	Gram-negative (n = 71)	Gram-positive (n = 107)	Univariate logistic	Multivariate logistic
			P	OR (95%CI)	P
PCT (ng/mL)	6.67 (1.38, 9.91)	4.29 (1.03, 9.64)	< 0.001	0.878 (0.817, 0.944)	< 0.001
CRP (mg/L)	29.68 (8.13, 45.97)	22.85 (4.46, 42.59)	0.024		0.257
IL-6 (pg/mL)	42.68 (6.17, 64.78)	28.35 (5.47, 61.47)	0.001		0.415
IL-1β (pg/mL)	33.77 (14.51, 48.70)	27.71 (11.24, 42.86)	0.053
IL-12 (pg/mL)	52.34 (43.00, 69.65)	55.25 (52.69, 73.90)	0.500
Lac (mmol/L)	3.72 ± 1.18	4.03 ± 0.84	0.045		0.113
Mortality rate	10 (14.08%)	6 (5.61%)	0.054

PCT Procalcitonin, CRP C-reactive Protein, Lac Lactate

Fig. 3.

ROC analysis of the efficacy of procalcitonin levels in distinguishing between Gram-negative and Gram-positive infections in patients with LOS

Discussion

Neonatal LOS is a common and serious complication within neonatal intensive care units, with diagnostic and therapeutic challenges stemming from the nonspecific clinical presentations and diversity of pathogens involved. Although traditional bacterial culture methods remain the gold standard, they are time-consuming and do not always provide timely and accurate diagnostic information [16]. Hence, identifying rapid, sensitive, and specific biomarkers for diagnosis and pathogen type determination is crucial for improving treatment efficacy and prognosis of LOS. This study evaluated the utility of various biomarkers in diagnosing LOS, aiming to provide more effective diagnostic strategies for clinical practice.

In this study, multiple biomarkers were significantly elevated in infants with LOS. Both PCT and CRP are robust biomarkers for sepsis diagnosis [7]. Elevated levels of PCT, particularly associated with bacterial infections and systemic inflammatory response syndrome, make it a valuable marker for diagnosis of neonatal sepsis [17]. Consistently, in this study, PCT was significantly elevated in LOS patients, and with high performance in diagnosing LOS. CRP, an acute-phase protein, is significantly elevated in septic patients. However, the sensitivity of CRP for detecting LOS is lower due to the delay between the onset of sepsis and the increase in CRP levels, as well as the fact that CRP can also increase under other infectious and non-infectious conditions [18]. In this study, CRP was significantly increased in LOS patients, but it had a lower sensitivity (78.7%) in ROC curve analysis for distinguishing LOS. Our study also confirmed inflammatory cytokines including IL-6, IL-1β, and IL-12 as important predictors of LOS. IL-6, a multifunctional cytokine, plays a key role in the acute-phase response and has been widely studied as a biomarker for sepsis [19]. Both IL-1β and IL-12 are involved in immune response modulation, further reinforcing the role of cytokine profiles in sepsis diagnosis [20]. Elevated levels of these cytokines indicate an enhanced inflammatory response, characteristic of sepsis. Lac levels were significantly higher in the LOS group, reflecting the common metabolic disturbances and tissue hypoxia in septic patients [21]. High Lac is a well-documented marker of sepsis severity and prognosis, highlighting its importance in the diagnosis and management of sepsis in the PICU [21]. As is well known, the composite biomarker model exhibits better performance than a single indicator in diagnosing sepsis. Our data showed that the combined use of CRP, Lac, and IL-1β display the highest diagnostic accuracy for LOS.

This study further explored the potential role of biomarkers in distinguishing LOS caused by Gram-negative and Gram-positive bacteria. Our results indicated that Gram-negative bacterial infections were associated with higher levels of PCT, and PCT showed moderate diagnostic performance in differentiating LOS caused by Gram-positive and Gram-negative bacteria. This aligned with findings by Guo, et al. [22]which highlighted elevated PCT levels as a useful marker for diagnosing Gram-negative bacterial infections. Gram-negative bacterial infections are typically associated with higher inflammatory responses and endotoxin release, leading to significant increases in PCT levels [23]. High levels of PCT clinically often indicate the need for rapid and aggressive antibiotic treatment to combat this type of bacteria, reducing the risk of mortality and complications. This finding suggested that PCT could have clinical utility in predicting infection types, aiding in more precise antibiotic therapy in LOS patients [24].

However, this study also has limitations. Firstly, this study is a single center retrospective study, which may limit the generalizability of the research results. Additionally, reliance on electronic medical records might lead to incomplete data collection, and some potential confounding factors could not be controlled. In addition, we also need to include a wider range of biomarkers, such as neutrophil CD64, to more comprehensively evaluate sepsis progression. Future studies should validate our findings through prospective, multicenter studies and explore the dynamic changes in biomarker levels during the disease process. Understanding the underlying pathophysiological mechanisms driving biomarker changes in sepsis could also provide insights into new therapeutic targets and strategies.

Conclusion

This study highlighted the diagnostic utility of PCT, CRP, IL-6, IL-1β, IL-12, and Lac as biomarkers for LOS. The combined diagnostic model of CRP, Lac and IL-1β provided a highly sensitive and specific approach for diagnosing LOS within pediatric intensive care settings. Additionally, PCT was not only a crucial biomarker for diagnosing LOS, but variations in its levels can also help clinicians differentiate LOS caused by Gram-negative and Gram-positive bacteria, thereby guiding more targeted treatment. Ongoing research and validation of these biomarkers are essential for refining diagnostic protocols for neonatal LOS and enhancing patient care.

Authors’ contributions

Y.S. proposed the research design, prepared all material, performed data analysis and wrote the original draft. G.L. supervised the process, validated the analysis and reviewed the draft. All authors have read and approved the final version of the draft.

Funding

Not applicable.

Data availability

The datasets generated and/or analysed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

The study was approved by the Ethics Committee of Northwest Women and Children’s Hospital. Due to the retrospective nature of the study, consent to participate was waived by the Ethics Committee of Northwest Women and Children’s Hospital.

Competing interests

The authors declare no competing interests.