This retrospective study was carried out in a single-center pediatric intensive care unit (PICU) from November 2017 to December 2020, approved by the ethical committee of Shanghai Children’s Medical Center (Approval number: SCMCIRB-K2018030), and written informed consent was obtained from patients’ guardians prior to enrollment. All enrolled patients met the criterion of “single bacterial infection” and were diagnosed with sepsis. They gave specimens for cultures and underwent cytokine measurements and pathogenic screening tests within 1 h of ICU admission. Patients with clinical conditions such as viral and fungal infection, primary immune deficiency, primary hemophagocytic lymphohistiocytosis or cytokine release syndrome were excluded from this study. Patients who underwent operations or trauma within 7 days were excluded. Patients with prior exposure to corticosteroids, granulocyte colony-stimulating factor, or biological agents were excluded.

Microbiologically documented patients without infectious clinical signs were randomly enrolled and considered as the control group. These patients were not healthy controls, instead they were non-febrile, non-infectious, critically ill patients admitted to the PICU. All case group patients (n = 146) were categorized into subgroups: 1) bloodstream infection and non-bloodstream infection; 2) acute respiratory distress syndrome (ARDS) and non-ARDS; 3) new or progressive multiple organ dysfunction syndrome (NPMODS) and non-NPMODS; 4) severe sepsis and non-severe sepsis. Subgroup analysis aimed to eliminate the impact of other inflammatory responses on cytokine levels.

With improved understanding of the sepsis pathobiology, the definition of Sepsis-3 and Sequential Organ Failure Assessment (SOFA) score were expected to be widely adopted in patients with suspected infection. However, there is a lack of diagnostic criterion and modified Pediatric Sequential Organ Failure Assessment (pSOFA) for pediatric population. In this study, clinical sepsis-related diagnosis was based on the 2015 pediatric septic shock expert consensus announced by the Chinese Pediatric Society, Chinese Medical Association. This clinical consensus was widely abided by pediatricians in Mainland China and was formulated based on the quantitative indicators of pediatric sepsis mentioned in the 2012 SSC international guidelines for management of severe sepsis and septic shock (12). Pediatric Risk of Mortality III (PRISM-III) was calculated using the worst physiological and laboratory indicators collected within 24 h of admission (13). The ARDS diagnosis follows the 2015 International Expert Consensus on Pediatric Acute Respiratory Distress Syndrome (14). NPMODS was defined as new-onset organ dysfunction or progressive deterioration of injured organ within 7 days of sepsis diagnosis (15). Organ function was assessed dynamically within admission day 1 to day 7.

Microbiological diagnosis is based on specimen cultures. For patients with more than one positive result, we only chose the predominating bacterial strain [(higher colony forming unit CFU)] and preferred results obtained from sterile specimens (blood, cerebrospinal fluid). The microbiological diagnosis confirmed by other pathogen tests was considered as mixed infection. Based on these rules, 14 patients (9.6%) had a positive blood culture associated with other body fluid cultures. All of these patients had the same bacteria isolated in both cultures.

Blood samples (2 ml each) were obtained from patients within 1 h of PICU admission and stored in tubes containing potassium EDTA at room temperature. Samples were immediately sent to the clinical laboratory to perform cytokine measurements. The cytometric bead array (CBA) was used as the cytokine detection methodology, with BD™ CBA Human Th1/Th2/Th17 cytokine kit II (from BD Biosciences) for cytokine measurement. Before CBA assays begin, negative controls and standards diluted in serial concentrations were prepared for the purpose of standard curve for calibration and validation of linearity. Bead populations were mixed, resuspended (mixed with serum enhancement buffer after the supernatant was removed), and added into assay tubes accordingly. For samples with a known high cytokine concentration, samples were diluted to obtain several desired dilutions with the assay diluent. Bead populations ready in a filter plate were mixed with samples to form the bead array that resolved in a red channel of the flow cytometer.

For the principles of CBA, beads conjugated with specific antibodies (with discrete fluorescence intensities) were captured and incubated with detection reagent and test samples to form sandwich complexes (beads + detection reagent + analyte). Such complexes provide proportional fluorescent signal relative to the concentration of that cytokine. Compared with conventional ELISA, flow cytometry could perform multi-analyte assays to detect series of cytokines by acquiring one single sample. Finally, Th1/Th2 cytokine levels were then quantitatively determined by a flow cytometer (FACSCalibur, BD Biosciences), and graphical results were generated by FCAP Array™ software. For the clinical reference range, upper normal limits for IL-2, IL-4, IL-6, IL-10, TNF-α, and Interferon-γ (IFN-γ) are 8.84, 3.88, 15.80, 4.98, 5.97, and 16.20 pg/ml, respectively.

Statistical significance between categorical variables was compared using the Fisher’s exact test or chi-square (χ²) test, while the Mann–Whitney U test was used for skewed distribution continuous variables. Propensity score matching (PSM) was applied to balance baseline differences including comorbidities, PRISM-III, and critical condition (septic shock, severe sepsis) between the two groups. Match tolerance was 0.4. ROC curves and Youden index were used to determine diagnostic performances of clinical indicators. The correlation between IL-6 and IL-10 was compared using Spearman rank correlation analysis, while a binary logistic regression model was used to calculate the incidence of NPMODS. A two-sided p-value of <0.05 was considered statistically significant.

A total of 226 patients were enrolled, of whom 146 sepsis patients with microbiologically documented infection comprised the case group and 80 non-infectious patients belonged to the control group. According to specimen culture results, 89 patients (61%) were categorized as G- sepsis, while 57 patients (39%) had G+ sepsis. PSM was used to balance the baseline characteristics. Finally, 78 matched patients were eligible for statistical analysis. As shown in Table 1 , baseline characteristics were comparable between the two groups (p > 0.05). In our study, median age was 1 year 11 months, and patients from the preschool age group (38.5%) accounted the most ( Table 1 ). Most patients (51.3%) were admitted from the emergency department, followed by hospital transfer (24.4%) or general wards (24.4%). Comorbid condition was particularly prevalent (93.6%) with no significant differences between G+ and G- sepsis (p = 0.123). Nearly half of the patients (56.5%) had accompanying hematologic malignancies or under post-liver transplant/post-hematopoietic stem cell transplantation (HSCT) state. Other underlying comorbidities include cardiovascular (19.2%), hepatobiliary (9%), neuromuscular (3.8%), and respiratory diseases (2.6%) and inherited metabolic disorders (2.6%).

The most common primary site of infection was the respiratory tract (50%), followed by bloodstream (42.3%), gastrointestinal (5%), and thoracic/abdominal infections (2.6%). Majority of the patients presented with a mild PRISM-III score (73.1%) at sepsis onset. Only 26.9% of the patients gained a severe or extremely severe PRISM-III score, which was milder on average than the overall PRISM-III score distribution observed before PSM. For common critical condition and advanced supportive care, no statistical differences were observed between G+ and G- sepsis groups ( Table 1 ). In this study, 50% of patients experienced septic shock, while 71.8% of the patients met the criteria of severe sepsis. Such high incidence might be attributed to the high prevalence of comorbid conditions among the enrolled patients.

Among all Th1/Th2 cytokines, IL-2, IL-6, IL-10, TNF-α, and IFN-γ were significantly elevated in infectious patients (p < 0.001). This supports the assumption that significant elevation of Th1/Th2 cytokines should be observed in febrile infectious patients ( Figures 1A –F ) . IL-2, IL-4, TNF-α, and IFN-γ were all slightly elevated, while IL-6 and IL-10 exhibited an exponential elevation proven by the logarithmic transformation scale ( Figures 1E, F ). Medians of IL-6 levels in G+ and G- sepsis groups were 54.24 and 271.37 pg/ml, respectively. While medians of IL-10 in the two groups were 15.55 and 23.95 pg/ml, respectively ( Supplementary Table S1 ). This indicated that more than half of the G- sepsis patients had their IL-6 exceeding 270 pg/ml, while the remaining 30.8% patients had IL-6 levels exceeding 1,000 pg/ml. In the G+ sepsis group, IL-6 levels only exceeded 270 pg/ml in 25.6% of the patients, while 15.4% of patients exceeded 1,000 pg/ml.

In earlier studies, bacteremia is often the fundamental inclusion criterion, as it is the most common cause of severe sepsis in children with hematologic malignancies. However, few studies have investigated cytokine patterns in non-bloodstream infectious patients. In most ICUs, non-bloodstream infection dominates and possesses equal risk in developing septic shock and MODS. Therefore, it is necessary to set up subgroups for analysis. ARDS is a common complication in progressively deteriorating critically ill patients. Although the precipitating cause and pathophysiology of ARDS are complex, central to its pathogenesis is acute lung inflammatory injury. Involvement of robust inflammatory processes urges the setup of ARDS subgroups.

Distribution patterns of IL-6 and IL-10 in G+/G- sepsis patients were significantly different in both bloodstream and non-bloodstream infections (p < 0.05) ( Supplementary Table S2 ). Discrepancies in the extent of IL-6 and IL-10 elevations were observed between subgroups of the same bacterial Gram types. For instance, in G- bacterial infection, IL-6 levels in bloodstream infection could reach 1,000–1,200 pg/ml while IL-6 levels in the corresponding non-bloodstream infection was barely 200 pg/ml. However, in the same subgroup, differences in cytokine levels between G+ and G- sepsis still exist. This supports the hypothesis that the discriminating power of IL-6 and IL-10 for G-/G+ sepsis is independent of the site of infection (bloodstream or non-bloodstream infection).

Regarding combined indicators, “IL-6 IL-10 differences” were statistically different between bacterial Gram types in the two subgroup analyses ( Supplementary Tables S2 , S3 ). G- sepsis patients showed relatively higher “IL-6 IL-10 differences.” This suggests that the extent of IL-6 and IL-10 elevation in G- sepsis patients was comparable regardless of site of inflammatory responses. However, “IL-6 IL-10 ratio” was relatively lower in G- sepsis patients with bloodstream infection and ARDS. While in the non-bloodstream and non-ARDS group, “IL-6 IL-10 ratio” was higher in G- sepsis patients but without statistical significance. One possible explanation for this adverse result might be the existence of other inflammatory responses that leads to a distinct IL-6–IL-10 variation pattern within bloodstream infection and ARDS patients.

Next, we evaluated the influence of critical conditions on Th1/Th2 cytokine levels by categorizing infectious patients into four subgroups: severe sepsis, non-severe sepsis, NPMODS, and non-NPMODS. In severe sepsis patients (n = 73) ( Supplementary Table S4 ), no statistical significance was observed except for IL-10 (p = 0.034). This was consistent with a previous perspective stating that IL-10 was the most useful predictor for G- bacterial infection in critically-ill sepsis patients (16). Although the extent of the elevation was lesser than that observed in severe sepsis, cytokine patterns in G- bacterial infection with non-severe sepsis (n = 37) were comparable to those observed in case group analysis (p < 0.05).

In both NPMODS and non-NPMODS subgroups, median levels of IL-6 and IL-10 in patients with G- sepsis were higher than those with G+ sepsis ( Supplementary Table S5 ). In the NPMODS subgroup, however, only IL-10 exhibited statistical differences between G+ and G- bacterial infection (p < 0.005), while IL-6 was significantly elevated regardless of bacterial Gram type ( Supplementary Table S5 ).

Based on the variation pattern of IL-6 and IL-10, we used combined indicators “IL-6 IL-10 ratio” and “IL-6 IL-10 differences” to further describe cytokine patterns among G+/G- sepsis patients ( Supplementary Table S1 ). As shown in Figure 2 , both indicators varied in their limited range within G+/G- sepsis groups with statistical significance (p < 0.05) ( Supplementary Table S1 ). “IL-6 IL-10 ratio” represents the extent of IL-6 elevation and its fixated correlation with IL-10. “IL-6 IL-10 differences” further emphasized the prominent elevation. Almost all G- sepsis patients experience a marked elevation of IL-6 compared to IL-10 at admission ( Figure 2A ), which was less commonly observed in the G+ sepsis group.

For non-NPMODS patients, “IL-6 IL-10 ratio” after G- bacterial infection was statistically higher (p < 0.05) with a non-overlapping confidence interval (4.295, 95% CI 3.16–8.27) compared to G+ bacterial infection (2.470, 95% CI 1.37–3.18) ( Figure 2A ). For G- NPMODS sepsis patients, “IL-6 IL-10 ratio” varied in a narrow range similar to that observed in G- non-NPMODS sepsis. In contrast, G+ NPMODS sepsis patients exhibited dispersed levels that implied an inconsistent relationship between IL-6 and IL-10 in such group of patients. Most G+ NPMODS sepsis patients experienced significantly higher levels of pro-inflammatory IL-6 without relevant IL-10 elevation.

Next, we look at the usefulness of “IL-6 IL-10 differences” ( Figure 2B ). For G- sepsis patients, the observed narrower confidence intervals (compared to G+ sepsis patients) in “IL-6 and IL-10 differences” proved the efficacy of using such indicator to describe the IL-6–IL-10 correlation. Similarly, a statistical difference was observed between Gram bacteria types (p < 0.05).

Binary scatter plots in Figure 3 further proved a monotonic linear relationship between IL-6 and IL-10 among G+ and G- sepsis patients. In Spearman correlation rank analysis, IL-6 was positively correlated with IL-10 (r = 0.654, p < 0.001) ( Figure 3 ), in which G- sepsis patients showed a stronger linear correlation between the two cytokines.

To discover the diagnostic performances of IL-6 and IL-10 in identifying G- bacterial infection, ROC curves were built based on IL-6, IL-10, C-reactive protein (CRP), and procalcitonin (PCT) of all of the matched patients (n = 78) ( Figure 4 ). Area under the curve (AUC) values of IL-6, IL-10, C-reactive protein (CRP), and procalcitonin (PCT) were 0.679 (95% CI 0.561–0.798), 0.637 (95% CI 0.512–0.762), 0.616 (95% CI 0.489–0.742), and 0.623 (95% CI 0.499–0.747), respectively ( Table 2 ). AUC values of IL-6 and IL-10 were superior than traditional indicators. Using Youden index, the optimal cutoff values of IL-6 and IL-10 were 76.77 and 18.90 pg/ml, in which specificity was 61.5% while sensitivity was 69.2% and 66.7%, respectively.

To explore the predictive power of cytokines on sepsis prognosis, patients in the NPMODS subgroup (n = 67) were separately analyzed. ROC curves were shown in Figure 5 , in which the AUC values for IL-6, IL-10, CRP, and PCT were 0.834 (95% CI 0.766–0.902), 0.781 (95% CI 0.701–0.860), 0.766 (95% CI 0.688–0.845), and 0.725 (95% CI 0.639–0.810), respectively. IL-6 and IL-10 showed remarkably higher AUC and non-overlapping optimal cutoff values ( Table 2 ). Although cytokines were measured at early onset of sepsis, IL-6 and IL-10 were sufficient enough for the prediction of NPMODS, in which their optimal cutoff values outperform that observed in diagnosing G- sepsis.

Binary logistic regression model was used to investigate the association of IL-6 and IL-10 with NPMODS incidence. Both cytokines were transformed into categorical variables presented as percentiles, in which the <P²⁵ was used as a reference for comparison. Factors such as age, gender, bacteria types, and underlying diseases were included for adjustment. In this model, elevation of IL-6 and IL-10 showed a linear relationship with increased risk of NPMODS incidence ( Table 3 ). With elevated levels of IL-6 and IL-10, the risk of NPMODS increased accordingly [higher odds ratio (OR)].