This prospective case-control study complied with the Declaration of Helsinki and was approved by the ethics committee of the hospital. Written informed consent was obtained from the legal guardians of all participants. For healthy controls, only a single non-invasive salivary sample and routine clinical/delivery data were obtained; no additional procedures were performed.

Because prior data for salivary biomarkers in EONP were unavailable, we used a precision-based approach informed by related neonatal pneumonia/sepsis studies reporting single-marker sensitivities/specificities around 0.75–0.90 (e.g., tracheal-aspirate presepsin in neonatal pneumonia (19); salivary CRP/platelet-based indices in late-onset neonatal pneumonia (20, 21). We targeted sensitivity and specificity of about 0.80 with 95% confidence-interval half-widths of 0.07–0.08 (two-sided α = 0.05). Standard formulas indicated a minimum of approximately 100 EONP cases and ∼120–130 controls to meet these precision goals; accordingly, we enrolled 100 EONP infants and 126 healthy controls. All enrolled participants provided analyzable saliva samples, so no additional allowance for dropout was required.

A total of 226 neonates were enrolled between June 2022 and June 2025, including 100 neonates diagnosed with EONP (2, 22) and 126 healthy controls. All enrolled infants had a gestational age of at least 34 weeks and were delivered in the study hospital. EONP cases were included if they met all of the following criteria: (1) onset or progressive worsening of respiratory distress within 72 h after birth, including a sustained respiratory rate >60 breaths/min for ≥2 h and/or clinical signs such as grunting, nasal flaring, chest retractions, and an increased requirement for oxygen or respiratory support [e.g., supplemental oxygen, continuous positive airway pressure (CPAP), or mechanical ventilation] to maintain adequate oxygen saturation; (2) chest radiograph and/or lung ultrasound demonstrating new or progressive patchy or diffuse infiltrates, consolidation, or reticulonodular changes consistent with infectious pneumonia; (3) at least one laboratory indicator of infection: abnormal peripheral white blood cell count and/or neutrophil differential [e.g., neutropenia or neutrophilia, or an immature-to-total neutrophil ratio (I/T ratio) > 0.16], elevated C-reactive protein (CRP ≥ 10 mg/L), elevated procalcitonin (PCT ≥ 0.5 ng/mL), or a positive blood and/or airway culture yielding a pathogen compatible with the clinical presentation; (4) complete clinical and laboratory data available; and (5) salivary samples obtained before initiation of any systemic antibiotic therapy, with sampling performed within 2 h after the first definitive diagnosis. Infants with or without concomitant bacteremia and early-onset neonatal sepsis were eligible, provided that they fulfilled all of the above clinical and radiologic criteria for early-onset neonatal pneumonia.

Healthy controls were neonates who had no clinical signs of infection or respiratory distress within the first 72 h after birth and no abnormal findings on physical examination or routine in-hospital investigations. None of the healthy controls received systemic antibiotic therapy before or at the time of saliva sampling. Healthy controls had complete basic clinical data and provided adequate salivary samples within a postnatal time window comparable to that of the EONP group.

The following exclusion criteria were applied to both the EONP and healthy control groups: (1) a definite diagnosis of noninfectious pulmonary disease, such as classic respiratory distress syndrome (RDS) or transient tachypnea of the newborn (TTN), in which the clinical course and imaging findings were not compatible with infectious pneumonia; (2) presence of major congenital cardiopulmonary malformations or genetic syndromes (e.g., complex congenital heart disease, severe pulmonary hypoplasia, congenital diaphragmatic hernia) that could substantially affect respiratory function or confound outcome assessment; (3) severe perinatal asphyxia (e.g., 5 min Apgar score ≤3 and/or prolonged chest compressions with epinephrine during resuscitation) with persistent, marked multiorgan dysfunction, making it difficult to distinguish pneumonia-related respiratory failure from hypoxic—ischemic injury; (4) confirmed or strongly suspected purulent meningitis at admission, defined by markedly abnormal cerebrospinal fluid cell counts, protein and glucose levels, and/or a positive cerebrospinal fluid culture; (5) infants fulfilling the clinical diagnostic criteria for EONP but without radiologic evidence of pneumonia on chest radiograph and/or lung ultrasound; (6) failure to obtain a salivary sample or provision of a salivary sample of inadequate quality for analysis; (7) coexistence of other severe underlying diseases, such as severe primary immunodeficiency or active malignancy; and (8) refusal of parents or legal guardians to permit use of the infant's medical data or failure to obtain written informed consent.

Salivary samples were collected after a definitive diagnosis of EONP had been established and before initiation of systemic antibiotic therapy. All sampling procedures were performed by neonatologists or neonatal nurses who had received standardized training. Before each procedure, hand hygiene was performed and sterile gloves were worn.

Oral samples were obtained using medical nylon flocked swabs (FLOQSwabs®, Copan, Brescia, Italy). In line with the approach described by Wunderlich et al. (23), and without interfering with clinical care, sampling was scheduled, whenever feasible, at least 30 min after the last feeding; in infants receiving donor human milk, sampling was preferably performed at least 60 min after feeding to minimize potential interference from residual breast milk or formula.

During collection, the infant's mouth was gently opened, and the swab was placed along the inner cheek and gumline on both sides. The swab was slowly rotated along the oral mucosa, approximately 5–10 rotations per side, for a total sampling time of about 10–20 s, avoiding the posterior tongue and pharynx to reduce gagging and discomfort. In infants with copious oral secretions, excess surface fluid was gently blotted with sterile gauze before swabbing.

Immediately after collection, the swab tip was placed into a pre-labeled 1.5 mL RNase-free microcentrifuge tube, the shaft was snapped off, and the tube was tightly capped. Then, 300–500 µL of prechilled normal saline was added to the tube and the contents were gently vortexed to ensure adequate elution of saliva into the preservation solution. All samples were temporarily stored at 4 °C and processed as soon as possible, within 4 h of collection, by centrifugation at 10,000 × g for 5 min at 4 °C. The supernatant was then aliquoted into RNase-free microcentrifuge tubes and stored at −80 °C until batch analysis. Each aliquot was thawed only once for biomarker measurement. Biomarker results are reported as concentrations in the eluted supernatant; because the original saliva volume absorbed by the swab was not directly quantified (e.g., by gravimetric weighing), the absolute concentrations should be interpreted as semi-quantitative.

Salivary PTX3, calprotectin, and IL-8 concentrations were measured using commercial ELISA kits according to the manufacturers' instructions: PTX3 with the Quantikine™ Human Pentraxin 3 Immunoassay (R&D Systems, Cat. DPTX30B); calprotectin with the Quantikine® Human Calprotectin Heterodimer Immunoassay (R&D Systems, Cat. DS8900); and IL-8 with the LEGEND MAX™ High Sensitivity Human IL-8 ELISA Kit (BioLegend, Cat. 431517).

For PTX3, the limit of detection was approximately 0.014 ng/mL, with a calibration range of 0.313–20 ng/mL, and intra- and inter-assay coefficients of variation of 1.2%–2.6% and 4.4%–6.9%, respectively. For calprotectin, the limit of detection was approximately 0.086 ng/mL, the calibration range was 0.625–40 ng/mL, and intra- and inter-assay coefficients of variation were 2.7%–4.5% and 3.2%–5.8%, respectively. For IL-8, the limit of detection was approximately 0.235 pg/mL, the calibration range was 0.78–50 pg/mL, and intra- and inter-assay coefficients of variation were 4.3%–5.3% and 5.7%–6.3%, respectively.

All standards and samples were assayed in duplicate. Laboratory personnel performing the ELISA assays were blinded to the clinical status and group allocation of the infants.

Continuous variables were assessed for distribution (Shapiro–Wilk). Skewed data are presented as median (IQR) and compared with the Mann–Whitney U test; categorical variables are presented as n (%) and compared with the χ² test or Fisher's exact test, as appropriate. Associations were examined using Spearman rank correlations. To assess whether salivary biomarkers were independently associated with EONP after accounting for potential confounders, we fitted a multivariable binary logistic regression model with EONP status as the dependent variable. Predictors included salivary biomarkers, with adjustment for gestational age and maternal intrapartum intravenous antibiotic prophylaxis. Diagnostic performance was evaluated using receiver operating characteristic (ROC) curve analysis, with discrimination quantified by the area under the ROC curve (AUC) and its 95% confidence interval (CI). For single biomarkers, the optimal cutoff value was selected by maximizing the Youden index (J = sensitivity + specificity−1). At the chosen cutoff, we report sensitivity and specificity, positive and negative likelihood ratios (LR+ and LR−), and positive and negative predictive values (PPV and NPV). AUCs and 95% CIs were estimated using the DeLong method. A combined model incorporating salivary PTX3, calprotectin, and IL-8 was developed using multivariable logistic regression to generate individual predicted probabilities. Apparent (development) performance was calculated by fitting the logistic regression model on the full development dataset and computing the ROC/AUC based on the resulting in-sample predicted probabilities. Internal validation was performed using out-of-fold predicted probabilities obtained from leave-one-out cross-validation (LOOCV) and 5-fold cross-validation. All tests were two-sided, with a significance level of α = 0.05. Analyses were performed using SPSS Statistics version 22.0 (IBM Corp.) and GraphPad Prism version 8.

The EONP and healthy control groups were broadly comparable in maternal and neonatal baseline characteristics (Table 1), including maternal age, parity, gestational diabetes, hypertensive disorders of pregnancy, gestational age, birth weight, sex, mode of delivery, 5 min Apgar scores, postnatal age at saliva sampling, and axillary temperature (all P > 0.05). As expected, intrapartum risk factors for early-onset infection were more frequent in the EONP group. Clinical chorioamnionitis was present in 25.0% of EONP pregnancies vs. 3.2% of healthy controls, intrapartum fever ≥38 °C in 22.0% vs. 2.4%, and premature rupture of membranes ≥18 h in 28.0% vs. 5.6% (all P < 0.001), respectively. Intrapartum intravenous antibiotic prophylaxis was administered to 52.0% of mothers in the EONP group and 23.8% in the healthy control group (P < 0.001). At the time of saliva sampling, infants with EONP had higher respiratory rates (72 breaths/min vs. 47 breaths/min) and heart rates (140 beats/min vs. 130 beats/min), and lower room-air SpO₂ (89% vs. 98%) than healthy controls (all P < 0.001), consistent with their underlying respiratory compromise.

In the EONP group (Table 2), signs of respiratory distress were frequent: nasal flaring in 75/100 infants (75.0%), chest wall retractions in 73 (73.0%), grunting in 62 (62.0%), and central cyanosis in 21 (21.0%). Inflammatory and hematologic markers were clearly elevated. Median high-sensitivity CRP was 45.0 mg/L (IQR 32.45–58.50), serum procalcitonin 4.65 ng/mL (2.90–7.10), white blood cell counts 25.5 × 10⁹/L (17.45–31.65), and absolute neutrophil count 12.85 × 10⁹/L (9.50–17.55). The immature-to-total neutrophil ratio had a median of 0.27 (0.20–0.36). Median platelet count was 182.0 × 10⁹/L (122.5–240.0), and median serum IL-6 concentration was 239.2 pg/mL (160.4–318.7).

Microbiologically, 34 infants (34.0%) had a positive blood culture, 37 (37.0%) had a positive upper airway bacterial culture, and respiratory pathogen PCR identified at least one pathogen in 44 (44.0%) cases. Chest radiographs most often showed patchy infiltrates in 60 infants (60.0%), with diffuse infiltrates in 22 (22.0%), consolidation in 32 (32.0%), and reticulonodular changes in 18 (18.0%); some infants had more than one radiographic pattern.

Compared with healthy controls, infants with EONP had substantially higher salivary PTX3, calprotectin, and IL-8 levels (Figures 1A–C). For PTX3, the median concentration in the EONP group was 2.11 ng/mL (IQR 1.36–3.08) vs. 0.79 ng/mL (IQR 0.52–1.25) in healthy controls (Mann–Whitney P < 0.001). For calprotectin, the median was 11.65 ng/mL (IQR 8.47–16.93) in EONP infants and 3.07 ng/mL (IQR 2.09–4.53) in healthy controls (P < 0.001). For IL-8, the median was 15.02 pg/mL (IQR 10.16–25.71) in the EONP group compared with 4.67 pg/mL (IQR 3.11–6.34) in healthy controls (P < 0.001). Together, these findings indicate that all three salivary biomarkers are significantly elevated in EONP and show substantial separation between cases and healthy controls. A multivariable binary logistic regression model was fitted to evaluate whether the salivary biomarkers were independently associated with EONP after adjustment for gestational age and maternal intrapartum IV antibiotic prophylaxis (Supplementary Table S1). After adjustment, salivary calprotectin remained an independent predictor of EONP (B = 0.772; adjusted OR = 2.163, 95% CI 1.639–2.855; P < 0.001), as did salivary IL-8 (B = 0.404; adjusted OR = 1.497, 95% CI 1.226–1.828; P < 0.001). In contrast, salivary PTX3 was not statistically significant in the adjusted model (B = 0.446; adjusted OR = 1.562, 95% CI 0.537–4.544; P = 0.413). Neither gestational age category (adjusted OR = 1.410, 95% CI 0.196–10.159; P = 0.733) nor intrapartum IV antibiotic prophylaxis (adjusted OR = 2.615, 95% CI 0.543–12.599; P = 0.231) was significant in this model. To assess robustness, we additionally performed bootstrap resampling (1,000 samples). The bootstrap results were consistent with the primary analysis: the regression coefficients for calprotectin (bootstrap B 95% CI 0.573–1.262; P = 0.001) and IL-8 (bootstrap B 95% CI 0.232–0.710; P = 0.001) remained significant and their confidence intervals did not cross zero, supporting the stability of these associations. In contrast, PTX3 did not show a significant association on bootstrap validation (bootstrap P = 0.350).

Using ROC-derived optimal cutoffs, all three salivary biomarkers showed good to excellent diagnostic performance for identifying EONP from healthy controls (Figures 1D–F, Table 3). For PTX3 (cutoff ≥ 1.38 ng/mL), the AUC was 0.865, with a sensitivity of 75% and a specificity of 84.1% (Youden index 0.59). The corresponding positive and negative likelihood ratios were 4.7 and 0.30, and the positive and negative predictive values in this cohort were approximately 78.9% and 80.9%, respectively. Salivary calprotectin (cutoff ≥ 5.49 ng/mL) yielded the highest diagnostic accuracy among single markers, with an AUC of 0.967, sensitivity 93%, specificity 92.1%, and a Youden index of 0.85. The positive and negative likelihood ratios were 11.8 and 0.08, with a PPV of about 90.3% and an NPV of 94.3%. For IL-8 (cutoff ≥ 8.69 pg/mL), the AUC was 0.930, with a sensitivity of 85% and specificity of 90.5% (Youden index 0.76). The positive and negative likelihood ratios were 9.0 and 0.17, and the corresponding PPV and NPV were 87.7% and 88.4%. A multivariable logistic regression model combining salivary PTX3, calprotectin, and IL-8 showed excellent discrimination for identifying EONP in the overall cohort (Figure 1G). The apparent AUC estimated on the development dataset was 0.978. Bootstrap resampling (1,000 iterations) yielded a mean AUC of 0.979 with a percentile 95% confidence interval of 0.963–0.991. The mean optimism estimated by bootstrap was 0.0015, resulting in an optimism-corrected AUC of 0.976, indicating minimal optimism. Consistent results were obtained using cross-validation: the LOOCV AUC was 0.974, and the 5-fold cross-validated AUC was 0.972 ± 0.024. Collectively, these findings support stable model discrimination with limited evidence of overfitting.

Salivary biomarkers were strongly inter-correlated (PTX3—calprotectin r = 0.75; PTX3—IL-8 r = 0.72; calprotectin—IL-8 r = 0.78, all P < 0.001; Figure 2A), whereas no significant correlations were observed among these markers in the healthy control group (all P > 0.05; Figure 2B). Each marker tracked systemic inflammation (all P < 0.05; Figure 2C), showing the highest correlations with hs-CRP (PTX3 r = 0.84; calprotectin r = 0.90; IL-8 r = 0.85), and moderate—strong correlations with PCT (0.72, 0.74, 0.69) and IL-6 (0.58, 0.67, 0.70). Associations with leukocyte indices were moderate (WBC 0.55, 0.59, 0.51; ANC 0.62, 0.70, 0.68; I/T ratio 0.64, 0.68, 0.60). All three were inversely related to platelet count with relatively strong negative correlations (−0.75, −0.66, −0.69), supporting that salivary PTX3, calprotectin, and IL-8 mirror systemic inflammatory burden.

Among infants with EONP, salivary PTX3, calprotectin, and IL-8 were higher in culture-positive than culture-negative cases (Figures 3A–C). Median (IQR) PTX3 was 2.97 (1.82–4.62) ng/mL vs. 1.78 (1.22–2.46) ng/mL (Mann–Whitney P < 0.001); calprotectin 15.30 (10.12–19.02) ng/mL vs. 10.31 (8.17–13.84) ng/mL (P = 0.003); and IL-8 22.34 (11.90–32.42) pg/mL vs. 13.58 (9.50–18.24) pg/mL (P = 0.002). In EONP infants, salivary biomarkers showed fair discrimination for blood-culture—positive bacteremia. Using ROC-optimized thresholds (PTX3 ≥ 2.51 ng/mL; calprotectin ≥14.57 ng/mL; IL-8 ≥ 18.97 pg/mL, Figures 3D–F, Table 3), performance estimates were: PTX3 AUC = 0.725, sensitivity = 64.71%, specificity = 80.30%; calprotectin AUC = 0.682, sensitivity = 55.88%, specificity = 78.79%; IL-8 AUC = 0.689, sensitivity = 58.82%, specificity = 80.30%). Using the combined salivary biomarker model score to discriminate culture-positive from culture-negative EONP cases demonstrated moderate performance (Figure 3G). The AUC was 0.702 using the apparent predicted probability. Internally validated predictions showed nearly identical discrimination, with an AUC of 0.706 for 5-fold cross-validated probabilities and 0.702 for LOOCV probabilities, supporting the robustness of this enrichment signal. These findings suggest that, within confirmed EONP, the salivary panel may help enrich for culture-positive cases, although discrimination is limited compared with its performance for EONP detection.