We conducted a prospective cohort study in the First Hospital of Tsinghua University in China from March 2021 to March 2022. The study received approval from the ethics committee of the First Hospital of Tsinghua University and conformed to the principles of the Declaration of Helsinki. All data were de-identified before analysis, and access was restricted to authorized investigators. Written informed consent was obtained from a parent or legal guardian before any study procedures were undertaken.

Infants were eligible if they were born at 24 + 0–33 + 6 weeks' gestation, admitted to the NICU within 6 h of delivery, and met standard clinical and radiographic criteria for RDS as assessed by the attending neonatologist. RDS was defined by: (1) onset of respiratory distress (tachypnea, nasal flaring, chest retractions and/or grunting) within the first 6 h of life; (2) requirement for supplemental oxygen or respiratory support to maintain target saturations; and (3) a characteristic chest radiograph showing a diffuse reticulogranular “ground-glass” pattern with air bronchograms and reduced lung volumes. Infants with clinical and radiographic features more suggestive of transient tachypnea, meconium aspiration, primary pneumonia or other non-RDS diagnoses were not enrolled. We excluded infants with major congenital malformations (such as congenital diaphragmatic hernia, pulmonary hypoplasia or complex congenital heart disease), proven TORCH infection, chromosomal disorders or severe perinatal asphyxia defined by a five-minute Apgar score of three or less. Families who declined consent or whose infants lacked either the lung ultrasound or procalcitonin measurement were also excluded.

We prespecified a recruitment target to compare discriminatory performance (AUC) of the combined LUS plus PCT model against a single-marker approach. Assuming an area under the receiver-operating-characteristic curve (AUC) of 0.75 for the leading single marker and 0.85 for the combined model (13), with a two-sided alpha of 0.05% and 80% power, Hanley and McNeil's method indicated that at least 154 infants with and 154 without the composite endpoint were required. Allowing for a 10% attrition rate, we set a recruitment target of 343 infants. Because recruitment was limited by the prespecified study period, 290 eligible infants were ultimately enrolled, of whom 110 experienced the composite endpoint.

Lung ultrasound score (LUS): LUS was performed within 6 h of birth by credentialed sonographers blinded to PCT results and to follow-up outcome status (The 6 h window primarily reflects the timing needed to complete the bedside LUS after initial cardiorespiratory stabilization and NICU admission). Scanning used a GE LOGIQ S8 9.0-14.0Mhz linear transducer following a standardized 12-zone protocol (anterior, lateral, and posterior zones bilaterally). Each zone was scored 0–3 (0 = normal aeration; 1 = interstitial pattern; 2 = severe interstitial pattern/coalescent B-lines; 3 = consolidation), yielding a total score range 0–36 as previously described (15). Inter-rater reliability was assessed in a pilot set (quadratic-weighted κ = 0.83). The standardized 12-zone LUS score was calculated and recorded for research purposes and was not communicated to the treating clinical team in real time, and it did not influence bedside clinical decision-making.

Umbilical cord-blood PCT: Cord blood was collected immediately after clamping, prior to postnatal therapeutic interventions, centrifuged within 30 min, and PCT was measured using the BRAHMS PCT III chemiluminescent platform. Analytical range was 0.02–100 ng/mL within-run CV <6%. For modeling, PCT was log-transformed due to right-skewness. Cord-blood PCT testing was performed for the purposes of this research protocol, and results were not returned to the treating team in real time and therefore did not influence clinical management.

Maternal and neonatal baseline data, including gestational age, birth-weight, sex, mode of delivery, antenatal corticosteroid exposure, maternal intrapartum antibiotic exposure, suspected/clinical chorioamnionitis or maternal fever/infection, prolonged rupture of membranes, hypertensive disorders of pregnancy, gestational diabetes, surfactant administration, ventilation parameters and early-onset infection status, were extracted from the electronic medical record. Dual data entry with automated logic checks ensured accuracy, and discrepancies were resolved against source documents.

Early-onset infection: Early-onset infection was defined as clinician-diagnosed infection/sepsis within 72 h of birth based on institutional criteria (maternal risk factors and neonatal clinical signs), with or without culture confirmation, and antibiotic treatment consistent with NICU protocol.

The primary outcome was a composite of bronchopulmonary dysplasia, assessed at 36 weeks' postmenstrual age using a severity-based definition aligned with contemporary NICHD/NRN criteria (16), intraventricular hemorrhage grade III or IV, necrotizing enterocolitis stage II or higher, culture-proven sepsis occurring after 72 h of age or all-cause death within 12 months corrected age. Secondary outcomes comprised each component of the composite, total days of hospitalization, duration of mechanical ventilation and re-hospitalization within the first year. Secondary analyses also included deriving clinically usable cut-off values for the combined model and examining additional independent risk factors.

Normality of continuous variables was assessed using visual inspection and the Shapiro–Wilk test to determine whether results are summarized as mean ± SD or median (IQR), and categorical variables as frequencies with percentages. Group differences were assessed using Student's t-test or the Mann–Whitney U-test for continuous data and χ² or Fisher's exact test for categorical data. Missing values, which accounted for less than five per cent of any variable, were addressed with multiple imputation by chained equations (m = 20).

A logistic regression model combined LUS and log-transformed PCT to generate individual predicted risk probabilities. Discrimination was quantified by the AUC with 95% CI. AUCs were compared using paired DeLong tests. Multiple comparisons were controlled using the Benjamini–Hochberg false discovery rate (FDR). The Youden index was used to define a single operating point for sensitivity/specificity and to construct a binary “high-risk vs. low-risk” classification used for Kaplan–Meier displays and subgroup risk ratios.

Time-to-event analysis: Cox proportional hazards regression estimated associations with time to first composite event, adjusting for clinically relevant covariates. Proportional hazards assumptions were assessed using Schoenfeld residuals.

Internal validation and sensitivity analyses: Model performance was internally validated by 1,000 bootstrap resamples with optimism correction. Prespecified sensitivity analyses included using a 6-zone LUS scheme and restricting analyses to complete cases.

A total of 343 preterm infants were screened. Fifty-three were excluded (35 not meeting eligibility criteria and 18 missing either LUS or PCT), leaving 290 (85.3%) infants with both index tests within 6 h for analysis (Figure 1). All included infants completed 12-month follow-up.

Infants reaching the composite endpoint (110/290, 37.9%) had lower gestational age (27.8 ± 2.5 vs. 29.5 ± 2.0 weeks, p = 0.001) and lower birthweight [1,125 [741–1,424] vs. 1,247 [873–1,548] g, p = 0.031] than those who remained event-free. The time to LUS examination was performed similarly after birth [median 2.2 (1.4–3.3) vs. 2.4 (1.6–3.5) hours, p = 0.23], while cord-blood PCT was collected at delivery. They more often had a 1 min Apgar score <7 (32.7% vs. 17.8%, p = 0.013). Repeat surfactant was less frequent in the event group (28.2% vs. 45.0%, p = 0.004) (Table 1). Maternal/perinatal baseline factors were also summarized in Table 1.

The composite endpoint occurred in 110 infants, the median time to first event was 21 days (IQR 9–35), and 72% of first events occurred during the index hospitalization. Overall, 180 infants (62.1%) remained free of the composite endpoint through 12 months' corrected age.

Receiver-operating-characteristic analysis demonstrated AUCs of 0.76 (95% CI 0.70–0.82) for LUS alone and 0.78 (0.72–0.84) for PCT alone, the latter not significantly superior (DeLong p = 0.35, FDR-adjusted q = 0.350) (Figure 2 and Table 2). Combining LUS with log-transformed PCT in a logistic model increased the AUC to 0.87 (0.83–0.92), exceeding each single marker (both p < 0.001, q < 0.005). At the Youden-optimal threshold the combined model achieved 82% sensitivity, 80% specificity, a positive predictive value of 0.69 and a negative predictive value of 0.89, while calibration across risk deciles was excellent (Hosmer–Lemeshow χ² = 6.4, p = 0.60) (Table 2).

After adjustment for gestational age, birth-weight, repeat surfactant, and early-onset infection, a high combined risk score remained the strongest independent predictor of the composite endpoint [hazard ratio (HR) 2.9, 95% CI 2.0–4.1, q < 0.001]. Gestational age <28 weeks (HR 1.8, 95% CI 1.2–2.7, q = 0.009) and early-onset infection (HR 1.6, 95% CI 1.1–2.4, q = 0.025) were additional risk factors, whereas each 100-g increase in birth-weight reduced risk by 8 % (HR 0.92, 95 % CI 0.88–0.97). The proportional-hazards assumption was satisfied (global Schoenfeld p = 0.42) (Table 3 and Figure 3).

Bronchopulmonary dysplasia was the most common individual complication (62/290 cases, 21.4%), followed by culture-proven sepsis (46/290 cases, 15.9%), severe intraventricular hemorrhage (grade ≥ III; 18/290, 6.2%), and necrotizing enterocolitis (stage ≥ II; 14/290, 4.8%). High-risk infants, as defined by the combined score, had significantly higher incidences of every component, with risk ratios ranging from 1.9 for severe intraventricular hemorrhage (95% CI 1.0–3.5, q = 0.060) to 2.8 for all-cause death (95% CI 1.5–5.2, q = 0.008); they also experienced median hospital stays and ventilation durations that were 12 days and 3 days longer, respectively (both p < 0.01) (Table 4).

Results were robust when the 6-zone LUS scheme was substituted (combined AUC = 0.86), when only complete cases were analyzed (n = 277), and across gestational-age strata, with no significant interaction between risk score and gestational age (interaction p = 0.27). Bootstrap optimism-correction yielded an internally validated AUC of 0.86, confirming model stability.

No complications attributable to lung ultrasound scanning or cord-blood sampling were observed, and no protocol deviations occurred during the study period.