This was a prospective, observational, multicenter study conducted in two academic tertiary neonatal intensive care units (NICUs). The protocol was approved by the institutional review boards at each site, and written informed consent was obtained from the legal guardians of all infants. This study has been approved by the Ethics Committee of Shangrao Maternal and Child Health Hospital, with the ethical approval number SRFB20231206037, dated December 6, 2023.

Eligible participants were term or post-term neonates (gestational age 37–42 weeks) admitted between 2019 and 2025. Inclusion criteria were: (1) grade III meconium-stained amniotic fluid [17]; (2) postnatal respiratory distress requiring NICU respiratory support (noninvasive ventilation or invasive mechanical ventilation); and (3) clinical diagnosis of MAS. Exclusion criteria were: (1) healthy rooming-in newborns not requiring NICU care; (2) major congenital disorders or severe congenital malformations; and (3) congenital lung disease.

Based on pilot data, an a priori sample size calculation was performed using G*Power 3.1. With α = 0.05, power (1−β) = 0.80, an anticipated AUC of 0.98 for the LUS score, and a null AUC of 0.90, the minimum required sample size was 156. Allowing for a 20 % loss to follow-up or incomplete data, the authors the authors planned to enroll 200 infants. Ultimately, 218 infants were included, meeting and exceeding power requirements.

Timing of Lung Ultrasound Examination: Bedside lung ultrasound was performed immediately upon admission to the NICU, with a median time of 10 min from admission to completion of the lung ultrasound (The maximum duration does not exceed 30 min). This rapid assessment captured the infant's pulmonary status prior to any major therapeutic interventions beyond initial stabilization. The FiO₂ value recorded reflected the fraction of inspired oxygen required at the exact moment of the lung ultrasound examination. Surfactant administration was always performed after the completion of the baseline assessment. The median time from admission to surfactant administration was 1 hour. To ensure timeliness, each center guaranteed 24-hour on-site coverage by at least one qualified ultrasonographer. Examinations used a Mindray M9 portable system, preferentially with a high-frequency linear probe (L12–4 s), switching to a low-frequency convex probe (C 11-3s) when greater penetration was required. Dynamic cine loops were archived for subsequent analysis. Lung zoning followed recent international expert consensus, using the anterior and posterior axillary lines to divide each hemithorax, for a total of ten neonatal-specific assessment zones across both lungs, accommodating the heterogeneous distribution typical of MAS (The protocol actually used in practice was the 6-zone method) [18,19].

MAS-specific LUS scoring criteria (Brat-based system adapted to MAS pathology)

• 0: normal ventilation with lung sliding and A-lines or < 3 isolated B-lines;

• 1: moderate reduction in ventilation with ≥ 3 B-lines and/or small focal consolidation;

• 2: severe reduction with confluent B-lines or larger consolidation involving more than two zones in aggregate;

• 3: extensive consolidation with air bronchograms and/or atelectasis, tissue-like pattern.

Each zone was scored 0–3 (total 0–18). All anonymized LUS cine loops were stored on a secure server and were independently scored by two experienced raters who were blinded to all clinical information, including the infant's respiratory support level, FiO₂ requirements, blood gas results, and whether surfactant was eventually administered. Discrepancies between raters were resolved by consensus or by a third senior reviewer. Inter-rater reliability was assessed on a random subset of 30 infants with double-blind rescoring, yielding a Fleiss’ kappa of 0.85 (95 % CI, 0.76–0.94), indicating high reproducibility. To ensure objectivity, the LUS examinations were performed by a dedicated team of trained sonographers who were not involved in the clinical management of the infants. The results of the LUS examination, including the real-time images and the subsequent scoring, were not disclosed to the attending clinical team responsible for making decisions regarding respiratory support and surfactant administration.

Surfactant administration followed institutional standards: among infants requiring respiratory support, treatment was given when FiO₂ > 0.5 and/or OI > 8 [20]. Crucially, as per the study protocol, the attending physicians were strictly prohibited from viewing the LUS images or scores. All decisions regarding surfactant administration were made exclusively based on the pre-defined institutional criteria (FiO₂ > 0.5 and/or OI > 8) and the infant's overall clinical assessment, without any influence from the research LUS data. This rigorous separation ensured that the predictive performance of LUS could be evaluated against an independent, real-world clinical gold standard. All surfactant was delivered via standard endotracheal instillation.

Collected data included demographics, perinatal characteristics, respiratory support modes and parameters, arterial blood gas results, surfactant use (including timing of the first dose), complications (e.g., PPHN, ECMO), and hospital outcomes. PPHN was diagnosed by echocardiographic evidence of right-to-left or bidirectional shunting in conjunction with clinical criteria.

All analyses were conducted in R (version 4.5.1). Two-sided p < 0.05 was considered statistically significant.

Data preprocessing and grouping: Prior to analysis, the raw dataset underwent comprehensive cleaning and preprocessing. Missingness was handled via complete-case analysis using the dplyr and tidyr packages, and outliers were screened and addressed. The authors The authors then defined the following analytic groups: Primary comparison groups—surfactant vs. no-surfactant—used for baseline comparisons and diagnostic performance evaluation. LUS strata—LUS < 6 vs. LUS ≥ 6 based on the ROC-derived optimal cutoff—to examine management differences across score ranges. Prespecified subsets for robustness analyses included: by support type (NIV vs. IMV); by disease severity (LUS 0–5 mild, 6–12 moderate, > 12 severe); by complication status (PPHN absent vs. present); by gestational age (37–39 vs. 40–42 weeks); and by risk tier (low: LUS 〈 6; intermediate: LUS 6–10; high: LUS 〉 10).

Descriptive statistics and between-group comparisons: Normality of continuous variables was assessed using the Shapiro–Wilk test. Normally distributed data are reported as mean ± SD and compared with Student’s t-tests; non-normal data are reported as median (IQR) and compared with Mann–Whitney U tests. Categorical variables are presented as counts and percentages and compared using chi-square tests or Fisher’s exact tests (Table 1).

Diagnostic accuracy and comparisons: ROC analyses were performed for LUS score, OI, FiO₂, and pH to predict surfactant requirement. The authors calculated AUCs with 95 % CIs. Because OI/FiO₂ defines the standard for surfactant decision-making, the authors used bootstrap resampling (1000 iterations) for robust AUC estimation and applied DeLong tests to compare LUS AUCs with traditional indices. The optimal LUS cutoff was determined by the maximum Youden index, with corresponding sensitivity, specificity, PPV, and NPV reported (Table 2; Figure 1).

Figure 1.

ROC curves of LUS and other variables.

Multivariable logistic regression and confounding control: To independently assess the association between LUS score and surfactant use while addressing potential confounding, the authors fit multivariable logistic regression models with surfactant use as the dependent variable. Prior to modeling, collinearity among continuous covariates was evaluated using variance inflation factors (all VIFs < 2.0). The final model included the LUS score (continuous) adjusted for gestational age, birth weight, and PPHN. Results are presented as adjusted odds ratios (ORs) with 95 % CIs (Table 3).

Model performance, calibration, and clinical utility:

Visualization and distribution: To visualize LUS distributions in surfactant vs. no-surfactant groups and depict overall dispersion with key statistics, the authors generated combined box-and-violin plots (Figure 2). To display the continuous relationship between LUS and the probability of surfactant use, the authors plotted restricted cubic spline (RCS) curves (Figure 3).

• — Decision curve analysis (DCA): Net clinical benefit across threshold probabilities was compared among four strategies: (1) treat-all; (2) treat-none; (3) LUS-guided treatment; and (4) treatment guided by pH combined with chest X-ray grade (Table 4, Figure 4).

  Table 4.

  Decision curve analysis - net benefits at key threshold probabilities.

  Threshold Probability	All Treat Net Benefit	None Treat Net Benefit	pH Only Net Benefit	Chest X-ray Only Net Benefit	pH + X-ray Net Benefit	LUS-based Net Benefit	LUS Advantage vs Combined	Best Strategy
  0.1	0.2966	0	0.3313	0.2966	0.3216	0.3660	0.0443	LUS-based
  0.2	0.2087	0	0.2661	0.2087	0.2672	0.3647	0.0975	LUS-based
  0.3	0.0957	0	0.2339	0.0996	0.2326	0.3630	0.1304	LUS-based
  0.4	−0.0550	0	0.2049	0.0015	0.2248	0.3609	0.1361	LUS-based
  
  

  Figure 4.

  Decision curve analysis demonstrating the superiority of the LUS-based strategy: within the clinically relevant threshold range (10–40 %), the LUS-based prediction strategy yielded the highest net benefit, significantly outperforming other strategies, including pH + Chest X-ray, Chest X-ray only, and pH only.

• — Calibration: The authors assessed agreement between predicted and observed probabilities using decile grouping with LOWESS smoothing (Table 5, Figure 5).

  Table 5.

  Calibration statistics for different predictors.

  Predictor	Calibration Error	Brier Score	Observations
  LUS Score	0.0162	0.0081	218
  Chest X-ray Grade	0.4617	0.2309	218
  pH	0.2259	0.1151	218
  Gestational Week	0.4644	0.2322	218

  Note: Lower values indicate better calibration and prediction performance.
  
  

  Figure 5.

  Calibration curves. LUS score demonstrates optimal calibration: the calibration curve for the LUS score (thick blue line) lies closest to the ideal reference line, indicating the best agreement between predicted and observed probabilities across the entire probability range.

Figure 2.

(A) comparison of variables between surfactant groups (Boxplots). (B) Variable distributions are visualized using violin plots, with embedded boxplots displaying the median, interquartile range, and outliers. - The "width" of the violin plot represents data density, while the boxplot intuitively illustrates central tendency and dispersion.

Figure 3.

Probability curve.

- risk stratification: low risk (LUS < 6): very low probability of surfactant therapy (< 10 %).

- Intermediate risk (6 ≤ LUS < 12): Probability increases rapidly, approaching 100 %.

- High risk (LUS ≥ 12): probability stabilizes at 100 %.

Subgroup analyses, interactions, and robustness checks: Prespecified subgroups assessed robustness of LUS diagnostic performance: ventilation mode (IMV vs. NIV), gestational age (≤ 39 vs. > 39 weeks), and PPHN status. Heterogeneity in AUC across subgroups and interaction P-values were evaluated using the Cochran–Mantel–Haenszel framework (Table 6).

All statistical code and datasets will be made publicly available for verification.

A total of 218 infants with confirmed MAS were included and stratified by receipt of surfactant therapy into a surfactant group (n = 80) and a non-PS group (n = 138). Demographic and clinical baseline characteristics are summarized in Table 1. Demographics: There were no significant differences in sex distribution (p = 1.00) or gestational age (median: 39 weeks in the non-surfactant group vs. 40 weeks in the surfactant group, p = 0.704). Clinical indices: The surfactant group had significantly higher LUS scores (median: 8.5 vs. 4, p < 0.001), higher fraction of inspired oxygen (FiO₂; median: 50 % vs. 30 %, p < 0.001), higher oxygenation index (OI; median: 10 vs. 7, p < 0.001), and lower pH (median: 7.08 vs. 7.21, p < 0.001).

Complications and length of stay: Length of stay was longer in the surfactant group (median: 8 vs. 6 days, p < 0.001). Chest X-ray grade distribution (p = 0.413) and the incidence of complications such as PPHN (p = 0.134) did not differ significantly between groups.

Receiver operating characteristic (ROC) analyses were used to evaluate the prediction of PS requirement (Table 2). LUS score: At the optimal cutoff of 6, the AUC was 0.999 (95 % CI, 0.997–1) with 100 % sensitivity, 98.6 % specificity, and 93.1 % accuracy. Its AUC was significantly higher than that of pH (AUC = 0.906, p < 0.001) and chest X-ray grade (AUC = 0.539, p < 0.001).

Comparison with conventional indices: Although OI and FiO₂ guided surfactant decisions in this study, the predictive performance of LUS (AUC = 0.999) numerically exceeded OI (AUC = 0.9673) and FiO₂ (AUC = 0.9527). DeLong tests showed significant AUC differences between LUS and OI (p = 0.021) and between LUS and FiO₂ (p = 0.015).

To control for potential confounding, the authors fitted multivariable logistic regression models (Table 3) to assess the independent association between LUS score and surfactant use. Model performance: Across three handling schemes (clinical grouping, winsorization, and L1 regularization), model AUCs all exceeded 0.98, indicating excellent discrimination. Key finding: After adjustment for gestational age, sex, FiO₂, chest X-ray grade, and pH, the LUS score remained a strong independent predictor of surfactant use (Scheme C: OR = 2.73, 95 % CI: 1.00–3.81). Other covariates (e.g., gestational age, sex, FiO₂) did not show significant independent associations in the final model.

Decision curve analysis (DCA) (Table 4): Within clinically relevant threshold probabilities of 10 %–40 %, an LUS-guided strategy provided higher net benefit than “treat-all,” “treat-none,” “pH only,” or “pH plus chest X-ray” strategies. For example, at a threshold probability of 0.3, the net benefit for the LUS strategy was 0.363, superior to alternatives.

Calibration (Table 5): The LUS-based model showed the lowest calibration error (0.0162) and a Brier score of 0.0081, outperforming chest X-ray grade (calibration error: 0.4617), pH (0.2259), and gestational age (0.4644), indicating excellent agreement between predicted and observed probabilities.

Subgroup analyses assessed robustness across clinical strata (Table 6). Ventilation mode: LUS demonstrated excellent and consistent performance in both the invasive mechanical ventilation group (n = 134, AUC = 0.997) and the noninvasive ventilation group (n = 84, AUC = 1.000). Gestational age strata: Predictive ability was comparable between 37 and 39 weeks (n = 113, AUC = 0.998) and 40–42 weeks (n = 105, AUC = 1.000). Complication status: Among infants without PPHN (n = 216), LUS maintained very high accuracy (AUC = 0.999). Conclusion: The predictive value of LUS for surfactant requirement is robust across patient subgroups, supporting broad clinical applicability.

Score distribution: LUS scores were concentrated at higher values in the surfactant group (median 8.5) and at lower values in the non-surfactant group (median 4), with a significant distributional difference (p < 0.001). Probability prediction: Restricted cubic spline (RCS) curves indicated a continuous dose–response relationship between LUS and the probability of surfactant use, with a sharp increase in probability once the LUS score exceeded 6 (Figure 6).

Figure 6.

Scatter plot.

Strong positive correlation: The LUS score demonstrated a significant strong positive correlation with FiO₂ (r = 0.763, p < 0.001), indicating that LUS effectively reflects the oxygenation requirements of neonates. - Early Identification Efficacy: Among infants who received surfactant therapy, 63.7 % could be identified early (LUS ≥ 6) before their FiO₂ requirement reached 50 %, providing a basis for timely intervention.