1. Introduction
Persistent pulmonary hypertension of the newborn (PPHN) is characterized by failure of the normal postnatal decline in pulmonary vascular resistance, with right‐to‐left shunting and severe hypoxemia [1]. Observational studies in very preterm and term or near‐term populations have documented substantial mortality and practice variation [2, 3]. Multicenter and registry data further show that etiology, management, and response to inhaled nitric oxide (iNO) differ across neonatal populations [4, 5]. iNO selectively vasodilates ventilated regions of the lung and reduces the composite outcome of death or extracorporeal membrane oxygenation (ECMO) in term and near‐term neonates, although mortality alone is not consistently reduced and incomplete response remains common [6–8].
High‐frequency oscillatory ventilation (HFOV) supports gas exchange with small tidal‐volume excursions and can promote lung recruitment [9, 10]. Neonatal studies have evaluated oscillatory support across preterm and near‐term respiratory failure [11, 12]. The physiologic rationale for combining HFOV with iNO is that recruitment may reduce intrapulmonary shunt and increase the ventilated vascular surface exposed to iNO, whereas iNO acts through endothelial nitric oxide signaling to reduce extrapulmonary right‐to‐left shunting [13, 14].
A recent PPHN review supports this complementary lung‐recruitment and pulmonary‐vasodilation framework [15]. Prior trials largely evaluated rescue or crossover use rather than a parallel‐group comparison of initial strategies [16, 17]. Broader pediatric pulmonary hypertension guidance also emphasizes age‐specific phenotyping and etiologic context when interpreting treatment response [18, 19].
We therefore compared first‐line HFOV + iNO with conventional mechanical ventilation (CMV) + iNO and HFOV alone in neonates with severe, echocardiographically confirmed PPHN. The primary objective was to compare death or ECMO within 120 days. Secondary objectives were to characterize oxygenation, pulmonary pressure and right‐ventricular function, circulating biomarkers, duration of support, and acute complications.
Because the trial was open‐label and single‐center, secondary physiologic and mechanistic outcomes were considered supportive and exploratory.
2. Methods
2.1. Study Design and Ethics
This prospective, single‐center, 3‐arm, open‐label randomized clinical trial was conducted at Xuzhou Central Hospital from January 2023 through June 2025. The institutional review board approved the study before enrollment (2023‐XZ‐23184), and written informed consent was obtained from at least 1 parent or legal guardian. The trial was conducted in accordance with the Declaration of Helsinki and Good Clinical Practice [20].
Reporting follows the CONSORT 2025 statement [21]. The trial was not prospectively registered; consequently, prespecification of outcomes and analyses cannot be independently verified against a public registration record, which is a substantive interpretive limitation.
2.2. Participants
Eligible neonates had a gestational age of at least 34 weeks, a postnatal age of 7 days or less, echocardiographic evidence of PPHN (estimated pulmonary artery systolic pressure greater than two‐thirds of systemic systolic pressure or right‐to‐left or bidirectional shunting through the ductus arteriosus or foramen ovale), and an oxygenation index of at least 25 on 2 consecutive arterial blood gas measurements while receiving an FiO2 of at least 0.80 [22, 23]. Exclusion criteria were congenital diaphragmatic hernia, cyanotic congenital heart disease other than patent ductus arteriosus or patent foramen ovale, lethal chromosomal or genetic abnormalities, severe perinatal asphyxia, birth weight below 2000 g, uncontrolled seizures, or an immediate requirement for ECMO.
2.3. Randomization and Masking
Participants were assigned 1:1:1 to CMV + iNO, HFOV + iNO, or HFOV alone using a computer‐generated sequence prepared before recruitment. The assignment was released only after eligibility and written consent had been confirmed, so the enrolling clinical team did not have access to future assignments. The archived analytic records did not preserve the identities of the individual who generated the sequence or the individual who released assignments.
The ventilation strategy and iNO administration could not be masked. Echocardiographic measurements and laboratory assays were interpreted without access to group labels whenever feasible. Because treating clinicians were aware of assignment, ventilator adjustment, rescue decisions, ECMO initiation, and liberation from support remained susceptible to performance bias.
The study was therefore open‐label and did not fully implement a formal prospective randomized open‐blinded end‐point design with systematically blinded outcome adjudication [24].
2.4. Interventions
HFOV was delivered with a SensorMedics 3100A or Stephanie ventilator. Initial frequency was 8–10 Hz, mean airway pressure was generally set 2–4 cm H2O above the preceding CMV mean airway pressure, and amplitude was adjusted to visible chest‐wall oscillation [9]. Recruitment was individualized using oxygenation, systemic perfusion, chest expansion, radiography, and serial echocardiography.
Mean airway pressure was reduced when hyperinflation, hypotension, impaired venous return, or worsening right‐ventricular loading was suspected. This strategy reflects the U‐shaped relation between lung volume and pulmonary vascular resistance: Recruitment toward functional residual capacity may improve ventilation‐perfusion matching, whereas overdistension may compress alveolar capillaries and increase right‐ventricular afterload [25, 26].
iNO was started at 20 ppm. Response was assessed after 30 min; a brief increase to 40 ppm was permitted for nonresponse, followed by stepwise weaning after improvement in oxygen requirement. Methemoglobin was monitored, consistent with neonatal iNO practice guidance [27].
CMV was delivered in pressure‐controlled or volume‐targeted modes with lung‐protective targets [28]. Surfactant could be administered for documented or strongly suspected surfactant deficiency; combined surfactant and iNO has previously been evaluated in neonatal PPHN [29]. Sildenafil and milrinone were clinician‐directed options for refractory pulmonary hypertension [30–32], whereas vasoactive support followed the infant’s hemodynamic condition [33, 34].
Participant‐level timing and exposure to surfactant, sildenafil, milrinone, and vasoactive support were not systematically captured in the locked analysis dataset; therefore, their distribution across groups and potential contribution to the observed outcomes could not be evaluated.
2.5. Rescue Treatment and ECMO
Rescue crossover to HFOV + iNO was permitted for persistent severe hypoxemia despite optimization of the assigned strategy. Participants who crossed over were excluded from the principal per‐protocol analysis; their possible outcomes were represented in randomized‐denominator sensitivity analyses according to their original assignment. The locked efficacy dataset did not preserve the timing of crossover or an ascertainable 120‐day primary outcome for either of the 2 crossover participants; therefore, separate as‐treated or time‐dependent crossover analyses were not possible.
ECMO was considered for potentially reversible cardiorespiratory failure with persistent severe hypoxemia despite maximal assigned and rescue therapy, generally an oxygenation index of approximately 40 or greater, after review of major contraindications, consistent with the center’s adaptation of neonatal ECMO guidance [35]. ECMO candidacy was not adjudicated by an independent masked committee; it was determined jointly by the treating neonatology and ECMO teams according to institutional practice. Because ECMO delivery is influenced by clinician judgment, expertise, and local resources, the ECMO component was not interpreted as a purely physiologic endpoint [36].
2.6. Outcomes
The primary outcome was death from any cause or initiation of ECMO within 120 days after randomization. Secondary outcomes were serial oxygenation index and PaO2/FiO2 ratio; estimated pulmonary artery systolic pressure, tricuspid annular plane systolic excursion, right‐ventricular to left‐ventricular basal diameter ratio, and Tei index; B‐type natriuretic peptide, endothelin‐1, cyclic guanosine monophosphate, and interleukin‐6; individual components of the primary outcome; and durations of mechanical ventilation, intensive care, hospitalization, and supplemental oxygen. Acute safety outcomes were grade III–IV intraventricular hemorrhage, pneumothorax, Bell stage II or greater necrotizing enterocolitis, stage III or greater retinopathy of prematurity, and hearing impairment.
Bronchopulmonary dysplasia was omitted because no single definition was appropriate across this predominantly term and late‐preterm cohort, whereas much of the iNO evidence focused on bronchopulmonary dysplasia prevention concerns extremely preterm infants [37].
2.7. Statistical Analysis
The sample‐size calculation required 42 analyzable infants per group, and the recruitment target was increased to 46 per group to allow for approximately 10% attrition. A total of 138 infants were randomized, 46 to each group. Reaudit of the locked participant‐level efficacy dataset identified 126 infants (42 per group) with an ascertainable 120‐day primary outcome who completed the assigned strategy without a major protocol deviation.
The principal efficacy analysis is, therefore, described as a complete‐case analysis of the per‐protocol population rather than a conventional intention‐to‐treat analysis. The 12 randomized infants excluded from this analysis had no ascertainable 120‐day primary‐outcome record in the locked dataset after major protocol deviation, parental withdrawal, equipment failure, or rescue crossover. No participant‐level outcomes were created or singly imputed.
Model‐based multiple imputation was not used because the missingness mechanism could not be established and a missing‐at‐random assumption could not be justified. To quantify the possible influence of the unavailable outcomes, post hoc randomized‐denominator extreme‐case sensitivity analyses retained 46 infants in each originally assigned group under 4 assumptions: all unavailable outcomes were nonevents; all were events; all 4 unavailable HFOV + iNO outcomes were events, while all 8 unavailable comparator outcomes were nonevents; and the converse differential best‐case assumption. These analyses assess robustness and do not constitute a fully observed intention‐to‐treat estimate.
The overall 3‐group primary comparison used the Pearson chi‐square test. The 2 planned comparisons of HFOV + iNO with each comparator used 2‐sided Fisher exact tests with Holm adjustment and were summarized with risk ratios and absolute risk differences with 95% confidence intervals. Secondary categorical outcomes were compared using Pearson chi‐square tests or Fisher–Freeman–Halton exact tests when expected cell counts were small.
Continuous variables are reported as mean (SD) or median (IQR), according to distribution. Repeated measurements were analyzed with linear mixed‐effects models containing group, time, and group‐by‐time interaction as fixed effects and participant as a random intercept; markedly skewed outcomes were log transformed. Duration outcomes used the Kruskal–Wallis test.
A post hoc etiologic analysis summarized events within meconium aspiration syndrome, respiratory distress syndrome, idiopathic PPHN, and sepsis/pneumonia strata; homogeneity across strata was assessed separately for each HFOV + iNO comparison using the Breslow–Day test with a continuity correction. Because of sparse cells and low event counts, this analysis was considered exploratory. Secondary‐outcome p values were not adjusted across the complete family and are exploratory.
Analyses were independently rerun from the participant‐level dataset using Python 3.13, SciPy 1.17, and statsmodels 0.14.6. All tests were 2‐sided with alpha = 0.050, and statistical reporting follows the SAMPL recommendations [38].
3. Results
3.1. Participant Flow and Baseline Characteristics
Between January 2023 and June 2025, 203 neonates were assessed for eligibility, and 65 were excluded before randomization: 41 did not meet eligibility criteria, 18 parents declined consent, and 6 were excluded for other reasons. A total of 138 infants were randomized, with 46 assigned to each group. The complete‐case per‐protocol efficacy population comprised 126 infants, 42 per group.
Twelve randomized infants were not included in the principal analysis: 5 because of major protocol deviations, 4 after parental withdrawal, 1 because of equipment failure, and 2 after rescue crossover. For the 2 rescue crossovers, neither crossover timing nor the 120‐day primary outcome was recoverable. The contemporaneous flow record did not preserve whether parental withdrawal referred only to study treatment or also to continued data collection; however, no ascertainable 120‐day primary outcome for these infants was available in the locked efficacy dataset.
All 138 randomized infants remained represented in randomized‐denominator extreme‐case sensitivity analyses according to their originally assigned group (Figure 1). Baseline characteristics of the per‐protocol population are summarized descriptively in Table 1.
3.2. Oxygenation
Oxygenation improved in all groups, with the largest and most rapid reduction in oxygenation index in the HFOV + iNO group (group‐by‐time interaction p < 0.001) (Table 2; Figure 2). At 72 h, the mean oxygenation index was 7.1 (SD, 2.8) with HFOV + iNO, 11.0 (SD, 3.8) with CMV + iNO, and 15.9 (SD, 4.5) with HFOV alone. The corresponding mean differences for HFOV + iNO were −3.9 (95% CI, −5.3 to −2.4) versus CMV + iNO and −8.7 (95% CI, −10.4 to −7.1) versus HFOV alone.
The PaO2/FiO2 ratio was right‐skewed and was summarized as median (IQR); the log‐scale group‐by‐time interaction was p < 0.001 (Table 2; Figure 3).
3.3. Echocardiographic and Biomarker Outcomes
Estimated pulmonary pressure and right‐ventricular indices improved most in the HFOV + iNO group (Table 3). At 72 h, mean PASP was 25.9 (SD, 5.2) mm Hg with HFOV + iNO, 34.1 (SD, 7.7) mm Hg with CMV + iNO, and 39.2 (SD, 8.8) mm Hg with HFOV alone (Figure 4). Group‐by‐time interactions were p < 0.001 for PASP, TAPSE, RV/LV ratio, and Tei index.
These secondary measures remain susceptible to measurement and performance bias in an open‐label trial.
Biomarker trajectories also favored HFOV + iNO (Table 4). At 72 h, mean cyclic guanosine monophosphate was 114.7 (SD, 39.0) pmol/mL with HFOV + iNO, 88.8 (SD, 24.2) pmol/mL with CMV + iNO, and 62.0 (SD, 16.4) pmol/mL with HFOV alone. B‐type natriuretic peptide and interleukin‐6 were right‐skewed and are summarized as median (IQR).
3.4. Primary and Clinical Outcomes
In the complete‐case per‐protocol population, death or ECMO occurred in 3 of 42 infants (7.1%) with HFOV + iNO, 11 of 42 (26.2%) with CMV + iNO, and 14 of 42 (33.3%) with HFOV alone (overall p = 0.012) (Table 5). For HFOV + iNO versus CMV + iNO, the risk ratio was 0.27 (95% CI, 0.08–0.91), the risk difference was −19.0 percentage points (95% CI, −34.5 to −3.6), and the Holm‐adjusted p value was 0.038. For HFOV + iNO versus HFOV alone, the risk ratio was 0.21 (95% CI, 0.07–0.69), the risk difference was −26.2 percentage points (95% CI, −42.4 to −9.9), and the Holm‐adjusted p value was 0.011 (Table 6).
The corresponding numbers needed to treat were 5.2 and 3.8, but estimates were imprecise because only 28 composite events occurred. In randomized‐denominator extreme‐case sensitivity analyses, the overall comparison remained significant when all 12 unavailable outcomes were assigned as nonevents (11/46, 3/46, and 14/46; p = 0.013) and when all were assigned as events (15/46, 7/46, and 18/46; p = 0.033). Under the differential worst‐case assumption for HFOV + iNO, the corresponding counts were 11/46, 7/46, and 14/46, and the overall comparison was no longer significant (p = 0.222).
The converse differential best‐case scenario remained significant (p < 0.001) (Table 7).
3.5. Etiology Subgroup Analysis
In the post hoc analysis, event counts by PPHN etiology were directionally favorable to HFOV + iNO in most strata but were sparse (Supporting Table S1). Breslow–Day tests did not detect heterogeneity of the association across etiologic strata for HFOV + iNO versus CMV + iNO (p = 0.521) or HFOV alone (p = 0.213). These tests had limited power, and the subgroup results should not be interpreted as evidence of a uniform treatment effect.
4. Discussion
In the observed complete‐case per‐protocol population, HFOV + iNO was associated with a lower incidence of death or ECMO than CMV + iNO or HFOV alone. The absolute risk differences were 19.0 percentage points versus CMV + iNO and 26.2 percentage points versus HFOV alone. The lower incidence of death or ECMO was accompanied by faster improvement in oxygenation, estimated pulmonary pressure, and right‐ventricular indices.
However, this was not a fully observed intention‐to‐treat estimate. The overall comparison remained statistically significant under symmetric extreme‐case assumptions for the 12 unavailable outcomes but was no longer significant under the most unfavorable differential assumption for HFOV + iNO. Accordingly, the apparent benefit is not robust to a differential missing‐outcome assumption that cannot be excluded, and this study does not establish efficacy.
The observed physiologic changes are biologically plausible and are broadly consistent with current understanding of pulmonary vascular physiology. iNO reaches pulmonary vessels adjacent to ventilated alveoli; recruitment of atelectatic lung may therefore improve ventilation‐perfusion matching and expose a larger vascular bed to selective vasodilation through endothelial nitric oxide signaling [13, 14]. The greater increase in cyclic guanosine monophosphate and decrease in endothelin‐1 with HFOV + iNO are compatible with enhanced nitric oxide‐soluble guanylate cyclase signaling, but circulating biomarkers are downstream correlates rather than direct measures of pulmonary iNO delivery [39, 40].
Recruitment must be balanced against right‐ventricular and systemic hemodynamics. Pulmonary vascular resistance is lowest near an appropriate functional residual capacity and may increase at both low and excessive lung volumes. Excessive mean airway pressure may overdistend alveoli, compress pulmonary capillaries, reduce venous return, and increase effective right‐ventricular afterload [41, 42].
The observed pulmonary artery systolic pressure, tricuspid annular plane systolic excursion, right‐ventricular to left‐ventricular ratio, and Tei‐index trajectories suggest that the observed echocardiographic trajectories are consistent with improved cardiopulmonary adaptation, but they do not exclude transient adverse hemodynamic effects. Individualized pressure titration using oxygenation, perfusion, lung expansion, and functional echocardiography is therefore preferable to routine pressure escalation.
Systems‐level studies in adult pulmonary hypertension provide methodological context for future neonatal research but should be interpreted only as mechanistic context for treatment of PPHN. Adult pulmonary arterial hypertension reviews and text‐mining studies illustrate mechanism‐informed drug‐repurposing frameworks [43, 44]. Trans‐right‐ventricular metabolomics provides an example of integrated right‐ventricular and pulmonary vascular phenotyping [45].
Mendelian randomization analysis in another cardiovascular domain illustrates a strategy for causal hypothesis generation rather than direct evidence of a microbiome mechanism in PH [46]. These studies neither justify neonatal use of sodium–glucose cotransporter 2 inhibitors or other repurposed agents nor establish a neonatal microbiome treatment target; their relevance is limited to generating age‐specific hypotheses for future prospective cohorts.
Acute complication rates did not differ detectably, but nonsignificant comparisons in this sample do not establish safety equivalence. The shorter duration of mechanical ventilation may be clinically meaningful, although ventilator weaning and extubation were clinician‐directed and could have been influenced by knowledge of treatment assignment. Duration and biomarker outcomes should be interpreted as exploratory.
This study has limitations. First, the principal efficacy analysis was a complete‐case analysis of the per‐protocol population rather than a conventional intention‐to‐treat analysis. Twelve of 138 randomized infants did not have an ascertainable 120‐day primary outcome in the locked efficacy dataset.
Although randomized‐denominator sensitivity analyses supported the overall result under symmetric assumptions, statistical significance was lost under the most unfavorable differential assumption for HFOV + iNO; informative postrandomization missingness and selection bias therefore remain possible, and the magnitude of the observed effect may be overestimated. Thus, both the direction and magnitude of the apparent treatment effect remain uncertain, and the complete‐case result cannot support an efficacy conclusion. Second, the open‐label design permits performance and ascertainment bias, particularly for ventilator adjustment, rescue decisions, ECMO initiation, echocardiographic measurements, and liberation from support.
Third, ECMO use depends partly on institutional expertise, resources, and clinician judgment and was not adjudicated by an independent masked committee. Fourth, the exact timing and observed outcomes of the 2 recorded rescue crossovers were not recoverable from the locked efficacy dataset. Accordingly, neither an as‐treated effect nor the direction of crossover‐related bias can be determined.
Fifth, the single‐center sample was small for mortality, uncommon adverse events, and treatment‐by‐etiology interactions; the post hoc etiologic analysis was underpowered and included zero‐event cells. Sixth, clinician‐directed cointerventions such as surfactant, vasoactive drugs, sildenafil, and milrinone were not systematically captured in the locked analysis dataset, so balance across groups could not be assessed. Differential cointervention use therefore cannot be excluded and may have contributed to between‐group differences, further weakening attribution of the observed outcomes to the assigned strategy.
Seventh, the documentation of allocation concealment was incomplete in the archived study records, and the archived records did not distinguish whether parental withdrawal applied only to assigned treatment or also to outcome follow‐up; these details were not reconstructed. Eighth, multiple secondary outcomes were examined without family‐wise multiplicity adjustment. Ninth, long‐term neurodevelopmental follow‐up was not available, although neurosensory outcomes are clinically relevant after major neonatal cardiopulmonary interventions [47].
Finally, the trial was not prospectively registered. This prevents independent verification that the reported outcomes and analyses were prespecified and increases concern about selective outcome or analysis reporting.
In conclusion, HFOV + iNO was associated with fewer deaths or ECMO events and faster improvement in oxygenation and right‐ventricular echocardiographic measures in the observed per‐protocol population. However, exclusion of 12 randomized infants, loss of statistical significance under the differential worst‐case missing‐outcome assumption, unrecoverable crossover data, unmeasured clinician‐directed cointerventions, and absence of prospective registration substantially limit causal confidence. These findings are exploratory and hypothesis‐generating; they do not establish efficacy, safety equivalence, a preferred first‐line strategy, or a basis for guideline change.
A prospectively registered multicenter trial with complete outcome ascertainment, documented allocation concealment, standardized rescue and ECMO criteria and cointerventions, prespecified heterogeneity analyses, and long‐term neurodevelopmental follow‐up is required.
Author Contributions
Bin Zhou, the corresponding author and manuscript guarantor, had full access to all of the data in this study and takes complete responsibility for the integrity of the data and the accuracy of the data analysis.
Funding
This study was supported by the Xuzhou Science and Technology Program Project (KC23184).
Disclosure
The guarantor affirms that this manuscript is an honest, accurate, and transparent account of the study; that no important aspects have been omitted; and that discrepancies from the study as planned and reported have been explained. The funder had no role in study design, recruitment, data collection, analysis, interpretation, manuscript preparation, or the decision to submit. All authors have read and approved the final version of the manuscript.
Conflicts of Interest
The authors declare no conflicts of interest.
Supporting Information
Additional supporting information can be found online in the Supporting Information section.
Supporting information