Preterm infants face a unique physiological predicament. Born before their organs have completed development, they must maintain blood pressure, thermoregulation, respiration and digestion simultaneously, often while receiving intensive respiratory and circulatory support. The kidneys and the gastrointestinal tract are both highly oxygen-dependent organs, and both are vulnerable to injury when oxygen delivery falls short of demand. Necrotizing enterocolitis, a devastating bowel disease, and acute kidney injury are among the most feared complications of prematurity, and both have been linked in previous research to periods of compromised perfusion and tissue hypoxia.

The research team used near-infrared spectroscopy, a bedside optical technique that shines near-infrared light through tissue and measures how much is absorbed by oxygenated and deoxygenated hemoglobin, to track regional oxygen saturation continuously. Probes placed over the flank captured renal oxygenation, while probes over the abdomen monitored intestinal oxygenation. Because the technique is entirely non-invasive and painless, it can be applied repeatedly or continuously in the smallest and sickest infants without adding to their physiological burden, a major advantage over blood sampling or imaging approaches that offer only snapshots.

What makes the study particularly valuable is its paired design. Rather than examining kidney or gut oxygenation in isolation, the investigators measured both regions simultaneously in the same infants, allowing direct comparison of how the two vascular beds behave under identical systemic conditions. This matters because the splanchnic and renal circulations respond differently to stress. The gut is exquisitely sensitive to feeding, expanding its metabolic and oxygen demands with every milk bolus, while the kidney’s oxygen consumption is dominated by active sodium reabsorption and is influenced by medications, oxygen therapy and hemodynamic swings.

The study evaluated how feeding timing interacted with these oxygenation patterns. Feeding practices in neonatal intensive care units vary widely: some units begin minimal enteral feeds within hours of birth, others delay until the infant is clinically stable, and the pace of advancing feeds is a subject of ongoing debate. The reasoning behind cautious feeding is that a gut with marginal oxygen reserve may be poorly equipped to handle the postprandial surge in demand, potentially increasing the risk of feed intolerance or necrotizing enterocolitis. Yet prolonged delays in feeding carry their own costs, including gut atrophy, prolonged dependence on intravenous nutrition and increased risk of infection and liver injury.

By correlating continuous renal and intestinal oxygen saturation measurements with feeding episodes, the researchers were able to characterize how preterm infants’ regional oxygenation responds around feeds and whether the magnitude or pattern of those responses differs between the kidney and the gut. The paired measurements revealed that intestinal oxygenation behaves dynamically in relation to feeding, whereas renal oxygenation follows a more independent course shaped largely by systemic hemodynamics and clinical interventions. This dissociation underscores that a single global measure of oxygenation, such as arterial oxygen saturation, cannot capture the redistribution of blood flow that occurs between organ beds in the transitional period after preterm birth.

The clinical implications are potentially significant. If intestinal oxygenation measurements can identify infants whose guts are experiencing relative hypoxia before clinical signs such as feeding intolerance, abdominal distension or bloody stools appear, monitoring could shift from reactive to preventive. Neonatologists could, in principle, titrate the initiation and advancement of feeds against a real-time physiological signal, feeding more aggressively when intestinal oxygenation is robust and holding back when it signals distress. Similarly, renal oxygenation trends could serve as an early warning for acute kidney injury, which currently is often detected only after serum creatinine rises, a lagging marker that reflects injury after it has already occurred.

Like all physiological monitoring studies in this population, the work comes with methodological caveats. Near-infrared spectroscopy measures a regional venous-weighted saturation rather than direct tissue oxygen tension, and probe positioning, skin thickness and local edema can influence readings. The kidney’s retroperitoneal location and the interposition of bowel gas make renal signals technically more challenging to acquire than cerebral signals, and interpreting absolute values requires normative data across gestational ages and postnatal days. The authors’ paired comparison design helps address some of these concerns by using each infant as their own physiological control, isolating differences between organ beds from confounding by systemic factors.

Broader context strengthens the relevance of the findings. Research over the past two decades has established that low regional oxygenation in the first days of life is associated with adverse outcomes, including necrotizing enterocolitis, and that postnatal adaptation involves complex redistribution of cardiac output away from some vascular beds. Yet most monitoring studies have focused on the brain, leaving the abdomen comparatively underexplored despite the gut’s central role in neonatal morbidity. By bringing the kidney and the intestine into a single analytical frame and tying both to feeding, a modifiable care practice, this study adds a practical dimension to a growing physiological literature.

For now, near-infrared spectroscopy of the abdomen and kidney remains a research and select-care tool rather than routine standard practice, and no oxygenation threshold has been validated as a trigger for feeding decisions. But the trajectory of the field is clear: as continuous multi-organ monitoring becomes more feasible, the goal of individualized, physiologically guided nutrition in preterm infants moves closer to reality. The study’s paired demonstration that kidney and gut oxygenation diverge in the same infant at the same moment is a reminder that preterm physiology cannot be read from any single monitor, and that the organs we cannot see are often the ones most in need of a window.

Subject of Research: Simultaneous monitoring of renal and intestinal oxygenation in preterm neonates in relation to feeding timing

Article Title: Evaluation of kidney oxygenation compared to intestinal oxygenation and feeding timing in preterm neonates

Article References: Condit, P. E., Lindstrom, R., Gunderson, S. N., & Harer, M. W. (2026). Evaluation of kidney oxygenation compared to intestinal oxygenation and feeding timing in preterm neonates. Journal of Perinatology. https://doi.org/10.1038/s41372-026-02915-3

Image Credits: AI Generated

DOI: 10.1038/s41372-026-02915-3

Keywords: preterm neonates, near-infrared spectroscopy, kidney oxygenation, intestinal oxygenation, feeding timing, necrotizing enterocolitis, acute kidney injury, neonatal intensive care, regional oxygen saturation, enteral feeding, neonatal hemodynamics, Journal of Perinatology

To understand why the authors believe artificial intelligence is needed, it helps to grasp what pulmonary hypertension actually is in a newborn. The condition is characterized by an abnormal elevation in pulmonary arterial pressure, most often driven by an increase in pulmonary vascular resistance, by altered pulmonary blood flow, or by impaired drainage of venous blood from the lungs. The World Health Organization classifies pulmonary hypertension into distinct groups based on these underlying mechanisms, a taxonomy that reflects the remarkable diversity of pathophysiologic pathways that can lead to pulmonary vascular disease. In adults, these categories tend to remain relatively distinct. In neonates, however, the authors emphasize that the mechanisms frequently overlap within a single patient, producing a dynamic and continuously evolving disease state that resists tidy classification.

That overlap is not merely an academic inconvenience. It means that a premature infant with bronchopulmonary dysplasia, a term baby with persistent pulmonary hypertension of the newborn, and an infant with congenital heart disease may all share elevated pulmonary pressures while requiring fundamentally different therapeutic strategies. Whatever the underlying cause, the central hemodynamic consequence is the same: increased afterload on the right ventricle, the chamber of the heart responsible for pumping blood through the lungs. The immature newborn myocardium is poorly equipped to tolerate this burden. Faced with sustained elevation in pulmonary arterial pressure, the right ventricle undergoes maladaptive remodeling, its contractility deteriorates, and eventually it begins to fail.

The consequences of right ventricular failure cascade through the entire fetal-to-neonatal circulatory transition. As the right ventricle weakens, its output falls. The enlarged, pressure-overloaded right ventricle can compress the left ventricle through a phenomenon known as interventricular dependence, flattening the interventricular septum and diminishing the left ventricle’s ability to fill with blood. Reduced left ventricular filling means reduced systemic cardiac output, and compromised perfusion of vital organs follows. Perhaps most treacherously, the authors highlight a diagnostic paradox: in severe cases, right ventricular failure may actually lead to a reduction in measured pulmonary arterial pressure, simply because the failing ventricle can no longer generate enough force to push blood forward. A diagnosis defined solely by pressure-based metrics can therefore miss the sickest patients, precisely the infants in whom the right ventricle has already decompensated.

This paradox exposes a fundamental weakness in how neonatal pulmonary hypertension has traditionally been diagnosed. Echocardiography remains the cornerstone of bedside assessment, and clinicians rely on a battery of derived measures: estimated pulmonary artery pressures from tricuspid regurgitant jets, the left ventricular end-systolic eccentricity index used to gauge septal flattening in infants, right ventricular size and function indices, and patterns of ductal shunting. Each of these metrics captures a single dimension of a profoundly multidimensional physiology. Guidelines from the American Heart Association and the American Thoracic Society, along with consensus statements from the European Pediatric Pulmonary Vascular Disease Network, acknowledge the difficulty, yet interobserver variability among clinicians interpreting the same images remains a persistent obstacle, particularly in the prematurely born infant whose transitional circulation shifts hour by hour.

The stakes of missing early disease are well documented. Research in extremely preterm infants has shown that measures of early cardiac function are associated with subsequent death, severe bronchopulmonary dysplasia, and the development of pulmonary hypertension, suggesting that the seeds of the disease are sown long before it becomes evident on standard screening. Infants with bronchopulmonary dysplasia represent an especially vulnerable population, and specialized evaluation and management pathways have been developed for them precisely because delayed recognition carries such a heavy price. Yet routine surveillance with expert-targeted echocardiography is resource intensive, operator dependent, and impractical to deploy at the frequency that a rapidly evolving neonatal circulation would demand. This is the gap that the authors believe artificial intelligence is positioned to close.

The evidence that machine learning can meet this challenge is accumulating quickly. In adult pulmonary arterial hypertension, investigators have demonstrated that artificial intelligence-based echocardiography can automate image acquisition, interpretation, and phenotyping, reducing the expertise threshold required for meaningful assessment. More striking for neonatology, a recent study published in Pediatric Research reported that a deep learning model could achieve automated detection of pulmonary hypertension directly from echocardiograms, performing image analysis tasks that previously required subspecialty-level training. Deep learning models of this kind are trained on large libraries of annotated ultrasound images, learning to recognize subtle patterns of ventricular geometry, septal motion, and Doppler flow that correlate with elevated pulmonary vascular resistance. Once trained, such models can process new studies in seconds, consistently and without fatigue.

The conceptual framework proposed by Altit and Sant’Anna extends this vision beyond simple binary detection. Because neonatal pulmonary hypertension is heterogeneous and overlapping in mechanism, the authors argue that algorithms should be designed to perform multidimensional phenotyping: simultaneously characterizing right ventricular afterload, ventricular interaction, cardiac output, and the trajectory of these variables over time. A model that tracks how an individual infant’s hemodynamic profile evolves across the first days and weeks of life could, in principle, identify the transition toward maladaptive remodeling far earlier than any single snapshot measurement. Such continuous, quantitative surveillance would convert echocardiography from an episodic diagnostic event into something closer to a longitudinal monitoring tool, analogous to how continuous pulse oximetry transformed the detection of hypoxemia decades ago.

The authors are careful to frame these possibilities as a direction of travel rather than a finished reality. Before artificial intelligence can be deployed safely in neonatal hemodynamics, models must be trained on diverse, well-annotated datasets that span gestational ages, disease etiologies, and the unique circulatory physiology of the transitional newborn, a population in whom adult-derived algorithms are unlikely to perform reliably. Validation against gold-standard references, transparent reporting of model performance across subgroups, and careful attention to how automated outputs integrate into clinical decision making will all be essential. There are also equity considerations: if training data are drawn predominantly from large academic centers, algorithms may underperform in the community hospitals where many at-risk infants are born.

Even so, the trajectory is clear. The convergence of increasingly accessible bedside echocardiography, powerful deep learning architectures, and a growing recognition that pressure-based definitions inadequately capture neonatal cardiovascular physiology has created a genuine inflection point. If the vision outlined in this commentary is realized, the next generation of neonatal intensive care could feature algorithms that silently review every echocardiogram performed in the unit, flagging the earliest signatures of rising pulmonary vascular resistance and prompting targeted intervention while the right ventricle is still compensating rather than failing. For a disease in which timing is everything, artificial intelligence offers what neonatology has long lacked: an objective, tireless, and multidimensional set of eyes on the newborn heart.

The commentary’s publication timing reflects a broader shift in how pediatric cardiovascular research is being conducted. Both authors are based in the neonatology division at Montreal Children’s Hospital, and the work received support from the Fonds de recherche du Québec – Santé, the Canadian Institutes of Health Research, and the American Thoracic Society, among other funders. The article was received in mid-June 2026, accepted within roughly a week, and published in September, a rapid turnaround that suggests the journal viewed the topic as timely for its readership of pediatric researchers and clinicians.

One notable aspect of the argument is its grounding in the pediatric pulmonary hypertension literature that has accumulated over the past decade. The authors draw on updated definitions and diagnostic frameworks developed by international consensus groups, including work published in the European Respiratory Journal and the Journal of Heart and Lung Transplantation, which have progressively refined how the disease is categorized in children. They also cite imaging research linking left ventricular geometry, measured invasively, to pulmonary vascular disease in pediatric patients, underscoring that echocardiographic surrogates have an evidence base extending beyond convenience into validated correlation with hemodynamic data.

The deep learning study highlighted in the commentary, published in Pediatric Research itself, reported automated detection of neonatal pulmonary hypertension from echocardiograms, appearing in a 2026 issue of the journal. Its appearance in a pediatric-specific venue matters, because it signals that the field is moving beyond adapting adult tools toward models built with newborn physiology in mind. The accompanying conceptual framework figure published with the commentary illustrates how detection and phenotyping might be integrated into a single workflow.

For clinicians and researchers following this area, the commentary serves as both a synthesis and a roadmap. It consolidates evidence spanning adult cardiology, pediatric imaging, and neonatal outcome studies into a coherent argument, while explicitly identifying the validation and dataset challenges that remain before algorithmic tools can responsibly enter routine neonatal practice.

Subject of Research: Artificial intelligence for detecting pulmonary hypertension in newborns through neonatal hemodynamic assessment.

Article Title: Artificial intelligence in neonatal hemodynamics: toward earlier and more reliable detection of pulmonary hypertension

Article References: Altit, G., & Sant’Anna, G. (2026). Artificial intelligence in neonatal hemodynamics: toward earlier and more reliable detection of pulmonary hypertension. Pediatric Research. https://doi.org/10.1038/s41390-026-05461-y

Image Credits: AI Generated

DOI: 10.1038/s41390-026-05461-y

Keywords: artificial intelligence, neonatal hemodynamics, pulmonary hypertension, echocardiography, deep learning, right ventricular function, preterm infants, bronchopulmonary dysplasia, neonatology, pediatric research, machine learning, ventricular afterload