Tracheostomy Risk Can Be Forecast in Severe Bronchopulmonary Dysplasia
In infants with severe bronchopulmonary dysplasia, the path toward tracheostomy is often decided through a mix of physiology, clinical trajectory, and bedside uncertainty. A retrospective cohort study from the BPD Collaborative registry shows that the trajectory can be modeled with useful precision at two clinically important time points: 36 and 44 weeks post-menstrual age.
Investigators developed harmonized prediction models using birth data, respiratory support, and medications to estimate the risk of tracheostomy or death before hospital discharge. The training set included 914 infants, and the validation set included 1,375. Within the training cohort, 854 infants were assessed at 36 weeks PMA and 319 at 44 weeks PMA. Model performance was strong at both time points, with an AUC of 0.89 at 36 weeks and 0.89 at 44 weeks in training, then 0.86 and 0.85 respectively in validation, alongside good calibration.
The practical value is immediate. Families and care teams are often forced to make long-range decisions while the infant’s respiratory course is still unfolding. A model that combines early life variables with the intensity of ventilatory support can turn that uncertainty into a more structured estimate of risk, which is especially relevant when counseling around prolonged ventilation and the possibility of a surgical airway.
Why Respiratory Support at 36 and 44 Weeks PMA Matters
The timing of the model is not arbitrary. By 36 weeks PMA, many preterm infants with BPD have declared part of their respiratory phenotype: some are improving, others remain dependent on invasive support, and a subset is drifting toward chronic ventilator dependence. By 44 weeks PMA, the signal may be even clearer, because persistent need for support reflects both lung injury and the limited reserve left in a developing respiratory system.
Severe BPD is not just a diagnosis of prematurity; it is a marker of enduring lung inflammation, disrupted alveolar development, and pulmonary vascular stress. In that setting, the need for escalating respiratory assistance is a direct bedside expression of physiologic fragility. Medications and ventilator support are therefore not merely treatment inputs in the model; they are proxies for disease burden, reserve, and the likelihood that an infant will remain dependent on intensive airway management.
The models incorporated birth data, medications, and ventilator support at both PMA landmarks. In clinical terms, that matters because no single variable captures the full burden of disease. Gestational history sets the stage, but current respiratory support tells the more urgent story. An infant who remains on substantial ventilatory assistance at 36 or 44 weeks PMA is already signaling a higher probability of a prolonged course.
Validation Strength Gives the Model Clinical Weight
Prediction tools in neonatal medicine often fail when moved beyond the cohort in which they were built. Here, performance stayed strong in validation, which is the main reason this work moves beyond a statistical exercise. The validated AUC values of 0.86 at 36 weeks and 0.85 at 44 weeks suggest that the model retained discrimination in a separate population, not just in the infants used to derive it.
Good calibration adds another layer of credibility. A model can rank risk well without estimating absolute risk accurately; calibration addresses whether predicted probabilities align with observed outcomes. For families and clinicians deciding whether to continue a path of intensive support or prepare for tracheostomy discussions, absolute risk estimates are the clinically meaningful output.
One sentence captures the core finding: the risk of tracheostomy/death was predictable from a combination of early clinical data and ongoing respiratory needs.
Bedside Consequences for Counseling and Planning
For neonatologists, pulmonologists, and surgical teams, this kind of model could reshape how conversations begin. Instead of waiting for a crisis point, clinicians can identify infants whose ongoing support needs place them in a high-risk stratum well before discharge decisions become urgent. That can improve the timing of multidisciplinary discussions involving respiratory management, feeding, growth, airway planning, and long-term family support.
The potential value is not limited to prediction alone. A more precise estimate of risk may help teams standardize referral timing, reduce center-to-center variability, and support shared decision-making in a domain where practice often varies widely. Severe BPD care already requires coordination across neonatology, pulmonology, surgery, otolaryngology, and developmental follow-up; a validated model can serve as a common frame of reference when those specialists weigh the same patient.
There is also a bedside implication for resource use. Tracheostomy is a major inflection point in the care of infants with chronic lung disease. It changes the care environment, the intensity of follow-up, and the family’s responsibilities. The ability to estimate risk earlier may help prepare the infrastructure needed for infants who are unlikely to wean from advanced support quickly.
What the Registry Data Can and Cannot Say
The study is retrospective, which means the models reflect associations in observed care rather than a tested intervention that changed outcomes. The registry design strengthens real-world relevance, but it cannot prove that using the model itself would improve survival, reduce tracheostomy rates, or alter long-term neurodevelopment. It also does not tell clinicians which management strategy should follow a high-risk prediction.
Still, the internal logic is persuasive. Infants with Grade 2/3 BPD who continue to require significant ventilatory support at 36 or 44 weeks PMA are clinically different from those who are already progressing toward respiratory stability. The model formalizes that difference and translates it into a probability estimate that can be updated at two points in the hospitalization.
A de-identified dataset will be available on request, which may allow future work to refine the model, test it in other cohorts, and explore whether adding additional biomarkers or physiologic measures improves prediction. For now, the main finding is straightforward: in severe BPD, the likelihood of tracheostomy or death prior to discharge can be estimated with substantial accuracy using information already available during routine care.