The Boston Children’s Experience: Hidden ICU Risk and AI-Driven De-escalation

Boston Children’s Hospital physicians say one of the greatest risks in pediatric ICUs is delayed de-escalation of care, which can expose children to prolonged ventilation, sedation, and vasoactive medications. They argue that AI-driven clinical intelligence can help clinicians identify safer timing for weaning and extubation, supporting recovery and improving long-term outcomes.

For much of the conversation around artificial intelligence (AI) in critical care, the focus has been on early warning: identifying deterioration sooner, escalating care faster and intervening before a crisis unfolds. These capabilities matter. But after more than a decade of using AI‑driven clinical intelligence at the bedside in pediatric intensive care, Drs. Peter Laussen, MBBS, FANZCA, FCICM, Executive Vice President of Health Affairs and Josh Salvin, MD, MPH, Chief, Division of Cardiovascular Critical Care; Senior Associate Cardiologist, Department of Cardiology at Boston Children’s Hospital have found that the greater, and often overlooked, risk to patients lies at the other end of the spectrum.

It is not only the events that are not caught that can harm children, it’s also the time they’re kept in the ICU longer than necessary.

Every additional day in intensive care carries cumulative downstream risk, particularly for pediatric and neonatal patients. Prolonged exposure to sedation, central lines, mechanical ventilation and vasoactive medications affects neurodevelopment, increases the risk of infection, delays recovery and can extend the physical and emotional toll on children and families. In many cases, the danger is not a missed escalation, but a delayed decision to de‑escalate.

Recently, Drs. Laussen and Salvin, shared their experience using AI-driven clinical intelligence decision-making tools, including their perspective on why AI must focus on de-escalation, not just early warning.

Why ICU duration matters more than we realize

The physiology of critically ill children is dynamic, nonlinear and highly individualized. Yet historically, clinicians’ ability to interpret that complexity has been limited. Much of the data generated at the bedside, such as high‑frequency vital signs, ventilator parameters or pressure measurements, appears momentarily on a monitor and then disappears. Clinicians are left to reconstruct hours or days of physiology from intermittent documentation and snapshots in the medical record.

This environment encourages caution. When uncertainty is high, teams may keep a child intubated “one more night,” maintain vasoactive infusions a bit longer, or defer extubation until conditions appear unquestionably safe. While well‑intentioned, these incremental delays compound risk. For infants and children, especially those recovering from cardiac surgery or critical illness, days spent sedated and mechanically ventilated are not benign.

“We have long known pharmacologically how vasoactive drugs, sedation and ventilatory support work, said Dr. Laussen. “What we have lacked is a reliable, longitudinal view of how an individual patient is responding in real time and how physiology is evolving minute‑to‑minute, not just shift‑to‑shift.”

Moving beyond alerts to trajectories

Early generations of clinical decision support focused largely on thresholds and alarms, such as isolated signals designed to flag immediate danger. Today, continuous clinical intelligence is enabling a more nuanced view of recovery, one that emphasizes trajectories over time rather than momentary alerts, and supports earlier, more coordinated intervention across individual patients and entire units.

“The initial promise of implementing a clinical intelligence platform in our hospital was simple: establish a trustworthy ‘source of truth’ by capturing continuous, high-fidelity physiologic data so we could replay events precisely—down to a five-second window—and infer the correct causal sequence,” said Dr. Salvin.

Building from that foundation, analytics then scale beyond single-patient forensics to population-level awareness, empowering a charge nurse to identify ‘hot spots’ on an overview screen and proactively allocate staff and resources to patients at risk of an unfavorable trajectory.

“We’ve seen firsthand how AI‑driven clinical intelligence supports clinical judgment, rather than replacing it,” said Dr. Salvin. “Instead of acting as an ‘if‑then’ directive or an automated decision‑maker, an AI-driven platform can aggregate continuous, high‑fidelity physiologic data into interpretable trajectories. It allows clinicians to see not just isolated values, but patterns over time: how oxygen delivery, ventilation, perfusion and metabolic demand interact as a child recovers.”

In the Boston Children’s ICUs, this longitudinal visibility has fundamentally altered how the team thinks about de‑escalation of care. Bedside clinicians use continuous risk indicators as secondary measures to cross‑check readiness for extubation and for weaning vasoactive medications. Charge nurses and team leaders monitor unit‑level views to identify patients whose risk profiles are improving or trending in the wrong direction before deterioration becomes obvious.

Critically, these insights do not dictate action. They prompt questions. They elevate discussion. They help teams align around a shared, data‑driven understanding of where a patient is along their recovery trajectory.

Supporting safer weaning and extubation

For critically ill pediatric patients, the transition off mechanical ventilation and vasoactive support represents one of the most fragile phases of recovery. Decisions around when to wean or extubate balance the risks of acting too soon against the harms of unnecessary delay, and even small misjudgments can carry significant clinical consequences. Advances in continuous physiologic monitoring and predictive analytics are helping teams move beyond episodic assessments, enabling more informed, data-driven decisions that support safer, more confident weaning and extubation.

“One of the most tangible benefits we have seen from using an AI-driven platform is in extubation readiness,” said Dr. Laussen. “Failed extubation is not a trivial complication; it exposes children to reintubation, additional sedation and prolonged ventilation, often undoing days of progress. Traditional assessments rely on spot checks such as a respiratory rate at a given ventilator setting, blood gas results, or brief trials of spontaneous breathing.”

By contrast, predictive analytics derived from continuous physiologic data adds a new layer of insight. In multi-center research led by Boston Children’s Hospital, elevated risk indices (IDO2 and IVCO2) in the hours leading up to extubation were associated with nearly double the odds of extubation failure following congenital cardiac surgery. When these analytics are incorporated alongside standard clinical assessment, teams can better time extubation without rushing or unnecessarily delaying the decision.

The same principle applies to vasoactive support. Historically, calculations such as coronary perfusion pressure were performed intermittently, if at all, due to their manual complexity. Continuous computation and trending of these parameters now allow clinicians to more precisely assess myocardial oxygen delivery and safely wean vasoactive infusions sooner. The result is not aggressive care, but more precise care.

Cultural change at the bedside

Perhaps the most underappreciated impact of AI‑driven clinical intelligence is cultural. Where older workflows required deliberate, intermittent calculations, the current generation of clinicians are seeing these continuous trends by default, elevating the conversation and broadening therapeutic options at the point of care.

“When all members of the care team, from bedside nurses to attending physicians, are looking at the same integrated data representation, conversations change. Rounds become more physiologically grounded. Teaching becomes more concrete,” said Dr. Laussen. “Assumptions are challenged earlier, and decisions are made with greater confidence. Over time, this shared situational awareness reduces variation, shortens ICU length of stay and supports earlier transitions out of intensive care.”

The power of de-escalation

As healthcare systems shift toward value‑based care, the most meaningful AI outcomes will not be measured solely by alerts fired or crises averted. They will be measured by whether the right level of care is provided at the right time and de-escalate when it is safe to do so.

For pediatric patients, that distinction matters profoundly. The goal is not simply survival, but recovery: cognitive development, physical resilience and the ability to return to life outside the ICU with as little residual harm as possible.

AI‑driven clinical intelligence, when thoughtfully implemented, helps clinicians answer one of the hardest questions in critical care: Is this child ready for less? The ability to answer that question with greater clarity may be the most important contribution AI can make to the future of pediatric intensive care.

Contact Us