The Ottawa Hospital is the first hospital in the world to evaluate the Extubation Advisor, an AI and machine learning-enabled prediction tool that helps ICU teams decide when patients are ready to breathe on their own. The device continuously monitors and analyzes a patient’s pattern of variation of their respiratory rate during trials of minimal support from the ventilator. It then uses AI to translate that altered breathing pattern into a prediction to help determine if the patient could be safely removed from the ventilator.
Turning ICU monitoring data into decision support for ventilation care
In the ICU, one decision comes up every day: is this patient ready to come off the breathing machine? If the team gets the timing right, it is a major step toward recovery. If the timing isn’t right and the patient has to go back on the ventilator, it can be a harmful adverse setback that can increase the risk of serious complications and can contribute to mortality. It also has a cost for hospitals. Patients who need to go back on the ventilator after being taken off stay an average of nine more days in hospital, at an additional cost of $50,000. Extubation failure is common, with about one in seven ICU patients experiencing it.
Extubation Advisor was created out of Dr. Andrew Seely’s long-standing focus on vital sign variability analysis and predictive monitoring in critical care. Dr. Seely is a critical care physician, thoracic surgeon and scientist based at The Ottawa Hospital and the Ottawa Hospital Research Institute (OHRI) and a professor at the University of Ottawa. He is also the Founder and Chief Science Officer of Therapeutic Monitoring Systems, and Extubation Advisor is one of its core products.
The work behind Extubation Advisor started with research, not software. The question was whether assessments of breathing variation contain early warning indicators of extubation failure. Dr. Seely and his colleagues found that loss of variation of breathing and heart rate were associated with higher risk.
From there, the team moved into building and testing prediction models. A major milestone was a large study involving 10 centres and more than 700 patients, supported through funding from the Canadian Institutes of Health Research (CIHR). That work was used to derive and validate the predictive models behind the Extubation Advisor tool. This research found that reduced respiratory rate variability gave the best prediction of extubation failure.
Next came the translation step: moving from models and analyses to a decision‑support product that could be used consistently at the bedside. Extubation Advisor was created and implemented in observational clinical studies and has now progressed to a randomized trial evaluation. Having completed a 100-patient, 10 centre, CIHR-funded multi-centre feasibility randomized clinical trial (RCT), Dr. Seely and his co-principal investigator, Dr. Karen Burns, have just submitted a grant for a large 700-patient definitive RCT to answer the bigger question: does extubation decision support improve outcomes in routine care at scale?
The tool has also moved into regulated and real-world evaluation. Health Canada approved Extubation Advisor as a Class III medical device, meaning it is regulated software as a medical device intended for clinical use. With the CE mark, the tool is also approved for clinical use in the EU. Meanwhile, at The Ottawa Hospital, Extubation Advisor is being implemented in the ICU for a one‑year term that began in October.
Dr. Seely highlights collaboration and teamwork as key factors in the project’s progress to date. He credits his team at the OHRI: PhD biomedical engineers Christopher Herry and Nathan Scales, research coordinator lead Jill Allen and respiratory therapist Emma Lee. He also describes broad collaboration across sites, with many clinicians, research staff and respiratory therapists contributing, and he points to clinical trial leadership beyond Ottawa, including Dr. Karen Burns at Unity Health.
Another key theme is partnership, including with the private sector, without losing sight of the public patient-centered purpose. Dr. Seely founded his company to help bring Extubation Advisor to the bedside and has described working with industry partners as part of what it takes to commercialize and move tools into real clinical use.
“Everyone is trying to do the same thing: improve clinical quality and efficiency of care. In that context, private sector partnership is not a distraction from patient care.”
Dr. Andrew Seely
— Physician-scientist, TOH & OHRI, uOttawa professor, Therapeutic Monitoring Systems founder
“Everyone is trying to do the same thing: improve clinical quality and efficiency of care,” says Dr. Seely. “In that context, private sector partnership is not a distraction from patient care. It can be an important part of how tools are built, supported and sustained over time.”
Although there has been significant progress, it has not been without some challenges. Dr. Seely highlights three barriers that can shape whether medical AI tools succeed.
The first is data readiness and availability. Predictive modelling requires linking two kinds of data: the inputs used to generate predictions and the outcomes the model is trying to predict. It is not enough to have monitoring information, which is challenging to collect. Teams also need consistent outcome definitions, such as extubation success versus failure and whether patients needed to be put back on the ventilator within a defined window. Dr. Seely noted that electronic health record (EHR) systems are a major investment, but they do not automatically produce standardized, connected datasets that are ready for this kind of modelling and validation.
The second is adoption. Decision support is meant to support clinical judgement, not replace it, but it only helps if clinicians actually use it. In practice, that depends on clinicians being interested, curious and open to additional decision-support information. It also depends on trust, which often comes from experience. Many clinicians want to try a tool only after it is rigorously tested, before it becomes part of routine practice.
The third is sustainability under regulation. For regulated clinical software, approval is not the finish line. Ongoing quality systems, audits, compliance work and significant costs continue year after year. Dr. Seely describes those ongoing requirements as necessary as well as a real cost driver that shapes what is realistic for maintenance, pricing and scale, especially in a category that is still new.
What keeps the work moving is a real clinical need, backed by steady research and testing over time. For Dr. Seely, it is also driven by passion for the work, the belief that it has value and the fact that he finds it enjoyable. At the same time, moving from a clinical problem to a regulated tool in patient care takes time and sustained effort.
Dr. Seely believes Ottawa is well positioned to lead in the clinical applications of AI tools, as a strong environment for making AI at the bedside real. The ingredients he points to: super supportive institutions, multidisciplinary research depth, motivated and innovative clinical teams, a remarkable culture of collaboration, and strong leadership across the Ottawa health care community. Extubation Advisor is one example, which sits within a broader ecosystem where there is growing and compelling medical AI work ready to emerge.