Medical-AI trial: predicting the success of a spontaneous breathing trial with machine learning — ICU ventilator weaning from biosignals and biomarkers (NCT05886803)

1Weaning from mechanical ventilation

A critically ill patient who cannot breathe well on their own is supported by a ventilator. As they recover, the ventilator is withdrawn (weaning), but judging the right moment is hard. Weaning too early can force reintubation, adding burden and risk, while weaning too late invites ventilator-associated complications and a longer stay.

Weaning indeed occupies much of an ICU stay; per the summary, the first attempt fails in about 20% of patients, and mortality can reach 38% in the most difficult cases.

2Predicting the outcome of a breathing trial

So this study examines, as an observational study, whether the success of the spontaneous breathing trial (SBT, in which ventilator support is reduced to see how well the patient breathes on their own for a set time) can be predicted in advance with machine learning. Per the registry summary, it uses biosignals such as heart rate and respiration and biomarkers from blood as cues to estimate whether the SBT will succeed.

As research applying AI to clinical decision support grows, from managing septic shock to predicting renal failure, this study focuses that lens on the weaning decision. The target is 500, and it is recruiting.

3Too early, and too late

Weaning from a ventilator harms the patient if it errs in either direction. That difficulty is the reason prediction is sought.

The study therefore asks whether machine learning can predict, in advance, the success of the spontaneous breathing trial conducted before weaning. It is an observational study using biosignals such as heart rate and respiration together with blood biomarkers, targeting 500 cases. Weaning decisions rest heavily on experience and vary by institution and clinician.