Ai identified intubated patients who pass standard extubation readiness tests but remain at twice the risk of returning to a ventilator.
RT’s Three Key Takeaways:
- Hidden High-Risk Patients: A machine learning model identified that nearly 1-in-5 patients who meet standard readiness criteria face more than double the rate of breathing tube replacement and higher mortality.
- Systemic Complication Factors: Underlying conditions such as sepsis, ARDS, fluid overload, and extended time on mechanical ventilation drive extubation failure beyond conventional breathing measurements.
- Earlier Weaning Opportunities: The model classified nearly 40% of patients who failed traditional weaning criteria as low risk, indicating potential opportunities to safely remove ventilators earlier.
A machine learning model has identified a group of mechanically ventilated patients who met traditional criteria for extubation but experienced more than twice the reintubation rate and higher mortality compared with patients classified as low risk by the model, according to a study presented at Chest 2026.
Extubation failure occurs in 8% to 15% of mechanically ventilated patients and independently doubles mortality, the researchers noted in the release. While traditional clinical predictors—such as rapid shallow breathing index (RSBI), fraction of inspired oxygen (FiO2), and positive end-expiratory pressure (PEEP)—assess respiratory mechanics, the authors reported that they often fail to capture systemic factors that contribute to reintubation.
To address these diagnostic blind spots, investigators developed an ML model aimed at predicting extubation failure and identifying patients at elevated risk despite meeting conventional weaning thresholds. Using data from the MIMIC-IV database, the research team analyzed 23,379 adult patients who underwent their first extubation following mechanical ventilation, according to the study release. Within the study cohort, extubation failure was defined as reintubation within 72 hours and occurred in 8.3% of patients.
Evaluating Occult High-Risk Patients
The investigators trained a gradient boosting machine using 52 pre-extubation clinical features, encompassing ventilator mechanics, hemodynamics, laboratory values, fluid balance, vasopressor status, ventilation duration, and comorbidities, the release stated. Patients were subsequently categorized by both traditional extubation guidelines and ML-predicted risk.
According to the study findings, the model identified 4,370 patients, or 18.7%, who met all traditional low-risk criteria but were classified as high risk by the algorithm. These occult high-risk patients experienced a 15.4% reintubation rate and an 18.7% mortality rate, compared with 7.3% and 11.1%, respectively, among patients who met traditional criteria and were categorized as low risk by the model.
The release indicated that patients in this occult high-risk cohort presented with longer durations of mechanical ventilation, a higher positive fluid balance, and increased rates of sepsis, pneumonia, congestive heart failure, and ARDS.
“Nearly one in five extubated patients meets all traditional weaning criteria but have twice the reintubation and mortality risk, driven by systemic factors like prolonged ventilation, ARDS, sepsis, and fluid overload, which is not readily reflected in conventional respiratory mechanics-based assessment,” said Jeril Lasington, MD, CHEST 2026 presenting author. “Integrating machine learning risk scores alongside standard weaning parameters could flag these occult high-risk patients for enhanced postextubation monitoring, prophylactic noninvasive ventilation, or delayed extubation.”
Potential for Earlier Weaning
In addition to detecting hidden vulnerabilities, the findings suggested that predictive models could expand extubation eligibility in select healthcare settings. The model classified 39.4% of patients who failed traditional weaning criteria as low risk. This group demonstrated a 4.2% reintubation rate and a 4.5% mortality rate, signaling that standard weaning criteria alone may delay extubation unnecessarily for certain individuals.
The authors concluded that combining systemic variables with conventional respiratory metrics provides a clearer picture of extubation readiness, helping clinicians pinpoint patients requiring closer postextubation support as well as candidates who can safely undergo earlier extubation.
