Date of Award

January 2024

Document Type

Open Access Thesis

Degree Name

Medical Doctor (MD)



First Advisor

Lloyd N. Friedman


The prediction of which patients can be liberated successfully from mechanical ventilation and which patients require continued support remains a challenge for clinicians. While premature extubation comes with many risks, including respiratory distress, loss of airway, complications of reintubation, and increased mortality, delayed extubation comes with risks including ventilator-associated pneumonia, ventilator-induced lung injury, airway trauma, and unnecessary sedation. With such serious clinical implications, there is a strong and persistent interest among clinicians to identify better models to predict extubation success. Multiple respiratory parameters for predicting weaning success have been assessed over the past several decades, with particular attention to respiratory rate, tidal volume, minute ventilation, and negative inspiratory force (NIF). Of these metrics, several predictive indices have been formulated and proposed, of which the Rapid Shallow Breathing Index (RSBI) has achieved the most widespread use in intensive care units (ICUs) across the United States. When originally validated, tidal volume and respiratory rate were measured after one minute of unsupported spontaneous breathing to calculate the index. Current practices in many ICUs across the country calculate the RSBI after 30 minutes of pressure supported-ventilation. No studies have compared current practices of weaning head-to-head with the original 1-minute method of weaning evaluated when the RSBI was first proposed. This study compares in each patient the originally proposed methodology of a 1-minute unsupported spontaneous breathing trial (SBT) to the more widely utilized 30-minute pressure-supported SBT, and further seeks to assess alternative respiratory metrics and indices that can be applied to the prediction of extubation outcome. From February 2023 to September 2023, we enrolled mechanically-ventilated adults to receive a 1-minute trial of unsupported spontaneous breathing followed by a ventilator maneuver to measure the maximum NIF, immediately followed by the 30-minute trial on pressure support ventilation that is the standard in the Yale-New Haven Hospital Medical ICU. The RSBI, tidal volume, respiratory rate, minute ventilation, and numerous other respiratory metrics were recorded both during the breathing trials and on resting ventilator settings. For our assessed outcome, we compared patients who required non-invasive ventilation and/or reintubation within 48 hours following extubation to patients with no additional pressure support requirements within the same timeframe, defined as failure and success, respectively. We found that the RSBI calculated during the 1-minute unsupported SBT was significantly higher than the RSBI calculated during the 30-minute pressure supported SBT, and furthermore that the unsupported RSBI was a slightly better predictor of extubation success. We compared the predictive power of the RSBI under both conditions to several new indices we have proposed, including [NIF*Compliance/TVvent] and [(Plateau – PEEP)/NIF]. Our new indices, particularly those incorporating NIF, proved to be superior predictors of extubation success. We conclude that indices incorporating NIF and compliance have the potential to serve as more accurate predictors of extubation outcome. With respect to recent strides in the development of artificial intelligence (AI), we recognized a path forward for an AI model that can reliably predict extubation outcome. We have demonstrated this potential in the form of a neural network that accurately assesses the probability of successful extubation. This proof-of-concept analysis can serve as a springboard for future AI innovation in ventilator management and critical care.


This is an Open Access Thesis.

Open Access

This Article is Open Access