JMIR Publications

ABSTRACT

Background:

Patients who rely on ventilators for extended periods often experience a lower quality of life. Prolonged ventilation times may also increase mortality rates and associated medical expenditure costs. Respiratory therapists must perform a complex and time-consuming ventilator weaning assessment, typically spanning 48 to 72 hours. Traditional weaning methods depend on manual assessments, which can be subjective, prone to human error, and time-consuming.

Objective:

In this study, we aimed to establish a predictive model based on spontaneous breathing test criteria derived from a respiratory information system's data. Our goal is to inspire and motivate the improvement of the quality of respiratory therapy for ventilator patients.

Methods:

The study utilized a retrospective cohort design and deep learning neural networks. It included patients aged 20 and older who had been admitted to the intensive care unit of a medical center in central Taiwan from January 1, 2016, to December 31, 2022. The study included physiological measurements from 3,686 patients taken at different times, with 6,536 records. Among these records, 3,268 passed the SBT, while the other 3,268 did not pass.

Results:

The model's performance in the training dataset was exceptional, instilling confidence in its reliability. The overall performance for all categories in the training set was also high, with an average precision of 91.91%, recall of 99.91%, and F1 score of 99.9%. These high scores reassure us of the model's reliability. In the testing dataset, the overall average for all categories was a precision of 87.46%, a recall of 87.47%, and an F1 score of 87.46%.

Conclusions:

The study successfully developed a predictive model for spontaneous breathing test outcomes using deep learning neural networks. The model demonstrated exceptional performance with high precision, recall, and F1 scores in training and testing datasets, instilling confidence in its reliability and effectiveness in clinical settings. Implementing this model can significantly enhance the efficiency and accuracy of ventilator weaning assessments, thereby improving the quality of respiratory therapy and reducing ventilator dependence, a promising step towards better patient outcomes.