From Internal Validation to External Validation: An Artificial Intelligence-Based Study on Predicting Optimal Timing for Mechanical Ventilation Weaning in ICU Patients

Abstract

Mechanical ventilation weaning is critical for ICU patients, as prolonged or premature use can cause adverse outcomes and resource waste. Using six years of ICU data, Chi Mei Medical Center developed two-stage AI predictive models to optimize the timing for "Trying Weaning" and "Actual Weaning." The original Chi Mei models compared to external validation at Kaohsiung Medical University Hospital demonstrated AUCs of 0.981 vs. 0.915 for "Trying Weaning" and 0.915 vs. 0.866 for "Actual Weaning." The findings demonstrate that AI-assisted tools, supported by strong external validation results, can effectively reduce ventilation time, complications, and costs. Future research will focus on model optimization and multi-center validation.