Introduction: Spontaneous awakening trials (SATs) and spontaneous breathing trials (SBTs) improve patient outcomes. However, measurement of their delivery in the electronic health record (EHR) depends on clinician documentation. We aimed to develop rule-based algorithms using structured EHR data to automate the detection of SAT and SBT delivery. Methods: We identified mechanically ventilated adults at 13 US hospitals across 5 health systems in the Common Longitudinal Intensive Care Unit Format (CLIF) consortium from January 1st, 2020, to December 31, 2022. SAT eligibility was defined on days with mechanical ventilation, no paralytics, and ≥4 hours of continuous sedatives or opiates use from 10 PM prior day to 6 AM current day. SBT eligibility required ≥12 hours of controlled ventilation and ≥2 hours of stable hemodynamics and respiratory status (FiO2 ≤50%, PEEP ≤8, SpO2 ≥88%, norepinephrine or equivalent ≤0.2 mcg/kg/min). SAT delivery was defined as ≥30 minutes off sedatives and opiates and RASS≥0 within 45 minutes. SBT delivery was defined as transition to support mode (pressure support and PEEP ≤8 or CPAP ≤5 cm H2O) for ≥2 minutes. Algorithm performance was evaluated at CLIF sites with available clinician flowsheet documentation, using accuracy, precision, recall, and specificity. Results: Among 68,862 mechanical ventilation encounters mean age 61 years (SD 16); 41% female, 102,172 patient-ventilator days were eligible for SAT detection. The SAT algorithm identified delivery on a median 17% (IQR 16%–20%) of eligible days, with moderate concordance against flowsheets: accuracy 0.59 (IQR 0.50–0.63), precision 0.61 (0.36–0.83), recall 0.23 (0.20–0.27), and specificity 0.89 (0.84–0.93). Of 159,715 SBT-eligible days, the algorithm identified SBTs in 34% (IQR 30%–41%), with stronger concordance: accuracy 0.72 (0.44–0.80), precision 0.44 (0.23–0.69), recall 0.80 (0.69–0.85), and specificity 0.76 (0.25–0.89). Conclusions: Structured EHR data can be used to detect SAT and SBT delivery via rule-based algorithms. Variable concordance with flowsheet documentation may reflect differences in documentation practices or clinical performance of SATs and SBTs. These findings support the use of algorithmic surveillance in detection of SATs and SBTs.
Jain et al. (Sun,) studied this question.