Evaluation of Computational Identification of Invasive Mechanical Ventilation at 3 Medical Centers

Monday, March 23, 2026

2:30 PM - 3:30 PM Central Time

Location: Connections Central - RST 11

AG

Anthony Gerlach, PharmD, FCCM

Ohio State University Wexner Medical Center, Ohio

Disclosure information not submitted.
JC

James Campbell, MD

Professor of Medicine
University Of Nebraska Medical Center

Disclosure information not submitted.
AK

Anna Krupp, PhD, MSHP, RN

Assistant Professor
University Of Iowa

Disclosure information not submitted.
KC

Katie Circo, BSN, RN, CCRN, CBRN

Clinical Quality Nurse Lead
University Of Nebraska Medical Center

Disclosure information not submitted.
EV

Eduard Vasilevskis, MD

Professor of Medicine
University Of Wisocnsin

Disclosure information not submitted.
CG

Carol Geary, PhD, MBA, RN

Associate Professor
University Of Nebraska Medical Center

Disclosure information not submitted.

Introduction: Pragmatic trials using computationally obtained electronic medical record (EMR) data can improve clinical trial efficiency but require clinical validation. This study aims to evaluate the accuracy of invasive mechanical ventilation (iMV) data obtained computationally at three sites through local clinician quality review (CQR).

Methods: BEST-ICU is a multi-institutional, pragmatic implementation trial comparing ABCDEF bundle performance supported by an RN facilitator compared to an electronic dashboard. Eligible patients include any adult (>18 years) receiving iMV in 12 ICUs. Institutions use the same commercial EMR (EPIC) and study data are shared through PCORnet approved dataset. At each site CQRs were done for one calendar day in 4th quarter 2024. Clinical identification of iMV was treated as the gold-standard and compared to electronic phenotypic identification The primary outcome was sensitivity and specificity to in iMV for computational versus CQR. Description statistics were used.

Results: Detailed data definitions and computational approach plans were developed including computable phenotypes for iMV collaboratively with knowledge of study requirements and clinical norms at each site. At all three sites both computational approaches and CQR identified all 169 patients admitted to ICUs (site A n=66; site B n=82; Site C n=21). Computationally 78 patients received iMV compared to 80 from CQR p=0.91. At site A, 3 (4.5%) patients were discrepant. All were noted to have procedural codes for intubation computationally, but study-relevant iMV was not identified in CQR. At site B, 5 (6.1%) patients were discrepant, and none at site C. The sensitivity was 93.5%, specificality 96.7% and accuracy of 95.3%.

Conclusions: There was high accuracy for iMV with computationally obtained EMR data and CQR. Site variances for iMV were heterogenous with unique site differences in clinical and documentation procedures that affect computation requirements. Validation of computationally obtained data elements using CQR is necessary to ensure adequate data quality. Studies are needed to elucidate most efficient data validation approaches.