A17-01 A Location Agnostic EHR Definition of Critical Illness Using the Common Longitudinal ICU Data Format (CLIF)

Background Traditional critical illness definitions in observational research rely on ICU admission, as documented in administrative data, potentially missing patients who receive life-sustaining interventions in other locations. Healthcare systems vary significantly in ICU admission practices, creating challenges for multi-center research and quality benchmarking. The Common Longitudinal ICU data Format (CLIF) enables standardized extraction of clinical interventions independent of location. We hypothesized that a location-agnostic, intervention-based definition would identify substantial critical illness outside ICUs and reveal heterogeneity in care delivery models across institutions. Methods We analyzed data from nine U.S. academic medical centers and MIMIC-IV (2011-2025) harmonized into CLIF format. CLIF captures demographics, vitals, laboratory values, medications, respiratory support, and location data with standardized terminology. We defined critical illness hierarchically: (1) any ICU admission; (2) advanced respiratory support (invasive/noninvasive ventilation, high-flow oxygen); (3) continuous vasopressor infusions; or (4) other deaths. We excluded hospitalizations where advanced respiratory support and vasopressor infusions were administered solely in the procedural units. Primary outcomes were prevalence of location-agnostic critical illness identification and variation in definitions across sites. Results We identified 873,258 hospitalizations from 681,163 unique patients across 51 hospitals. The cohort was diverse (median age 65 years,60.9% White,20.9% Black) with median SOFA score of 3 (site range: 2-4). Hospital mortality was 27.5% (range: 14.6%-26.4%). Median ICU and hospital length of stay were 1.7 days (range: 0.8-5) and 6.9 days (range: 2.6-18.6), respectively. 10.97% (n = 95,795) of the cohort met the critical illness criteria without ever being admitted to the ICU. Health-system level variation in ICU admission among critically ill patients ranged from 80.7% to 94.7% (I² = 99.4%, p < 0.001), demonstrating substantial heterogeneity in care delivery models, including variable availability of intermediate care units across institutions. Of 257,816 mechanically ventilated hospitalizations, 25.19% were first intubated outside the ICU, including in the emergency department (12.44%), procedural units (8.67%) and ward (2.13%). Conclusion CLIF enables reproducible, location-agnostic identification of critical illness independent of administrative location data. Nearly one in ten critically ill patients received care outside ICUs. Marked variation across institutions suggests different care delivery models and ICU admission thresholds. Standardized EHR-based definitions are essential for equitable multi-center critical care research and quality improvement initiatives.