Regional variation in critical care use has been described, but the reason for this variation remains unclear. Multiple factors contribute to the decision to admit patients to the intensive care unit (ICU), including local expertise, bed availability, and the perceived need for ICU care. We used retrospective data from a national sample of Veterans Affairs (VA) acute care hospitals, to better understand hospital variation in triage practices on admission.


We constructed a longitudinal cohort of adult admissions to any VA acute care hospital from April 2007 to June 2010, using data from the VA Inpatient Evaluation Center. We included the first admission for all non‐surgical patients who were admitted from the emergency department or outpatient clinic. We excluded VA hospitals with fewer than 10 ICU admissions. For each admission, we estimated predicted mortality using either the inpatient severity score (ISS), for those admitted to a non‐ICU ward, or the VA‐ICU score (ICUS), for those admitted directly to the ICU. Both scores are based on clinical, laboratory, and demographic variables collected in the 24 hours surrounding admission. We measured the proportion of patients that each hospital admitted directly to the ICU, stratified by ascending quintiles of mortality risk (as measured by the ICUS, if admitted directly to the ICU, or the ISS, if admitted directly to a non‐ICU ward). Each quintile contained equal numbers of patients.


During the study period, 716,023 unique patients were admitted to 121 VA hospitals. Of these patients, 68,389 (9.6%) were initially admitted to the ICU, and 647,634 (90.4%) were admitted elsewhere. Direct admissions to the ICU had a higher predicted mortality (ICUS, 0.084; 30‐day mortality, 7.7%) than admissions to a non‐ICU ward (ISS score, 0.042; 30‐day mortality, 3.5%). Transfers to the ICU from a non‐ICU ward had the highest predicted mortality (ICUS, 0.194; 30‐day mortality, 19.9%). The proportion of patients that hospitals triaged to the ICU increased with mortality risk (patient‐level quintiles of predicted mortality: < 0.6%, 0.6%–1.1%, 1.1%–2.2%, 2.2%–5.3%, 5.3% to 1.0; corresponding mean proportion admitted to the ICU: 5%, 8%, 11%, 15%, and 21%). For a given mortality risk, the proportion of patients sent to the ICU varied among hospitals (range by quintile of predicted mortality: 0%–33%, 0%–35%, 8%–41%, 8%–47%, and 5%–53%). Results were comparable when we restricted our sample to the 51 level 1 hospitals with similar ICU capabilities (mean proportion admitted to the ICU, by severity quintile: 0%–16%, 0%–31%, 2%–39%, 3%–47%, and 10%–51%).


Hospitals vary widely in their propensity to send patients with the same predicted mortality to the ICU. Access to critical care services may depend in part on the hospital where a patient seeks his or her care.


L. Chen ‐ none; M. Render ‐ none; A. Sales ‐ none; E. Kennedy ‐ none; W. Wiitala ‐ none; T. Hofer ‐ none