Session number: 1050
Abstract title: Variation in risk adjusted VA intensive care unit hospital mortality
Author(s):
MLRender - VAMC Cincinnati and University of Cincinnati College of Medicine, Dept of Pulmonary/ Critical Care
HMKim - VA Ann Arbor HSR&D Center of Excellence
DWelsh - VA Ann Arbor HSR&D Center of Excellence
JJohnston - VAMC - Cincinnati and the University of Cincinnati College of Medicine, Dept of General Internal Medicine
KBickel - VAMC Cincinnati
TJHofer - VA Ann Arbor HSR&D Center of Excellence and the University of Michigan Department of Medicine
Objectives: To examine risk adjusted variation in hospital mortality of intensive care unit (ICU) patients from 17 Veterans Affairs (VA) hospitals and 44 ICUs.
Methods: We used a database of 18,021 first ICU admissions between February 1996 through July 1997, drawn from tertiary, large, and moderate size VA hospitals from 5 regions defining mortality at hospital discharge from the Patient File in local VA databases and merge with records from the Beneficiary Information Record Location Subsystem. We then created predicted mortality risks for the ICU patients using a recently validated VA ICU measure (AUROC 0.886) based on binary variables for diagnosis, comorbid illness, source of admission, and continuous weights created with cubic splines for continuous variables (age, and laboratory values). To create conservative standardized mortality rate (SMR; observed to predicted mortality), we used a random effects hierarchical model to account for the clustering of patients within a site and within units and included individual patient predicted mortality risk and observed deaths. We conducted a Monte Carlo simulation with 50,000 repetitions of the site effect on estimates of predicted mortality. A hierarchical model allows explicit modeling of differences related to site, random variation in site effects, and as clustering or nesting of cases.
Results: Hospital mortality was 11.2% for the 5,916 operative cases and 12,105 non-operative cases. An average of 1157 (S.D. 296) patients were contributed per hospital. Raw (unadjusted) in-hospital mortality rate and standardized mortality rates for the 17 sites varied significantly and rank ordering of sites by raw and standardized mortality rate yielded significantly different orders. Similar ranges and rank order effects have been reported from public and private sector ICUs using APACHE III (Acute Physiology and Chronic Health Evaluation III).
Conclusions: The underlying complexity of intensive care limits the use of raw mortality as an index of quality of care. Significant variation in SMR remain among VA sites, not accounted for by patient factors included within the risk adjustment model. Further work is needed to elucidate the site factors important in ICU outcome
Impact statement: Examination of the contributors to site variation in ICU mortality begins the process of evidence based ICU care.