PURPOSE:The use of models to risk stratify patients (APACHE, SOFA) has become common in clinical research and is now being used for resource allocation and to compare non-research outcomes. Physician specific assessments of observed/expected mortality are now available to a broad audience external to medical center. The University Hospital Consortium (UHC) methodology is used to compare outcomes from Academic Hospitals. Unlike the APACHE IV, the UHC methodology has not been validated in ICU patients. We compare APACHE IV and the University Health System Consortium (UHC) mortality prediction models for patients admitted to an MICU.
METHODS:A prospective observational study was conducted on patients admitted to the University of Virginia MICU between April 30 and July 31, 2007. APACHE IV was calculated on data available during the initial 24 hours. UHC data was available in an administrative database and data could be matched to patient and care episode. Predicted mortality scores from both methodologies were then compared using Pearsons Correlation Coefficient using SPSS v 16.0 for windows.
RESULTS:We are presenting preliminary data on 73 of 200 patients to be analyzed. The average predicted mortality for the UHC model was 20.34% and for the predicted mortality for the APACHE IV model was 26.98%. The Pearson Correlation Coefficient is .46 (95% CI 0.26, 0.63) with a p-value less than 0.01. The r^2 for this is 0.22, suggesting the APACHE IV and the UHC models only agree twenty-two percent of the time.
CONCLUSION:We expected poor correlation between APACHE IV and UHC prediction models due to fundamental differences in their respective parameters. The UHC model is more heavily weighted by co-morbidities while the APACHE IV accounts for some co-morbidities but weighs physiologic parameters more heavily. Additionally, the UHC model does not account for severity of illness on admission which we believe to be a significant factor in determining mortality.
CLINICAL IMPLICATIONS:Given the small sample size used for our preliminary data, it would be premature to endorse any conclusion at this time.
DISCLOSURE:Katherine Schafer, No Financial Disclosure Information; No Product/Research Disclosure Information