Back to Annual Meeting Page
133rd Annual Meeting & Exposition
December 10-14, 2005
George J. Stukenborg, PhD1, M. Norman Oliver, MD, MA2, Douglas P. Wagner, PhD1, Frank E. Harrell, PhD3, Steven Heim, MD2, Andrew M. Wolf, MD4, and Alfred F. Connors, MD5. (1) Health Evaluation Sciences, University of Virginia School of Medicine, PO Box 800821, Charlottesville, VA 22908, 434-924-8437, firstname.lastname@example.org, (2) Department of Family Medicine, University of Virginia, Box 800729, Charlottesville, VA 22908, (3) School of Medicine, Department of Biostatistics, Vanderbilt University, S-2323 MCN 2158, Nashville, TN 37232, (4) Department of Internal Medicine, University of Virginia School of Medicine, PO Box 800744, Charlottesville, VA 22908, (5) Medicine, Case Western Reserve University, 2500 MetroHealth Drive, Cleveland, OH 44109
Mortality risk adjustment models are used in studies comparing hospital mortality rates for patients hospitalized with fresh acute myocardial infarction. This retrospective observational cohort study used California hospital discharge abstract data to examine in-hospital mortality among 120,706 acute myocardial infarction patients. This study closely reproduces mortality risk adjustment methods originally developed to compare observed to expected hospital mortality rates in the 1996-1998 California Hospital Outcomes Project. Those original models are compared to models using alternative methods for measuring comorbid disease, including the method of Elixhauser and colleagues and a new method using present-at-admission diagnoses. Model statistical performance was validated by using the developed models to predict mortality outcomes in an identically defined independent study population. The model using present-at-admission diagnoses obtained a validated C statistic of 0.86, which substantially exceeded the performance achieved by the other models we evaluated. The model using the Elixhauser et al. method had a validated C statistic of 0.79. The model that was originally developed for use in the California study had a validated C statistic of 0.76, which was nearly equivalent to the C statistic value of 0.77 reported in the original study. Our findings indicate that states that require hospitals to identify which secondary diagnoses are present at admission can use this information to substantially improve the statistical performance of mortality risk models that adjust for patient differences in comorbid disease, and can achieve more accurate adjustments for patient differences in studies comparing observed to expected hospital mortality rates.
Keywords: Statistics, Treatment Outcomes
Presenting author's disclosure statement:
I wish to disclose that I have NO financial interests or other relationship with the manufactures of commercial products, suppliers of commercial services or commercial supporters.
The 133rd Annual Meeting & Exposition (December 10-14, 2005) of APHA