Objective: To develop a prediction model of death within 30 days of hospital admission for Medicare patients with acute myocardial infarction that would permit use of risk-adjusted mortality rates as hospital quality measures.
Design: Retrospective cohort study using data created from medical charts and administrative files.
Setting: All acute care hospitals in Alabama, Connecticut, Iowa, or Wisconsin.
Patients: A cohort of 14,581 patients with acute myocardial infarction covered by Medicare in 1993.
Results: The unadjusted 30-day mortality rate was 21%, ranging from 18% in Connecticut to 23% in Alabama. The 4 largest contributors to variability in mortality rates were mean arterial pressure, age, respiratory rate, and serum urea nitrogen level. The area under the receiver operator characteristic curve was 0.79 in a developmental sample of 10 936 patients and 0.78 in a validation sample of 3645 patients. Based on admission variables, we were able to explain 27% of the variability in 30-day mortality rates. During the index admission, aspirin, beta-blockers, angiotensin-converting enzyme inhibitors, and thrombolytic agents were used in 72%, 39%, 32%, and 15% of patients, respectively. Explained variation increased by 6 percentage points to 33% when drug therapies and revascularization procedures performed during the index admission were added to the model predictors.
Conclusions: Short-term mortality remains high for elderly patients with acute myocardial infarction, and a large percentage of variation remains unexplained after controlling for admission severity. Part of the unexplained variability can be explained by the location of the admitting hospital; some of the remaining unexplained variation may reflect differences in quality of care or unmeasured differences in disease severity. Researchers should develop quality indicators based on process measures for acute myocardial infarction and should incorporate these measures into mortality models to determine whether quality accounts for variation in 30-day mortality rates beyond that explained by clinical status at admission.