Serial acquisition of data to predict one-year mortality rate after acute myocardial infarction

Eight hundred sixty-six patients with acute myocardial infarction (AMI) were enrolled in a prospective study to determine optimal predictors of long-term prognosis. During 12-month (mean) follow-up there were 65 cardiac deaths and 21 nonfatal repeat AMIs. Twenty-nine variables (from the history, physical examination, serum chemistries, ambulatory monitor, radionuclide ventriculogram and exercise test) were arranged into in 5 sequential groups according to the time at which results became available during hospitalization. Multivariate analysis (logistic regression) and receiver-operator characteristic curves were used to assess improvement in prediction of mortality or repeat AMI by addition of each group of variables. The first group of independent predictors included rales, left bundle branch block and symptom status at 1 month before admission. Addition of information from ambulatory monitoring or serum chemistry did not improve prediction. Radionuclide ejection fraction made a statistically significant, independent contribution to mortality prediction. Of the final group the only exercise test variable that contributed independently to prediction was whether the patients took the test. However, receiver-operator characteristic curves showed that improvement in sensitivity and specificity by addition of information from the radionuclide scan and exercise test was clinically insignificant. Our results imply that costly tests after AMI should be reserved for specific indications and not applied universally for prognosis. Although these tests were highly predictive individually, each test generally added little to preexisting prognostic information.