Current issues of ACP Journal Club are published in Annals of Internal Medicine


Prognosis

Prediction of in-hospital mortality after myocardial infarction was not accurate

ACP J Club. 1991 Jan-Feb;114:26. doi:10.7326/ACPJC-1991-114-1-026


Source Citation

Alemi F, Rice J, Hankins R. Predicting in-hospital survival of myocardial infarction. A comparative study of various severity measures. Med Care. 1990;28:762-75.


Abstract

Objective

To compare the accuracy of various indices to predict in-hospital survival after myocardial infarction compared with the arbitrary prediction that all patients will live.

Design

Retrospective hospital record review of all charts with the diagnosis of acute myocardial infarction (occurring up to 8 weeks before hospitalization), excluding those occurring during recovery from noncardiac surgery.

Setting

12 hospitals ranging in size from 40 to > 300 beds, in the New Orleans metropolitan area in the United States.

Patients

775 Medicare patients (mean age 73 y, range 40 to 96 y) discharged in 1985 with a diagnosis of myocardial infarction (ICD- 9-CM codes 410.0 through 410.9) including cardiac surgeries requiring pump support.

Assessment of prognostic factors

2 heart disease-specific, indices (Ischemic Heart Disease Index [IHDI] and the Predictive Index for Myocardial Infarction [PIMI], and 5 general severity indices (Acute Physiological and Chronic Health Evaluation [APACHE II], Medisgroups [MDGRP], Computerized Severity Index [CSI], Patient Management Categories [PMC], and Computerized Disease Staging [CDS], were compared.

Main outcome measures

Receiver operating characteristic curves (ROCS) were used to evaluate the sensitivity and specificity of each index in predicting mortality compared with a random prediction of mortality. The scores of each index were compared with in-hospital survival using logistic regression.

Main results

The assumption that all patients would survive correctly predicted 78% of the outcomes. Scoring with IHDI, APACHE II, and PIMI did not improve correct classification of outcome. MDGRP, PMC, and CDS improved the percent of correctly classified cases by 3%, 3%, and 5%, respectively. Each index explained a small but statistically significant portion of in-hospital survival (P < 0.001). 17% to 22% of the deaths were not predicted by these prognostic indices.

Conclusion

The various indices were somewhat useful in predicting in-hospital mortality in patients, after myocardial infarction, although up to 22% of deaths remained unexplained.

Sources of funding: Health Care Financing Administration.

Address for article reprint: Not given.


Commentary

The study shows that several indices of disease "severity" do not accurately predict hospital survival of patients with myocardial infarction. There are often methodologic reasons why these predictive indices show poor accuracy (1, 2). First, when an index is "fitted" statistically to 1 group (the "training sample") and then applied to another (the "validation" sample), the fit is usually less good. Second, in the validation sample, there are usually differences in selection of patients and measurement of variables and outcomes. For example, in the study by Alemi and colleagues, patients were included who had had a myocardial infarction up to 8 weeks before hospitalization, whereas most studies of indices among such patients would include only those who were seen within a much shorter time from the onset of their event. Third, readers should note that the "severity measures" discussed in this study included some that were not designed to predict survival. As examples, PIMI was designed to predict whether a person presenting with chest pain would prove to have a myocardial infarction, and the CDS, CSI, and PMC indices were developed for case mix or financial management. It is not surprising that these indices did not predict survival well. Moreover, the findings of this study do not undermine the use of these measures for the purposes for which they were developed. This study does not provide information of practical value for the clinical management of patients with myocardial infarction, except in the negative sense that the indices that were assessed should not be used to predict the prognosis of patients with myocardial infarction.

Allan Detsky, MD, PhD
Toronto HospitalToronto, Ontario, Canada


References

1. Wasson JH, Sox HC, Neff RK, Goldman L. Clinical prediction rules: applications and methodologic standards. N Engl J Med. 1985; 313:793-9.

2. Charlson ME, Ales KL, Simon R, MacKenzie CR. Why predictive indexes perform less well in validation studies. Arch Intern Med. 1987;147:2155-61.