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


Clinical Prediction Guide

Benefit from thrombolysis in acute MI was predicted in the emergency department

ACP J Club. 1998 Jan-Feb;128:25. doi:10.7326/ACPJC-1998-128-1-025


Source Citation

Selker HP, Griffith JL, Beshansky JR, et al. Patient-specific predictions of outcomes in myocardial infarction for real-time emergency use: a thrombolytic predictive instrument. Ann Intern Med. 1997 Oct 1;127:538-56.


Abstract

Objective

To develop a clinical prediction guide that identifies, in the emergency department, patients with acute myocardial infarction (MI) who are likely to benefit from thrombolytic therapy.

Design

Derivation and validation of a thrombolytic predictive guide using patient-specific data from 13 clinical trials.

Setting

Clinical trial data were from 107 hospitals in the United States.

Patients

4911 patients who were ≥ 1 mm in ≥ 2 contiguous leads on the electrocardiogram (ECG), and no contraindications to thrombolytic therapy. For mortality outcomes, the database of patients was randomly allocated (2:1 ratio) to a development data set (n = 3263) and a test data set (n = 1648).

Description of prediction guide

Logistic regression was used to develop component predictive instruments for each of 5 outcomes that included clinically important and statistically significant variables. 2 ECG-based variables were produced to show 2 indicators of the effect of thrombolytic therapy: acute infarction size based on ST segments, and a sign that the infarction was still early in its course based on T-wave changes.

Main outcome measures

Predictors of 30-day mortality, 1-year mortality, cardiac arrest within 48 hours of the first ECG, intracranial hemorrhage, and bleeding requiring transfusion. The areas under receiver-operating characteristic (ROC) curves were calculated to evaluate the performance of each component instrument in both the derivation and validation sets.

Main results

The overall Thrombolytic Predictive Instrument Database comprised 4911 patients. 3483 patients (71%) received thrombolytic therapy. The predictors for 30-day mortality were patient age {(odds ratio [OR] 1.75/10 y), systolic blood pressure in anterior or posterior acute MI (OR 0.76/10 mm Hg), history of diabetes (OR 2.52), heart rate (OR 1.31/10 beats per min), Q waves without ST-segment elevation in inferior acute MI (OR 1.38), right bundle-branch block (OR 1.78), and use of thrombolytic therapy (OR 0.30)}*. The predicted probabilities of dying within 30 days ranged from 0.2% to 80%. For all outcomes, the areas under the ROC curves ranged from 0.77 for cardiac arrest to 0.84 for 30-day mortality.

Conclusion

A thrombolytic predictive guide was helpful in quantifying the probability that an individual patient with acute myocardial infarction would benefit from thrombolytic therapy or have an adverse outcome caused by thrombolytic therapy.

Source of funding: Agency for Health Care Policy and Research.

For article reprint: Dr. H.P. Selker, Center for Cardiovascular Health Services Research, Division of Clinical Care Research, New England Medical Center, 750 Washington Street #63, Boston, MA 02111, USA. FAX 617-636-8023.

*Numbers calculated from data in article.


Commentary

Clinicians must incorporate information from an ever-increasing number of large studies into their practices. How can they directly apply these results to everyday care of their patients?

Selker and colleagues provide a good example of a way to address this question. From the data of 13 clinical trials, they developed a clinical prediction guide to help clinicians who offer thrombolytic therapy to patients with acute MI.

The guide incorporates information from the history, physical examination, and emergency department ECGs. The various components could be entered into a computerized ECG that rapidly calculates the probability that a patient will encounter a particular outcome, such as 30-day mortality or hemorrhagic stroke with and without treatment.

As described in the editorial in this issue (1), the creation of a prediction guide is a complex, multistep process. Selker and colleagues followed these steps but did so retrospectively. Therefore, before we can enthusiastically apply such a guide, it needs to be tested prospectively to evaluate accuracy in a clinical setting. If accuracy is proved, then the guide's effect on clinician behavior and patient outcomes should be assessed, ideally in a randomized controlled trial comparing patient outcomes when the guide is used with patient outcomes when it is not used.

Further work on this model is clearly needed, but it is an exciting example of the possibilities for prediction guides and how they may assist clinicians in everyday care of their patients. However, it is important to remember that these guides are meant to complement clinical acumen and not to replace it.

Thomas McGinn, MD
Montefiore Medical CenterBronx, New York, USA

Thomas McGinn, MD
Montefiore Medical Center
Bronx, New York, USA


Reference

1. McGinn T, Randolph A, Richardson S, Sackett D.Clinical prediction guides. ACP J Club. 1998 Jan-Feb;128:A14-15. Evidence-Based Medicine. 1998 Jan-Feb;2:4-5.