Diagnosing acute myocardial infarction The comparison of various ECG interpretive criteria for use by ambulance paramedics in the diagnosis of acute myocardial infarction (AMI)I raises several interesting issues regarding the interpretation of diagnostic tests. In particular, three of the authors’ implied assumptions are questionable: first, that the post-test probability of disease (ie, positive and negative predictive values) can be transferred from a diagnostic test evaluation performed in one clinical setting to another; second, that diagnosis can be made from test results without recourse to the pre-test probability of disease; and finally, that a single criterion (or set of criteria) should be selected as the basis of clinical decisions. The positive and negative predictive values of a test are affected by the performance of the test and the population in which they are estimated. They have no meaning when applied to a different population. The study by O’Rourke et al. I was conducted in a heterogeneous population of coronary care inpatients and various cardiology outpatients. The overall probability of Ah41 in these patients (4.0%) differs markedly from the probability of AM1 in patients for whom the ECG criteria are being selected, people with chest pain seen by ambulance paramedics(43%in their prospective series). Hence, the application of predictive values from this study to the intended target population is not appropriate. The role of tests in the diagnostic process is to modify the best clinical estimate of the probability of disease being present.2 A ‘positive’test increasesand a ‘negative’test decreases the prob ability. Pre-test probabilities, based on the prevalence of disease in a given clinical setting, are converted to post-test probabilities, which are the basis of subsequent management decisions. The pre-test probability, therefore, clearly influences the interpretation of test results. For example, an equivocal ECG in a patient with convincing clinical features of AM1 (ie, a high pre-test probability) will result in a greater post-test probability of Ah41 than the same ECG in a patient with less convincing clinical features (lower pre-test probability). Dichotomising the outcome of multilevel tests is wastehl of the information provided by such tests. This occurs if only one level of the ECG interpretative algorithms is selected for use. In fact, each of the test results raises (or lowers) the probability of disease to a varying degree and this can be quantified by calculatinglikelihood ratios (LRs)for a hierarchy of test results.2 These are related to the sensitivity and specificity of a test and are independent of the prevalence ofdisease in the study population. The post-test probability ofdisease is derived from the pre-test probability and the LR by a simple arithmetic manipulation or by using a published nomogram.z T o use LRs effectively in the context of this study two decisions must be made, What is the pre-test probability of AM1 and what is the post-test probability at which the decision to use thrombolytic therapy would be appropriate? The pre-test probability is estimated from the prevalence of the disease in the specified setting. It should be possible to measure the LETI‘EKS AND CASE REPORTS
prevalence of AM1 in three or four (or more) clinical scenarios likely to be encountered by paramedics. The post-test probability at which thrombolysis is appropriate depends on the probability and severity of adverse events. Using the prhaples of decision analy~is”~ an appropriate post-test probability threshold for treatment can be established. This approach to diagnostic decision making is an efficient and, we believe, honest one which accords more closely to high quality clinical diagnostic practice than the dichotomous, isolated interpretation of a single diagnostic test result as has been suggested for ambulance paramedics. It will not, however, be necessary for paramedics, huddled in the back of an ambulance, careering through the traffic at high speed, to struggle with nomograms and the intricacies of medical decision analysis. All the probabilities, LRs and thresholds can be estimated before the programme is established. Combinations of clinical features relevant to diagnosis of AM1 and the risk of adverse effects of thrombolysis, and ECG reports which should lead to the initiation of thrombolysis can be prescribed. An approach such as this should lead to optimal use of thrombolysis by paramedics in the amelioration of adverse outcomes from AMI. The extent to which this does occur should be the subject of ongoing evaluation. G. B. MARKS, NH&MRC Postgraduate Research Scholar, Department of Medicine, University of Sydney, NSW. R. G. CUMMING, Lecturer, Department of Public Health, University of Sydney, NSW. References 1. ORourke MF, Cook A, Carroll G er al. Accuracy of a portable interpretive ECG machine in diagnosis of acute evolving myocardial infarction. Aust NZ J Med 1992; 22: 9-13. 2. Sackett DL, Haynes RB, Guyatt GH, Tugwell P. Clinical epidemiology. A basic science for clinical medicine, 2nd ed. Boston: Little, Brown and Company, 1991. Chapter 4 Interpretationofdiagnostic tests. 3. Sox HC, Blatt MA, Higgins MC, Marton KI. Medical decision making. Boston: Butterworths, 1988, 151-60. 4. Glaziou P. Threshold analysis via Bayes nomogram. Med Deck Making 1991; 11: 61-2.
REPLY As William Osler stressed in his teachings, medicine is an art which is best practised by application of theoretic principles to disease processes. Drs Marks and Cumming are correct in presenting the ideal for evaluation and application of diagnostic tests; the reality is that the ideal is rarely practicable in real life clinical situations such as acute thrombotic coronaIy occlusion with evolving myocardial infarction, where the electrocardiogram comprises just one aspect of the total assessment, but where a decision for thrombolytic therapy must be made promptly. 387 Aust NZ J Med 1992; 22