Application of information theory to clinical diagnostic testing. The electrocardiographic stress test.

Abstract

SUMMARY The inherent imperfection of clinical diagnostic tests introduces uncertainty into their interpretation. The magnitude of diagnostic uncertainty after any test result may be quantified by information theory. The information content of the electrocardiographic ST-segment response to exercise, relative to the diagnosis of angiographic coronary artery disease, was determined using literature-based pooled estimates of the true-and false-positive rates for various magnitudes of ST depression from < 0.5 mm to 2 2.5 mm. This analysis allows three conclusions of clinical relevance. First, the diagnostic information content of exercise-induced ST-segment depression, interpreted by the standard 1.0-mm criterion, averages only 15% of that of coronary angiography. Second, there is a 41% increase in information content when the specific magnitude of STsegment depression is analyzed, as opposed to the single, categorical 1-mm criterion. Third, the information obtained from ECG stress testing is markedly influenced by the prevalence of disease in the population tested, being low in the asymptomatic and typical angina groups and substantially greater in groups with nonanginal chest pain and atypical angina.The quantitation of information has broad relevance to selection and use of diagnostic tests, because one can analyze objectively the value of different interpretation criteria, compare one test with another and evaluate the cost-effectiveness of both a single test and potential testing combinations.INFORMATION THEORY was developed as a means for quantifying the information content and the noise content in the audio signal transmitted over telephone channels.' Because the general theory provides a formal mathematical basis for quantifying information and uncertainty, it has been widely applied in other sciences as diverse as economics and engineering, but has been little used in clinical medicine,2' 3 presumably because its relevance was not apparent.The application of information analysis to diagnostic testing in medicine derives from the recognition that all clinical tests are imperfect. This imperfection introduces uncertainty (or "noise") into the interpretation of the test. The information (and, conversely, the uncertainty existing before and after a diagnostic test) can be quantified if one knows its three determinants. These are the pretest probability of disease in the individual being tested (prevalence), the conditional probability of the test response observed in a diseased population (the true-positive rate), and the conditional probability of this same response in a nondiseased population (the false-positive rate). The change in diagnostic uncertainty that occurs as a result of testing, then, is a measure of the test's effectiveness ( fig. 1 where N is the number of patients from which the value of P is derived. The complete bibliography for the literature review from which these data were obtained is given in reference 5.where a = p(Tj D+) = true positive rate, b = p(TjThe derivation of this equation is ...

Risk and information evaluation of prioritized genes for complex traits: Application to bipolar disorder

“…Estimating risk information for complex traits is useful, especially for measurement uncertainty and data incompleteness inherited in different study designs. Early studies have adopted the concept of information theory to quantify the risk for diagnosing coronary artery disease correctly using the electrocardiographic STsegment response to exercise [Diamond et al, 1981]. One advantage gains from using information theory is to estimate the average uncertainty (i.e.…”

CDC/AHA Workshop on Markers of Inflammation and Cardiovascular Disease

“…If a screening test is performed and a second test is applied afterward but the probability of disease did not change, then the result is a line of unity. 39,40 Such results occur rarely, and an "envelope" around the estimation line is the typical result. Positive results with a second test, shown by arrows pointing up, lead to an increased risk of disease, and the posterior probability is greater than the prior probability.…”

Théorie de l’information et diagnostic médical : une analyse coût-efficacité

“…Pour raison de simplicité de l'exposé, nous traitons ici le cas de tests dichotomiques (parfois appelés « catégoriques ») c-à-d dont le résultat est exprimé seulement par deux valeurs possibles (+ ou -). Le raisonnement est étendu aisément au cas de tests appelés « compartimentaux » dans la littérature médicale (résultats compris dans certains intervalles) (voir [14]) ou même de tests avec résultats continus.…”