Likelihood ratios: Getting diagnostic testing into perspective

A. Halkin, J. Reichman, M. Schwaber, O. Paltiel, M. Brezis*

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

120 Scopus citations

Abstract

In modern medicine, sophisticated laboratory tests and imaging studies are often emphasized at the expense of history and physical examination, rather than complementing clinical assessment. Ancillary testing often fails to advance the diagnostic process, and increases patient risk and the expense of medical care. The relative value of clinical evaluation and technological methods is rarely considered, and the power of the clinical evaluation is therefore underestimated. The likelihood ratio (LR) is a semi-quantitative measure of the performance of diagnostic tests which indicates how much a diagnostic procedure modifies the probability of disease, and is calculated from the sensitivity and specificity of the test (or directly from the change in probability associated with the test result). We review the performance of frequently-used tests by their LRs, and compare them to the power of clinical assessment, with clinical cases to illustrate the application of LRs in the diagnostic process. The discriminative power of clinical assessment and ancillary tests is often similar, and the combination of the two greatly increases accuracy in the diagnostic process. Clinical assessment is indeed frequently more informative than current technical modalities. LRs assist in putting the value of testing in proper perspective. Practice in evaluating pre-test probabilities of disease and in the application of LRs should be enhanced in medical training.

Original languageAmerican English
Pages (from-to)247-258
Number of pages12
JournalQJM: An International Journal of Medicine
Volume91
Issue number4
DOIs
StatePublished - 1998
Externally publishedYes

Fingerprint

Dive into the research topics of 'Likelihood ratios: Getting diagnostic testing into perspective'. Together they form a unique fingerprint.

Cite this