Other chapters of this book have made the case for the usefulness of likelihood ratios (LRs) in the process of diagnosis (see Chapter 16, The Process of Diagnosis, and Chapter 18, Diagnostic Tests). In this chapter, we present some examples of LRs, along with their associated 95% confidence intervals (CIs), for many diagnostic tests. For each test, we describe the population to whom the test was applied and the range of prevalence (pretest probability) found for each target condition (disease). Our choice of conditions has been idiosyncratic and represents the interests of the authors. We restricted ourselves to tests in current use and so do not offer a technical description of the tests. The authors conducted all searches and summaries, without duplicate adjudication of eligibility or data extraction.
Methods for Summarizing the Information on Likelihood Ratios
For each test and target condition under consideration, we included studies that met each of the following criteria:
The study authors presented LRs or sufficient data to allow their calculation.
The investigators compared the test with a reference standard (criterion standard or gold standard) that was defined in advance and that met the following criteria: (1) at the time of the study it was in wide use and no better standard was available; (2) when the decision to apply the criterion standard was unrelated to the results of the test, it was applied to at least 50% of eligible patients; and (3) when the decision to apply the criterion standard may have been influenced by test results, it was applied to 90% of eligible patients or it was blindly applied.
The investigators enrolled patients similar to those treated in clinical practice for whom the test might be reasonably applied.
Publications were in English or Spanish.
We excluded studies that met the following criteria:
The study was concerned with predicting long-term outcomes.
The study evaluated diagnostic models, including multiple tests such as decision trees, diagnostic algorithms, neural networks, or computer-based pattern recognition systems.
Our original search included Best Evidence (1991–2000) and MEDLINE (1966–2000). In addition, we hand-searched the JAMA series entitled The Rational Clinical Examination (1992–2000) and references from a diagnostic textbook.1 We also reviewed the citations of articles we found for additional potentially eligible studies. Examples have been updated to 2013, and new examples have been added.
For every pair of target condition and test, we searched the databases with the following search strategy template, using both Medical Subject Headings (MeSH) and text words (Figure 19.2-1). An example of the typical search strategy is shown in For every pair of target condition and test, we searched the databases with the following search strategy template, using both Medical Subject Headings (MeSH) and text words (Figure 19.2-2.