Interpretation and Application of the Likelihood Ratio to Clinical Practice in Thoracic Oncology

Likelihood ratios (LRs) are a method to evaluate diagnostic test performance and assist in clinical decision making. While sensitivity and specificity are useful for binary tests, they cannot be directly applied to tests with >2 possible test results. LRs can be used for diagnostic tests with 2 or more possible test results and are also suitable for tests with continuous results. In this paper we review the concepts of LRs and how they relate to sensitivity and specificity. Practical examples from the pulmonary literature of how LRs are used to calculate posttest disease probabilities using Bayes' theorem are provided. These include examples when there are 3 or more categorical test results that have distinct interpretations (eg, cytology results from endobronchial ultrasound) as well as continuous test results (eg, computed tomography lymph node size and probability of metastasis). We also highlight some problems, pitfalls, and misunderstandings about LRs in clinical practice. We use the example of how the Nodify XL2 test incorrectly calculates and applies LRs, which may lead to falsely low estimates of the probability of cancer in some pulmonary nodules.