Likelihood Ratios

Concept and Formula

Likelihood ratios are derived from applying Bayes' theorem to diagnostic testing. There are two key forms: the positive likelihood ratio (LR+) and the negative likelihood ratio (LR-). The formulas are: LR+ = sensitivity / (1 - specificity) and LR- = (1 - sensitivity) / specificity. LR+ tells you how many times more likely a positive test result is among people with the condition compared to those without it. LR- makes the same comparison for a negative test result. An LR equal to 1 means the test provides no diagnostic information at all.

From Pre-Test to Post-Test Probability

To use a likelihood ratio, first convert the pre-test probability to pre-test odds: pre-test odds = pre-test probability / (1 - pre-test probability). Then multiply the pre-test odds by the LR to obtain the post-test odds. Finally, convert back to probability: post-test probability = post-test odds / (1 + post-test odds). This calculation allows a clinician or researcher to formally combine prior knowledge with the new test result. The Fagan nomogram is a classical graphical tool that makes this conversion visually intuitive without requiring manual arithmetic.

Common Misuses and Misconceptions

The most common error is treating a likelihood ratio directly as a probability: an LR+ of 5 does not mean the patient has an 80 percent chance of having the condition. The LR only becomes meaningful when combined with the pre-test probability. A second misconception is overlooking that LR values are prevalence-independent: because sensitivity and specificity remain constant across populations, LRs travel well between settings, unlike positive and negative predictive values which shift with prevalence. A third error is reporting sensitivity and specificity separately without computing the LR, thereby missing the combined diagnostic strength that a single ratio communicates.

Why It Matters and How to Report It

Likelihood ratios are among the most robust ways to compare test accuracy in clinical research and systematic reviews. When reporting, both LR+ and LR- should be presented alongside their 95 percent confidence intervals. A widely cited rule of thumb holds that an LR+ of 10 or more provides strong evidence for a condition, while an LR- of 0.1 or less largely rules it out; these thresholds should nonetheless be interpreted in context. The STARD (Standards for Reporting of Diagnostic Accuracy Studies) guidelines recommend that LR values be transparently reported together with study methods to enable readers to apply the results to their own clinical or research settings.

Sources