What does the likelihood ratio tell you?

A diagnostic likelihood ratio tells you how much a particular test result changes the probability that a person has the condition. The positive likelihood ratio shows how strongly a positive result raises that probability, and the negative likelihood ratio shows how strongly a negative result lowers it.

What does a likelihood ratio of 0.5 mean?

A likelihood ratio of 0.5 means the result is half as likely in someone with the condition as in someone without it. As a negative likelihood ratio it lowers the probability of disease only modestly, and a value between 0.5 and 1 is generally considered largely uninformative.

What is the formula for the likelihood ratio?

The positive likelihood ratio equals sensitivity divided by one minus specificity, the true-positive rate over the false-positive rate. The negative likelihood ratio equals one minus sensitivity divided by specificity, the false-negative rate over the true-negative rate.

What does a likelihood ratio of 1.0 mean?

A likelihood ratio of 1.0 means the test result is equally likely in people with and without the condition, so it provides no diagnostic information. Multiplying the pre-test odds by 1.0 leaves them unchanged, so the post-test probability equals the pre-test probability.

Likelihood Ratios in Diagnostic Tests

A likelihood ratio in diagnostic testing tells you how much a particular test result changes the probability that a person has the condition. There are two: the positive likelihood ratio is how many times more likely a positive result is in someone with the disease than in someone without it, and the negative likelihood ratio is how many times more likely a negative result is in someone with the disease than in someone without it. They combine sensitivity and specificity into a single figure per result, and they let you move from a pre-test probability to a post-test probability.

Why likelihood ratios beat sensitivity and specificity alone

Sensitivity and specificity describe a test, but they do not directly answer the clinical question, which is: given this result in this patient, how likely is the disease now? Predictive values answer that but shift with the prevalence of the condition, so a value measured in one setting does not transfer to another. The likelihood ratio is the useful middle ground. It is a property of the test, like sensitivity and specificity, so it is reasonably stable across settings, yet it plugs directly into a calculation that updates an individual’s probability. That makes it the natural currency for diagnostic test accuracy work and the reason it is preferred for synthesis, where the goal is a transportable summary of test performance rather than a single setting’s predictive values.

The formulae and a worked example

Both ratios are built from sensitivity and specificity:

Positive likelihood ratio equals sensitivity divided by (one minus specificity), that is the true-positive rate over the false-positive rate.
Negative likelihood ratio equals (one minus sensitivity) divided by specificity, that is the false-negative rate over the true-negative rate.

Take a test with a sensitivity of 0.90 and a specificity of 0.85. The positive likelihood ratio is 0.90 ÷ (1 − 0.85) = 0.90 ÷ 0.15 = 6.0, so a positive result is six times more likely in a person with the disease than in one without. The negative likelihood ratio is (1 − 0.90) ÷ 0.85 = 0.10 ÷ 0.85 ≈ 0.12, so a negative result is about one eighth as likely in a diseased person, which is reassuring. A strong test pushes the positive likelihood ratio well above 1 and the negative likelihood ratio well below 1. The structure here is a ratio of rates, the same family of measures as the ratio measures used for binary outcomes, and you can compute related ratios and intervals with our odds ratio and risk ratio calculator.

From pre-test to post-test probability with the Fagan nomogram

Likelihood ratios work through odds, not probabilities directly. The rule is: pre-test odds × likelihood ratio = post-test odds. Convert the pre-test probability to odds, multiply by the relevant likelihood ratio, then convert the post-test odds back to a probability. Suppose the pre-test probability of disease is 0.20, which is odds of 0.20 ÷ 0.80 = 0.25. A positive result with a positive likelihood ratio of 6.0 gives post-test odds of 0.25 × 6.0 = 1.5, which converts back to a probability of 1.5 ÷ (1 + 1.5) = 0.60. The positive result has raised the probability from 20 percent to 60 percent. The Fagan nomogram performs this update graphically: draw a straight line from the pre-test probability through the likelihood ratio, and it crosses the post-test probability axis at the answer, no arithmetic required. Because the update is multiplicative on the odds scale, the same likelihood ratio shifts probability more when the pre-test probability is moderate and less when it is already near 0 or 1.

Interpretation thresholds

Rough rules of thumb help judge whether a result meaningfully changes probability. For the positive likelihood ratio, values above about 10 produce large, often conclusive increases in the probability of disease; values from 5 to 10 give moderate increases; and values between 1 and 2 barely move the probability at all. For the negative likelihood ratio, values below about 0.1 produce large decreases, effectively helping to rule the condition out; values from 0.1 to 0.2 give moderate decreases; and values between 0.5 and 1 are largely uninformative. These bands are guides, not hard cut-offs, because what counts as a useful shift depends on the consequences of acting, but they capture the key intuition that a likelihood ratio close to 1 means the test result barely helps. Always read the ratio with its confidence interval as shown on a forest plot, since a wide interval can span clinically different categories.

Likelihood ratios in diagnostic test accuracy reviews

In a diagnostic test accuracy review, the data for each study are the counts of true positives, false positives, false negatives, and true negatives against a reference standard. Sensitivity and specificity are correlated and reported as a pair, because moving a test’s threshold trades one for the other, so they cannot be pooled as two independent numbers the way a single effect size is combined in a conventional quantitative synthesis. Instead, reviews use a bivariate or hierarchical summary receiver operating characteristic model that respects this pairing, and the pooled sensitivity and specificity are then turned into summary likelihood ratios. Because a likelihood ratio is built from both, it inherits any between-study variation in the underlying estimates, much of which comes from studies using different positivity thresholds. Risk of bias in these studies is appraised with the QUADAS-2 instrument, the diagnostic counterpart of the design-matched tools used for appraising intervention studies.

Common mistakes

Three errors recur. The first is multiplying the pre-test probability rather than the odds by the likelihood ratio, which is simply wrong; the conversion to odds and back is essential, and skipping it is the most common arithmetic slip. The second is treating a single likelihood ratio as if it applied at every threshold, when in fact the ratio shifts as the positivity cut-off moves, which is why threshold differences drive so much heterogeneity in reviews. The third is quoting the point estimate without its confidence interval, hiding genuine uncertainty about whether the test is informative. Reported with its interval, computed at a stated threshold, and applied through the odds, the likelihood ratio is one of the most practical statistics in diagnosis, because it answers the question a clinician actually has: how much should this result change my mind?