Likelihood Ratios in Diagnostic Testing

When a clinician orders a test, the question is not simply whether the result is positive or negative — it is by how much the result changes the probability that the patient has the condition. A test that comes back positive in a patient who was already almost certainly diseased adds little; the same positive result in a patient with borderline clinical suspicion may be decisive. The tool that captures this context-dependence is the likelihood ratio.

The likelihood ratio is grounded in Bayesian reasoning. Before the test is performed, there is already some probability — the pre-test probability — that the patient has the disease, based on prevalence, symptoms, signs, and clinical judgement. The test result updates that probability. How much it shifts depends on how strongly the result distinguishes diseased from non-diseased patients, which is what the likelihood ratio measures. The measures of risk, odds, and their ratios in the context of epidemiological and treatment studies are covered in the Risk, odds and effect measures chapter.

Sensitivity and Specificity

Likelihood ratios are derived from two fundamental properties of a diagnostic test, introduced in full in the Sensitivity, specificity and predictive values chapter. The sensitivity is the probability that the test is positive given that the disease is truly present: $\text{Se} = P(\text{test}+\mid\text{disease}+)$. The specificity is the probability that the test is negative given that the disease is truly absent: $\text{Sp} = P(\text{test}-\mid\text{disease}-)$.

Neither sensitivity nor specificity alone tells a clinician how to respond to an individual test result, because neither accounts for the prevalence of disease in the population being tested. The likelihood ratio combines both properties into a single number that can be applied directly to an individual patient's pre-test probability.

The Likelihood Ratio

The positive likelihood ratio (LR+) is the ratio of the probability of a positive test in diseased patients to the probability of a positive test in non-diseased patients:

The negative likelihood ratio (LR−) is the ratio of the probability of a negative test in diseased patients to the probability of a negative test in non-diseased patients:

The LR+ is always ≥ 1: a positive result raises the probability of disease. The LR− is always ≤ 1: a negative result lowers it. A test that provides no useful information has LR+ = LR− = 1. As a practical guide, an LR+ above 10 or an LR− below 0.1 represents strong diagnostic evidence, substantially shifting the post-test probability in most clinical settings. Values between 2 and 5 (or between 0.2 and 0.5) represent moderate evidence; values close to 1 add little diagnostic information.

Bayesian Updating

Likelihood ratios are applied by working in the odds domain. The pre-test probability is converted to pre-test odds, multiplied by the likelihood ratio to obtain post-test odds, and then converted back to a post-test probability:

This three-step calculation is a direct application of Bayes' theorem and is the mathematical foundation of how clinicians should update their diagnostic thinking when a test result arrives. The multiplication step — pre-test odds times LR — is the entire mechanism: a large LR+ multiplies the odds substantially upward; a small LR− multiplies them substantially downward.

Worked Example: Troponin in the Emergency Department

Suppose a high-sensitivity troponin assay has a sensitivity of 95% and a specificity of 90% for acute myocardial infarction. In patients presenting to the emergency department with chest pain, suppose the pre-test probability of MI is 20% — estimated from the clinical assessment before the result is known.

First, compute the likelihood ratios:

The pre-test probability is 0.20, so the pre-test odds are $0.20\,/\,0.80 = 0.25$.

If the troponin is positive:

If the troponin is negative:

A positive troponin triples the probability of MI from 20% to 70% — a result that should prompt urgent management. A negative troponin effectively rules out MI, reducing the probability to 1.4%. This is why likelihood ratios, which incorporate both sensitivity and specificity into a single number for each possible result, are more useful for individual clinical decisions than sensitivity and specificity considered in isolation.

Fagan's nomogram is a graphical alternative to the three-step calculation above: a straight line drawn from the pre-test probability through the likelihood ratio on a log-odds scale lands directly on the post-test probability, with no arithmetic required.

The Role of Pre-Test Probability

The magnitude of the shift produced by any likelihood ratio depends critically on the pre-test probability. In a low-risk population — say, young adults with atypical chest pain — where the pre-test probability of MI is 2%, the same positive troponin (LR+ = 9.5) raises the post-test probability to only about 16%, which is not nearly sufficient to diagnose MI or initiate thrombolytic therapy. In a high-risk population with a pre-test probability of 60%, the same positive result pushes the post-test probability above 93%. The test result cannot be interpreted without knowing where the patient starts.

This pre-test dependence is both the strength and the practical challenge of the likelihood ratio framework. It forces explicit acknowledgement of the prior probability — which is always present, whether or not it is made explicit — and prevents the blind application of test results regardless of context. Estimating pre-test probability well, through clinical scoring systems and knowledge of local disease prevalence, is therefore just as important as the test performance characteristics themselves.

Summary