Negative Predictive Value

Negative predictive value (NPV) is the probability that a person with a negative test result truly does not have the condition. Like positive predictive value, it is read across a row of the 2x2 table rather than down a column, so it depends on the prevalence of the condition as well as on the test's intrinsic accuracy.

Definition

Negative predictive value is the conditional probability that disease is truly absent given a negative test result, calculated as the number of true negatives divided by the total number of negative results (true negatives plus false negatives).

Scope

This entry defines NPV as the proportion of true negatives among all negative results, explains its dependence on disease prevalence, contrasts it with sensitivity and specificity, and relates it to the Bayesian updating of pre-test to post-test probability. It is a methodological topic and does not advise on the use of any particular test.

Key concepts

• Probability of no disease given a negative result
• Dependence on prevalence (pre-test probability)
• True negatives versus false negatives
• Post-test probability
• Bayes' theorem in diagnosis
• Relationship to negative likelihood ratio

Mechanisms

NPV is computed across the negative-result row of the 2x2 table: of all subjects the test calls negative, it is the fraction whose disease status is truly negative. Because the absolute number of false negatives is generated from the diseased group, NPV falls as disease becomes more prevalent and rises as it becomes rarer, even when sensitivity and specificity are held fixed. NPV is therefore a joint product of the test's intrinsic accuracy and the pre-test probability of disease in the tested population. The relationship is formalised by Bayes' theorem, which updates the pre-test probability to a post-test probability using the test's likelihood ratios; NPV corresponds to one minus the post-test probability of disease following a negative result.

Clinical relevance

NPV expresses how much reassurance a negative result provides in a given setting and is therefore central to interpreting screening and diagnostic results in context. The concept supports critical appraisal of diagnostic evidence; it describes how test results are interpreted across populations and is not a basis for individual diagnostic or treatment decisions.

Epidemiology

In low-prevalence settings, NPV tends to be high simply because most people are disease-free, which can make a negative result look strongly reassuring even when a test's sensitivity is modest. Conversely, as prevalence rises, NPV declines, so like all predictive values it must be reported in relation to the relevant population rather than treated as a fixed test attribute.

History

The dependence of predictive values on prevalence was clarified as the diagnostic accuracy framework matured in the twentieth century, and the distinction between intrinsic test properties and population-dependent predictive performance was made accessible to clinicians through expository statistical writing in the 1990s.

Debates

Does a high negative predictive value mean a test is good at ruling out disease?

A high NPV may largely reflect low prevalence rather than strong test performance, so reassurance from a negative result should be judged against the underlying probability of disease and the test's likelihood ratios, not NPV alone.

Key figures

• Douglas Altman
• Martin Bland
• Jonathan Deeks
• David Grimes
• Kenneth Schulz

Seminal works

• altman-bland-1994b
• deeks-altman-2004
• grimes-schulz-2002-screening

Frequently asked questions

Why is negative predictive value often high for rare conditions?

When a condition is rare, most people genuinely do not have it, so most negative results are correct, pushing NPV high largely because of low prevalence rather than test quality.

Is negative predictive value a property of the test?

No. It depends on the prevalence of the condition in the tested population as well as on the test's sensitivity and specificity, so the same test yields different predictive values in different settings.