Negative Predictive Value (NPV)

Negative predictive value (NPV) is the proportion of all negative test results that are true negatives — the probability, after a negative result, that the condition being tested for is genuinely absent. It is calculated within the test-negative row of a diagnostic accuracy study’s 2×2 table, the mirror of positive predictive value (PPV), which is calculated within the test-positive row of the same table. See Sensitivity vs. Specificity for the full build-up of the 2×2 table this article assumes.

Formula

NPV = TN / (TN + FN) — of everyone who tested negative, what fraction is truly free of the condition.

Why NPV is the metric that matters for a rule-out test

PPV and NPV answer different clinical questions, and a test optimized for one is not automatically good at the other. A common teaching mnemonic, SnNout (a test with high Sensitivity, when Negative, rules out the condition), captures why NPV specifically depends on a test’s sensitivity: a highly sensitive test misses very few true cases, so a negative result on that test is strong evidence of true absence — a high NPV. A test can therefore be built or chosen specifically to maximize NPV for screening purposes (where the goal is confidently excluding a condition in a negative result) even if its PPV, and therefore its usefulness for confirming a diagnosis on a positive result, is comparatively weak.

Worked example: NPV moves opposite to PPV as prevalence changes

Take a test with 95% sensitivity and 90% specificity, applied to 1,000 people.

High-prevalence population, 30% prevalence (300 with the condition, 700 without): TP = 285, FN = 15, FP = 70, TN = 630.

• NPV = 630 / (630 + 15) ≈ 97.7%
• PPV, for comparison = 285 / (285 + 70) ≈ 80.3%

Low-prevalence population, 2% prevalence (20 with the condition, 980 without), same test, same sensitivity and specificity: TP = 19, FN = 1, FP = 98, TN = 882.

• NPV = 882 / (882 + 1) ≈ 99.9%
• PPV, for comparison = 19 / (19 + 98) ≈ 16.2%

NPV moved the opposite direction from what PPV does as prevalence fell: NPV rose from 97.7% toward 99.9%, while PPV collapsed from 80.3% to 16.2%. This is the same underlying mechanism as the PPV prevalence-dependence, running in reverse — as a condition becomes rarer, there are proportionally more true negatives and fewer false negatives relative to the shrinking true-positive pool, so a negative result becomes an even more reliable rule-out, while a positive result becomes a less reliable rule-in.

The reassurance trap: NPV can look excellent in a low-prevalence population by default

Because NPV rises automatically as prevalence falls, a very high NPV in a low-prevalence screening population is not, by itself, strong evidence that a test discriminates well — a nearly useless test (one that predicts ‘negative’ for almost everyone) can still post a high NPV purely because so few people in that population actually have the condition. The number that isolates a test’s genuine discriminating power, independent of prevalence, is sensitivity, not NPV; NPV should always be read alongside the prevalence it was calculated at, and ideally alongside sensitivity/specificity, rather than cited on its own as proof a negative result can be trusted.

NPV vs. PPV

	Negative Predictive Value (NPV)	Positive Predictive Value (PPV)
Answers	Given a negative result, how likely is the condition truly absent?	Given a positive result, how likely is the condition truly present?
Formula	TN / (TN + FN)	TP / (TP + FP)
Table row used	Test-negative row	Test-positive row
Effect of lower prevalence	Rises	Falls
Most informative when	Ruling a condition out	Ruling a condition in
Driven mainly by	The test’s sensitivity	The test’s specificity

Reporting NPV

Diagnostic accuracy studies reporting NPV should follow the STARD reporting guideline, which requires stating the prevalence (or pretest probability) the predictive value was calculated at, since an NPV figure reported without that context cannot be validly transferred to a population with a different prevalence — the same caveat that applies to PPV, in the opposite direction.

Frequently asked questions

What’s the difference between NPV and specificity?

Specificity is TN / (TN + FP) — calculated within people who are truly free of the condition (a column of the 2×2 table) and used to describe the test itself. NPV is TN / (TN + FN) — calculated within people who tested negative (a row of the table) and used to interpret a specific negative result. Specificity does not change with prevalence; NPV does.

Why is NPV usually high in screening programs?

Screening programs typically test a low-prevalence general population, and NPV rises as prevalence falls for a fixed sensitivity and specificity — so a screening test’s NPV is often reassuringly high largely because the condition is rare in the screened population, not necessarily because the test itself discriminates especially well. Sensitivity, not NPV, is the number that reflects the test’s intrinsic ability to catch true cases.

Can a test have high NPV and low PPV at the same time?

Yes, and this is common in low-prevalence screening settings — the worked example above shows a single test scoring 99.9% NPV and only about 16% PPV in the same 2% -prevalence population. High NPV and low PPV together mean a negative result is highly trustworthy while a positive result needs confirmatory follow-up testing before acting on it.

Does a higher NPV always mean a better test?

No. NPV is prevalence-dependent, so comparing NPV across two different populations, or across two tests evaluated in populations with different prevalence, does not isolate which test is actually better. Comparing sensitivity and specificity, which are in principle prevalence-independent, is the valid way to compare two tests’ intrinsic performance.