Skip to content
All articles

Statistics

Sensitivity, Specificity, PPV and NPV: Screening and Diagnostic Tests Explained

Every test is judged against a gold standard using four measures. Sensitivity is the proportion of people with the disease who test positive; specificity is the proportion without it who test negative. These are fixed properties of the test.

Resora Research Team · 14 Nov 2026 · 5 min read

Every test is judged against a gold standard using four measures. Sensitivity is the proportion of people with the disease who test positive; specificity is the proportion without it who test negative. These are fixed properties of the test. Positive predictive value (PPV) is the chance that a positive result is truly disease, and negative predictive value (NPV) is the chance that a negative result is truly disease-free. These depend heavily on prevalence: the same test has a low PPV when the disease is rare, as in community screening. Likelihood ratios, the ROC curve and the area under it summarise accuracy across cut-offs. Choose a sensitive test to rule disease out, a specific test to rule it in, and always interpret results in light of how common the disease is.

The 2×2 table

Compare the test result against a gold standard (biopsy, culture, expert panel) in everyone studied:

Disease presentDisease absentTotal
Test positivea (true positive)b (false positive)a + b
Test negativec (false negative)d (true negative)c + d
Totala + cb + dn
MeasureFormulaQuestion it answers
Sensitivitya / (a + c)Of all who have the disease, how many does the test catch?
Specificityd / (b + d)Of all who do not, how many does the test correctly clear?
PPVa / (a + b)If the test is positive, what is the chance of disease?
NPVd / (c + d)If the test is negative, what is the chance of no disease?
Accuracy(a + d) / nHow often is the test right overall?

Mnemonics: SnNout: a highly Sensitive test, when Negative, rules out disease. SpPin: a highly Specific test, when Positive, rules in disease.

Worked example

A screening test is used in 1,000 people. The disease prevalence is 10%, so 100 have it. The test has a sensitivity of 90% and a specificity of 80%.

  • Of 100 diseased: 90 test positive (true positives), 10 test negative (false negatives).
  • Of 900 healthy: 720 test negative (true negatives), 180 test positive (false positives).
Disease presentDisease absentTotal
Test positive90180270
Test negative10720730
Total1009001,000
MeasureCalculationResult
Sensitivity90 / 10090%
Specificity720 / 90080%
PPV90 / 27033.3%
NPV720 / 73098.6%
Accuracy(90 + 720) / 1,00081%

What this tells us. A negative result is very reassuring (NPV 98.6%). But two out of three positives are false alarms (PPV only 33.3%), even though the test looks quite good on paper. Positives need a confirmatory test.

Why prevalence changes everything

Sensitivity and specificity stay the same wherever the test is used. PPV and NPV do not. Using the same test (sensitivity 90%, specificity 80%) in three settings:

Disease prevalencePPVNPVTypical setting
1%4.3%99.9%General community screening
10%33.3%98.6%High-risk group
30%65.9%94.9%Specialist clinic

At 1% prevalence, 198 of 207 positives are false. This is why a screening test that works well in a hospital can flood a community programme with false alarms, and why PPV from one study cannot be copied into another population.

Only sensitivity and specificity can be carried from one population to another. Recalculate PPV and NPV for your own prevalence.

Likelihood ratios

Likelihood ratios combine sensitivity and specificity into a single number that does not depend on prevalence.

LR+ = \frac{\text{Sensitivity}}{1 - \text{Specificity}} \qquad LR- = \frac{1 - \text{Sensitivity}}{\text{Specificity}}

For our test: LR+ = 0.90 / 0.20 = 4.5 and LR− = 0.10 / 0.80 = 0.125.

  • LR+ above 10 gives strong evidence to rule in disease; LR− below 0.1 gives strong evidence to rule it out.
  • Multiply the pre-test odds by the LR to get post-test odds. With 10% pre-test probability (odds 0.111), a positive result gives odds of 0.111 × 4.5 = 0.5, which is a post-test probability of 33.3%, matching the PPV above.

The ROC curve and choosing a cut-off

Many tests give a number (blood glucose, haemoglobin, a score), and the cut-off decides who counts as positive. Lowering the cut-off catches more cases (higher sensitivity) but creates more false positives (lower specificity). Raising it does the opposite.

The receiver operating characteristic (ROC) curve plots sensitivity (y-axis) against 1 − specificity (x-axis) across every possible cut-off.

  • The area under the curve (AUC) summarises overall accuracy: 0.5 is no better than a coin toss, 0.7–0.8 acceptable, 0.8–0.9 good, above 0.9 excellent, and 1.0 perfect.
  • Youden's index (J = sensitivity + specificity − 1) picks the cut-off that best balances both. For our test J = 0.9 + 0.8 − 1 = 0.7.
  • If missing a case is dangerous, favour sensitivity. If a false positive causes harm (invasive follow-up, anxiety), favour specificity.

Screening versus diagnostic tests

Screening testDiagnostic test
Done onApparently healthy peoplePeople with symptoms or a positive screen
AimSort into likely and unlikely diseaseEstablish the diagnosis
PriorityHigh sensitivity, low cost, simpleHigh specificity, accuracy
ResultNot a diagnosis; needs confirmationBasis for treatment

The classic Wilson and Jungner criteria ask whether the condition is an important health problem, has a recognisable early stage, has an accepted treatment, and whether a suitable, acceptable and affordable test exists.

Using more than one test

  • In series (sequential): positive only if both tests are positive. Specificity and PPV rise; sensitivity falls.
  • In parallel (simultaneous): positive if either is positive. Sensitivity and NPV rise; specificity falls.

Common mistakes

  1. Treating PPV from one study as fixed for every population.
  2. Quoting sensitivity and specificity without the gold standard or the confidence intervals.
  3. Judging a screening test by accuracy alone. At low prevalence, a useless test that calls everyone negative looks 99% "accurate".
  4. Confusing sensitivity (disease → test) with PPV (test → disease).
  5. Testing only people already suspected of disease, which inflates sensitivity (spectrum bias).
  6. Ignoring lead-time and length bias when judging screening programmes by survival.

Keywords

sensitivity and specificity, positive predictive value, negative predictive value, PPV NPV, screening test, diagnostic test accuracy, 2x2 table, likelihood ratio, ROC curve, AUC, Youden index, SnNout SpPin, prevalence and predictive value, screening vs diagnostic test, community medicine

References

  • Park K. Park's Textbook of Preventive and Social Medicine — chapter on screening for disease.
  • Altman DG, Bland JM. Diagnostic tests 1: sensitivity and specificity. BMJ. 1994;308:1552.
  • Altman DG, Bland JM. Diagnostic tests 2: predictive values. BMJ. 1994;309:102.
  • Bossuyt PM, et al. STARD 2015: an updated list of essential items for reporting diagnostic accuracy studies. BMJ. 2015;351:h5527.
  • Wilson JMG, Jungner G. Principles and Practice of Screening for Disease. WHO; 1968.