Sensitivity, Specificity and Predictive Values
Plain-language definitions of sensitivity, specificity and predictive values, why a positive result can still be wrong, and how to read these numbers for a real test.
Content last updated . This page has not been medically reviewed.
Four numbers that describe a test
Test accuracy is usually reported with four measures. Two describe the test, and two describe what a result means in a particular group of people. This page explains them and does not interpret anyone's result.
| Measure | Plain meaning | What it depends on |
|---|---|---|
| Sensitivity | Of people who truly have the condition, the share the test flags as positive | The test |
| Specificity | Of people who truly do not have it, the share the test calls negative | The test |
| Positive predictive value (PPV) | Of people with a positive result, the share who truly have the condition | The test and how common the condition is |
| Negative predictive value (NPV) | Of people with a negative result, the share who truly do not have it | The test and how common the condition is |
The FDA defines sensitivity as the proportion of people with the target condition in whom the test is positive, and specificity as the proportion without the condition in whom it is negative. [1] A 2021 review gives the same definitions and describes the predictive values. [2]
Why predictive values change
Sensitivity and specificity are generally considered stable for a given test, and they move in opposite directions: raising one tends to lower the other. Predictive values, by contrast, vary with the pre-test probability, for example the prevalence of the condition in the population tested. [2]Here is an invented example with made-up numbers, not a real test. Imagine a test with 90% sensitivity and 95% specificity used on 1,000 people, of whom 10 (1%) have the condition. It would correctly flag about 9 of the 10. Among the 990 without the condition, about 5% (roughly 50) would be flagged by mistake. So about 59 people test positive and only 9 of them truly have the condition, a PPV near 15%. If the same test were used in a group where half have the condition, the PPV would be far higher. Same test, different group, different meaning.
This is why a test used in people with symptoms and referred to a specialist can behave differently from the same test used on healthy people with no symptoms. One researcher argues that predictive values are the more informative numbers in screening settings, and that studies should report all four. [3]
A real-world illustration
In the FDA's description of a cleared blood test for Alzheimer's-related plaques, 499 samples from cognitively impaired adults were compared with amyloid PET or spinal fluid results. Of positive results, 91.7% had plaques on the reference test, and of negative results 97.3% did not. Fewer than 20% of results were indeterminate. [4] Those figures describe that study group, which was already showing symptoms. The FDA states the test is not intended as a screening or stand-alone diagnostic test. [4] Applying those percentages to healthy people would be a mistake.Where to go next
For what happens when positives turn out to be wrong, read false positives, multiple testing and incidental findings. For effect sizes in trials rather than tests, see effect sizes and confidence intervals.Frequently asked questions
Is a test with 95% sensitivity 95% accurate for me?
Not necessarily. Sensitivity describes people who truly have the condition. What a positive result means for you also depends on how common the condition is among people like you. [2]
Why can a good test still give many false alarms?
When a condition is rare, even a small false-positive rate produces many false alarms compared with true cases, so the positive predictive value can be low. [2]
Which numbers should a test report?
Researchers recommend giving all four measures and how they were derived. [3] The FDA recommends stating the reference standard and intended-use population. [1]
Can I compare sensitivity across different studies?
Only with care. Different populations, reference standards and cut-offs can change the numbers. [1]
More details: evidence, limits and cautions
The details behind this page: cautions, limits of the evidence and where sources disagree. Open any note to read it.
Limits and common traps
- The reference standard matters. If a new test is compared with something other than a true reference standard, the FDA says unbiased sensitivity and specificity cannot be calculated. [1]
- Estimates carry uncertainty and bias. They come from a sample, and the FDA notes they can be systematically too high or too low. [1]
- A cut-off is a choice. Moving the line for "positive" trades missed cases against false alarms. [2]
- "Accuracy" alone is vague. Ask which of the four measures is meant, in which group, and against which reference.
- The measures say nothing about benefit. A highly accurate test may still not lead to better health. See clinical validity and clinical utility.
Safety note
References
- U.S. Food and Drug Administration. Statistical Guidance on Reporting Results from Studies Evaluating Diagnostic Tests. Issued 2007. Read 2026-10-22.
- Monaghan TF, et al. Foundational Statistical Principles in Medical Research: Sensitivity, Specificity, Positive Predictive Value, and Negative Predictive Value. Medicina (Kaunas). 2021. PMID 34065637. Read 2026-10-22.
- Trevethan R. Sensitivity, Specificity, and Predictive Values: Foundations, Pliabilities, and Pitfalls in Research and Practice. Front Public Health. 2017. PMID 29209603. Read 2026-10-22.
- U.S. Food and Drug Administration. FDA Clears First Blood Test Used in Diagnosing Alzheimer's Disease. Press announcement, May 16, 2025. Read 2026-10-22.