Short answer
Sensitivity asks how often a test is positive among people who have the target condition; specificity asks how often it is negative among people who do not. Both describe performance against a reference standard, but neither directly gives the chance that an individual positive or negative result is correct—that is a predictive-value question. 1 2
On this page
At a glance
Scroll to compare all columns →
| Attribute | Sensitivity | Specificity |
|---|---|---|
| Starting group | People with the condition | People without the condition |
| Correct result counted | Positive | Negative |
| Formula | TP / (TP + FN) | TN / (TN + FP) |
| Error excluded from numerator | False negative | False positive |
TP and TN mean true positives and true negatives; FP and FN mean false positives and false negatives. Multiply each formula by 100 for a percentage. 2
What each thing is
These are two conditional proportions: condition status comes first, and test result comes second. A reference standard establishes whether the target condition is present or absent; it can be a single method or a combination of methods, including clinical follow-up. The test being evaluated supplies the positive or negative result. 2
Key differences
Sensitivity captures detection within the condition-present group. Specificity captures negative classification within the condition-absent group. Their denominators are separate, so one percentage does not supply the other. A performance report needs both to describe these two sides of classification. 2
How to tell them apart
Check the denominator, not just the word “positive.” If the denominator contains everyone with the condition, the measure is sensitivity; if it contains everyone without it, the measure is specificity. This shortcut requires reliable condition-status labels: comparison with a non-reference test may measure agreement rather than accuracy. 2
Where they overlap
Both summarize classification performance for a defined target condition and population. Study values are estimates, not exact constants: another sample, or testing at another time, may yield different numbers. Confidence intervals describe sampling uncertainty; they do not turn either measure into an individual prediction. 2
Edge cases
The target’s definition matters. In cervical cancer screening, NCI explains that a positive Pap result followed by a high-grade intraepithelial lesion is not counted as a false positive when that precursor lesion is a screening target. “No invasive cancer” therefore does not automatically mean “target absent.” 1
Why the distinction exists
Separating the two condition-status groups keeps missed cases distinct from false alarms. Reversing the question creates different measures: positive predictive value starts with positive results, and negative predictive value starts with negative results. Those measures depend on prevalence as well as sensitivity and specificity. 1
Common misconceptions
High sensitivity does not mean a positive result has the same high probability of being correct. For fixed sensitivity and specificity, lower prevalence lowers positive predictive value and raises negative predictive value. Nor does agreement between two tests prove correctness: both can agree while misclassifying condition status. 1 2
Examples
Case 1—hypothetical numerical study: a stated reference standard identifies 100 people with condition X and 900 without it. The test detects 90 of the 100 and correctly calls 810 of the 900 negative: sensitivity and specificity are each 90%. Yet only 90 of 180 positive results are true positives, giving a 50% positive predictive value. 1 2
Case 2—hypothetical report: two tests frequently agree, but neither establishes condition status. Their agreement alone cannot establish sensitivity or specificity. 2