Diagnostic tests and statistics
Sensitivity and specificity describe performance within disease groups. Predictive values answer the patient-facing question after a result and depend on the population being tested.
Quick revision summary
Worked example: 1,000 people
Assume disease prevalence 10%, sensitivity 90% and specificity 90%. There are 100 affected people: 90 test positive and 10 negative.
Work from a two-by-two table
Place disease status in one dimension and test result in the other. True positives and false negatives make up the affected group; true negatives and false positives make up the unaffected group.
Keep probability, odds and uncertainty distinct
Probability is the proportion with the event; odds compare the event with its absence. A probability of 20% corresponds to odds of 0.2 divided by 0.8, or 0.25.
Common mistake
Sensitivity describes affected people; it is not the probability of disease after a positive result.
Diagnostic accuracy, prevalence and likelihood ratios
Sensitivity and specificity describe performance within disease groups. Predictive values answer the patient-facing question after a result and depend on the population being tested.
Keep the denominators straight
| Measure | Calculation | Meaning |
|---|---|---|
| Sensitivity | TP / (TP + FN) | Proportion of affected people testing positive. |
| Specificity | TN / (TN + FP) | Proportion of unaffected people testing negative. |
| Positive predictive value | TP / (TP + FP) | Proportion of positive results that are true disease. |
| Negative predictive value | TN / (TN + FN) | Proportion of negative results that are truly unaffected. |
| Positive likelihood ratio | Sensitivity / (1 − specificity) | How much a positive result shifts the odds. |
| Negative likelihood ratio | (1 − sensitivity) / specificity | How much a negative result shifts the odds. |
Worked example: 1,000 people
Assume disease prevalence 10%, sensitivity 90% and specificity 90%. There are 100 affected people: 90 test positive and 10 negative. There are 900 unaffected people: 810 test negative and 90 positive. Even with apparently good sensitivity and specificity, only half of the positive results are true disease.
| Result | Disease present | Disease absent |
|---|---|---|
| Positive | 90 true positives | 90 false positives |
| Negative | 10 false negatives | 810 true negatives |
Apply the result to a clinical question
In this example PPV = 90 / 180 = 50%, while NPV = 810 / 820 ≈ 98.8%. The positive likelihood ratio is 0.9 / 0.1 = 9. Pre-test odds at 10% are 0.1 / 0.9; multiplying by 9 gives post-test odds of 1, equivalent to a 50% probability. Odds and probability are different quantities.
At fixed sensitivity and specificity, lower prevalence usually reduces PPV and increases NPV. The clinical consequences matter: missing an airway threat or cancer is different from investigating a benign low-risk finding. A negative test with limited sensitivity should not override a strongly suspicious examination. Spectrum, verification and selection bias can also limit whether a published accuracy estimate applies to your patient.
Work from a two-by-two table
Place disease status in one dimension and test result in the other. True positives and false negatives make up the affected group; true negatives and false positives make up the unaffected group. Sensitivity and specificity use those disease-status groups as denominators. Predictive values instead start from the people with a particular test result.
Changing the threshold often trades sensitivity against specificity. Whether false negatives or false positives are more consequential depends on the condition, available confirmation and burden of follow-up. A screening pathway therefore needs more than a promising test statistic: it needs a suitable population, confirmatory assessment and a management pathway.
Keep probability, odds and uncertainty distinct
Probability is the proportion with the event; odds compare the event with its absence. A probability of 20% corresponds to odds of 0.2 divided by 0.8, or 0.25. Likelihood ratios update odds, not probability directly. After multiplying pre-test odds by the appropriate likelihood ratio, convert back using odds divided by one plus odds.
Published accuracy estimates have uncertainty and may not transfer to a different patient group. Disease severity, test technique and how the reference standard is applied can change observed performance. A negative result should not close a strongly suspicious clinical problem without considering the test’s limitations. Use the result to inform a decision, not to replace the history and examination.
A distinction to remember
Sensitivity describes affected people; it is not the probability of disease after a positive result.
In the worked example, why are there as many false positives as true positives?
Reveal the reasoning
The unaffected group is much larger: 10% of 900 gives 90 false positives, while 90% of 100 gives 90 true positives. Prevalence changes the meaning of a positive result.
Guidance to read alongside this chapter
Test your recall
Why can a positive test be less convincing in a low-prevalence group?
Show answer
False positives make up a greater share of positive results.
