Diagnostic Accuracy & Test Interpretation
Use a two-by-two table to distinguish test characteristics from the probability a patient has disease.
Quick revision ↓ Medical Dictionary
01Start with the table
Compare the index test with an appropriate reference standard in a clinically relevant sample. Classify results as true positive, false positive, false negative and true negative. Sensitivity is the proportion of people with disease testing positive; specificity is the proportion without disease testing negative. State which threshold defines positivity. A reference standard can itself be imperfect, so agreement is not automatically equivalent to truth.
02Predictive values and prevalence
Positive predictive value is the proportion of positive results that truly have disease; negative predictive value is the proportion of negative results without disease. These depend strongly on disease frequency in the tested population. Even a reasonably accurate test can produce many false positives when disease is uncommon. Use the patient's pretest context rather than importing a predictive value from a very different population.
03Likelihood ratios and updating
The positive likelihood ratio is sensitivity divided by one minus specificity; the negative likelihood ratio is one minus sensitivity divided by specificity. Convert pretest probability to odds, multiply by the relevant likelihood ratio, then convert odds back to probability. A likelihood ratio of one leaves odds unchanged. Test characteristics may vary across settings and disease spectra, so transportability should still be assessed.
04Worked two-by-two example
Among 100 people, 20 have disease. A test detects 16 of those 20 and is negative in 72 of the 80 without disease. Thus TP=16, FN=4, TN=72 and FP=8. Sensitivity is 80%, specificity 90%, PPV 16/24 or 66.7%, and NPV 72/76 or 94.7%. LR+ is 8 and LR− about 0.22. Keep each denominator visible when interpreting these different questions.
05Appraise the accuracy study
Look for a representative clinical spectrum, prespecified thresholds, independent interpretation and appropriate reference testing regardless of index result. Verifying only positive tests can bias estimates. Avoid an artificial comparison of obvious severe cases against exceptionally healthy controls when clinical use involves uncertain presentations. A ROC curve shows threshold trade-offs; a good area under that curve does not establish improved patient outcomes from using the test.
Diagnostic two-by-two table
| Index test | Disease present | Disease absent |
|---|---|---|
| Positive | TP = 16 | FP = 8 |
| Negative | FN = 4 | TN = 72 |
Sensitivity = 16/20 • Specificity = 72/80 • PPV = 16/24 • NPV = 72/76.
High-Yield Revision
- Sensitivity and specificity condition on disease status.
- Predictive values condition on the test result.
- Post-test odds = pretest odds × likelihood ratio.
Common Pitfalls
- Do not confuse high sensitivity with a high positive predictive value.
- A test with good accuracy may still lack evidence of clinical benefit.
References & Further Reading
Use the full methods and relevant chapters when appraising a paper. These sources support further study; the examples and questions above are original teaching exercises.
- Oxford CEBM — Critical appraisal tools
- Cochrane Handbook — Methods for evidence synthesis
- CDC — Analysing and interpreting epidemiological data
- ASA — Statement on p-values and statistical significance
- Oxford CEBM — Likelihood ratios
- Cochrane — GRADE and summary of findings
- WMA — Declaration of Helsinki, 2024 revision
- EQUATOR — Reporting guideline library
- NIH OBSSR — Mixed-methods research resources
- Institute for Healthcare Improvement — Model for Improvement