Diagnostic Testing and Clinical Reasoning
What this domain tests
This domain tests quantitative reasoning about tests: what sensitivity, specificity, predictive values, and likelihood ratios mean, how prevalence changes interpretation, and when a test is worth ordering at all.
Core rules
- Sensitivity = TP/(TP+FN); specificity = TN/(TN+FP). Both are properties of the test and the population studied, independent of local prevalence.
- Positive predictive value = TP/(TP+FP) and negative predictive value = TN/(TN+FN). Both depend heavily on prevalence, so a test that performs well in a referral center can mislead in a screening population.
- Likelihood ratios combine both: LR+ = sensitivity/(1 − specificity); LR− = (1 − sensitivity)/specificity. Multiply the pretest odds by the LR to get post-test odds. LR+ above 10 or LR− below 0.1 shifts management meaningfully; values near 1 do not.
- Highly sensitive tests are useful to rule out (a negative result is informative); highly specific tests are useful to rule in.
- Screening tests are applied to asymptomatic people and must have proven benefit, an accepted treatment for detected disease, and acceptable harms; diagnosis in symptomatic patients is a different question.
- Reference ranges are statistical, not thresholds for treatment; incidental findings and cascades of repeat imaging carry real cost.
Worked example
A test has 90% sensitivity and 95% specificity; prevalence is 1%. In 10,000 people: 100 have disease, giving 90 true positives and 10 false negatives; 9,900 do not, giving 495 false positives and 9,405 true negatives. PPV = 90/585 ≈ 15.4%. So a positive result means roughly a one-in-six chance of disease, and 495 worried healthy people are generated. LR+ = 0.90/0.05 = 18, so a positive result does raise probability substantially, but from a very low starting point. The correct next step is a confirmatory test, not immediate treatment.
Traps that cost marks
- Quoting sensitivity when the question asks what a positive result means for this patient (that is PPV).
- Forgetting prevalence in predictive-value calculations.
- Substituting a likelihood ratio for a probability, or adding rather than multiplying odds.
- Treating a screening test as diagnostic.
- Ignoring overdiagnosis and lead-time bias when survival appears to improve.