Questions: Diagnostic Test Properties: Sensitivity and Specificity

Question 1

A blood test for disease X has 90% sensitivity and 85% specificity. The test is used in Clinic A (disease prevalence 5%) and Clinic B (disease prevalence 40%). Which of the following changes between the two settings?

• A. Sensitivity — it will be lower at Clinic A because fewer true positives are available to detect
• B. Specificity — it will be higher at Clinic A because there are more healthy people to correctly clear
• C. Both sensitivity and specificity — both metrics depend on the composition of the patient population
• D. Neither sensitivity nor specificity — both are fixed properties of the test; what changes is the predictive value of a positive or negative result

Answer: (D)

Sensitivity and specificity are calculated within each disease-status group: sensitivity = TP/(TP+FN) among those *with* disease; specificity = TN/(TN+FP) among those *without* disease. These calculations are independent of how many diseased vs. healthy people are in the tested population — they describe the test's behavior conditional on disease status. What changes dramatically with prevalence is the positive predictive value (PPV): in a low-prevalence setting, even a highly specific test will generate many false positives relative to true positives, because the pool of healthy people is so large. This is a critical distinction: sensitivity/specificity are test properties; predictive values are population-dependent.

Question 2

A physician is designing a screening protocol for a rapidly progressing infection where missing a case could be life-threatening. Which test property should she prioritize, and why?

• A. Specificity — she needs to avoid false positives to prevent unnecessary treatment
• B. Sensitivity — she needs to minimize false negatives so that no true cases are missed (SnOUT: high sensitivity rules out disease when negative)
• C. Neither — she should use the test with the highest overall accuracy regardless of the sensitivity/specificity tradeoff
• D. Specificity — because high specificity means fewer people need follow-up testing

Answer: (B)

When missing a case (false negative) is the primary danger — as with a rapidly progressing, life-threatening infection — the physician should maximize sensitivity. A highly sensitive test rarely misses true cases: a negative result from a highly sensitive test is highly reassuring (SnOUT mnemonic: high Sensitivity, when Negative, rules Out disease). The cost of low sensitivity here is missed diagnoses; the cost of low specificity is false alarms requiring follow-up. When a false negative is more dangerous than a false positive, optimize for sensitivity. Conversely, when a false positive triggers a dangerous or expensive intervention, prioritize specificity (SpIN: high Specificity, when Positive, rules In disease).

Question 3

Lowering the diagnostic cutoff for a continuous test (e.g., reducing the blood glucose threshold for diabetes diagnosis) will increase sensitivity and decrease specificity.

• T. True
• F. False

Answer: True

True. Lowering the threshold means more people test positive, including some who genuinely have the disease who would have been missed at the higher threshold (more true positives, fewer false negatives → higher sensitivity). But it also means more healthy people cross the threshold and test positive (more false positives → lower specificity). The reverse happens when the threshold is raised. This trade-off is inescapable and is visualized by the ROC curve: moving along the curve represents changing the cutoff, and every point on the curve represents a different sensitivity/specificity combination. There is no setting that simultaneously maximizes both.

Question 4

A test with 95% sensitivity correctly classifies 95% of most patients tested — both those with and without the disease.

• T. True
• F. False

Answer: False

False. Sensitivity = TP / (TP + FN) — it is calculated only among patients *who have the disease*. It says nothing about how the test performs on healthy patients. A test with 95% sensitivity correctly identifies 95% of sick patients but could have very poor specificity, misclassifying most healthy patients as positive. Confusing sensitivity with overall accuracy is the most common misinterpretation of this metric. Overall accuracy = (TP + TN) / total patients, which weights both sensitivity and specificity by the prevalence of disease in the tested population. Sensitivity and specificity describe performance within each disease-status group, independently of each other.

Question 5

Why can't you maximize both sensitivity and specificity simultaneously, and what determines the optimal tradeoff in a clinical setting?

Model answer: Sensitivity and specificity are in inherent tension because they are determined by the same diagnostic cutoff. Lowering the cutoff increases sensitivity (catches more true cases) but also increases false positives, lowering specificity. Raising the cutoff improves specificity (fewer false alarms) but causes more true cases to be missed, lowering sensitivity. The two metrics measure performance on different populations (sick vs. healthy), and any single threshold divides the measurement distribution in a way that affects both simultaneously but in opposite directions. The optimal tradeoff depends on the clinical context: specifically, the relative costs of false negatives (missed cases) and false positives (unnecessary follow-up, treatment, or patient anxiety). For dangerous conditions where missing a case is catastrophic, accept lower specificity to achieve high sensitivity. For conditions where false positives lead to harmful interventions, accept lower sensitivity to protect specificity.

The core insight is that sensitivity and specificity are not independently adjustable — they are two sides of the same cutoff decision. Understanding why they trade off requires seeing that any threshold creates a boundary between two overlapping distributions (test values in sick vs. healthy patients). Moving the boundary in one direction always helps one metric and hurts the other. The clinical judgment is not 'which is more important in general' but 'what are the consequences of each type of error in this specific situation?'