What Are Sensitivity and Specificity?
Sensitivity is the proportion of people who have the disease that the test correctly identifies as positive. Specificity is the proportion of people who do not have the disease that the test correctly identifies as negative. Together, they measure how accurately a diagnostic test performs across both the sick and healthy populations. A perfect test would have both at 100%; in practice, raising one often lowers the other.
Imagine a screening program for tuberculosis in a hospital. Some patients genuinely have TB; others do not. A chest X-ray is used to screen everyone. The key questions are: among patients who truly have TB, what fraction does the X-ray flag as positive? That fraction is sensitivity. Among patients who genuinely do not have TB, what fraction does the X-ray correctly flag as negative? That fraction is specificity.
These two measures capture different types of test accuracy. Sensitivity is about not missing the disease. Specificity is about not raising false alarms. Clinical decisions about which to prioritise depend on the consequences of each type of error, not on a single universal rule.
Understanding the Confusion Matrix
Every diagnostic test produces four possible outcomes. A patient either has the disease or does not, and the test either flags them as positive or negative. These four combinations form the confusion matrix, the foundation from which every diagnostic accuracy metric is calculated.
Each cell has a precise meaning that carries clinical weight:
- True Positive (TP): The patient has the disease and the test correctly identifies them as positive. The test worked as intended.
- False Positive (FP): The patient does not have the disease but the test incorrectly flags them as positive. Also called a Type I error. Can lead to unnecessary anxiety, further invasive testing, or harmful treatment.
- False Negative (FN): The patient has the disease but the test incorrectly reports them as negative. Also called a Type II error. The most dangerous outcome when missing a diagnosis leads to delayed or absent treatment.
- True Negative (TN): The patient does not have the disease and the test correctly identifies them as negative. The test correctly ruled out disease.
Sensitivity is calculated by reading down the left column (among disease-positive people). Specificity is calculated by reading down the right column (among disease-negative people). Predictive values are calculated by reading across the rows. The column you read determines whether prevalence affects the result — column-based measures do not change with prevalence; row-based ones do.
Sensitivity Explained
Formula and Calculation
TP = True Positives (sick people who test positive)
FN = False Negatives (sick people who test negative)
TP + FN = Total number of people who actually have the disease
The denominator is the total count of disease-positive individuals, regardless of what the test said. Sensitivity asks: among every person who truly has the disease, what proportion did the test correctly catch?
Interpretation
A sensitivity of 0.95 means the test correctly identifies 95 out of every 100 true disease cases. The remaining 5 are false negatives — missed diagnoses. A sensitivity of 0.70 means 30 out of every 100 cases are missed, which would be unacceptable for a cancer screening test but potentially tolerable for a low-stakes initial triage.
The mnemonic SnNout helps with clinical application: a test with high Snsitivity, when it returns a Negative result, is good at ruling out disease. A highly sensitive test that comes back negative provides strong evidence the patient does not have the condition.
When High Sensitivity Matters
High sensitivity is the priority when missing a diagnosis is more dangerous than making a false alarm. Screening programs that cast a wide net over a population — checking for HIV in blood donors, detecting COVID-19 in high-risk contacts, or ruling out meningitis in a child with a fever — need high sensitivity above all else. Missing a true case in these contexts has severe consequences: delayed treatment, ongoing transmission, or preventable death.
Increasing sensitivity typically reduces specificity. A test set to catch every possible case will also flag many healthy people as positive. Those false positives then face unnecessary follow-up testing, potential harm from further procedures, anxiety, and wasted healthcare resources. This trade-off cannot be eliminated — only managed through thoughtful threshold selection and two-step diagnostic pathways.
Specificity Explained
Formula and Calculation
TN = True Negatives (healthy people who test negative)
FP = False Positives (healthy people who test positive)
TN + FP = Total number of people who actually do not have the disease
Interpretation
A specificity of 0.90 means the test correctly identifies 90 out of every 100 healthy people as negative. The remaining 10 receive false positive results. In a population where millions of people are screened, even 90% specificity generates enormous numbers of false alarms.
The mnemonic SpPin applies here: a test with high Specificity, when it returns a Positive result, is good at ruling in disease. A highly specific test that comes back positive provides strong evidence the patient genuinely has the condition.
When High Specificity Matters
High specificity is the priority when a false positive carries serious consequences. Confirmatory testing for conditions where positive results trigger invasive or irreversible interventions — such as HIV confirmation before starting lifelong antiretroviral therapy, biopsy decisions in cancer diagnosis, or surgical referrals — must minimise false positives. A wrong positive here causes real harm: unnecessary surgery, stigma, medication side effects, or treatment toxicity in a patient who was never sick.
Sensitivity vs. Specificity: Complete Comparison
Sensitivity measures how well a test detects true disease (true positive rate). Specificity measures how well a test rules out disease in healthy people (true negative rate). Sensitivity = TP / (TP + FN). Specificity = TN / (TN + FP). High sensitivity means few missed diagnoses. High specificity means few false alarms. The two cannot both be maximised simultaneously at a single threshold; clinicians choose based on whether missing disease or falsely diagnosing it causes greater harm.
| Feature | Sensitivity | Specificity |
|---|---|---|
| Definition | Proportion of truly sick people the test correctly flags as positive | Proportion of truly healthy people the test correctly flags as negative |
| Clinical question | "Of all patients with the disease, how many does this test detect?" | "Of all patients without the disease, how many does this test correctly clear?" |
| Formula | TP / (TP + FN) | TN / (TN + FP) |
| Also called | True Positive Rate, Recall | True Negative Rate |
| Measures performance among | Disease-positive patients only | Disease-negative patients only |
| Error it minimises | False Negatives (missed cases) | False Positives (false alarms) |
| Clinical mnemonic | SnNout: high Sensitivity, Negative result rules Out | SpPin: high Specificity, Positive result rules In |
| Priority when | Missing diagnosis is dangerous (HIV screening, cancer detection, meningitis) | False positive triggers harmful treatment (surgery, toxic therapy, stigma) |
| Best study phase | Initial screening of a broad population | Confirmatory testing after a positive screen |
| Affected by prevalence? | No — calculated within disease-positive group only | No — calculated within disease-negative group only |
| Trade-off | Raising sensitivity typically lowers specificity | Raising specificity typically lowers sensitivity |
| Threshold effect | Lowering the cutoff increases sensitivity, decreases specificity | Raising the cutoff increases specificity, decreases sensitivity |
Related Diagnostic Accuracy Measures
Positive and Negative Predictive Values
Sensitivity and specificity describe what a test does to known groups. Predictive values answer the question a clinician faces in practice: given this result, what does it actually mean for the patient in front of me?
TP = True Positives
FP = False Positives
PPV rises as disease prevalence rises and as specificity rises
TN = True Negatives
FN = False Negatives
NPV rises as disease prevalence falls and as sensitivity rises
Sensitivity and specificity are calculated within each disease group separately — they are properties of the test. PPV and NPV mix both groups in their denominators, so the ratio of sick to healthy people (prevalence) directly shifts their values. In a low-prevalence population, even a highly specific test generates many false positives relative to true positives, making PPV poor. Clinicians testing high-risk populations get better PPV from the same test than when screening low-risk populations.
Accuracy
Overall accuracy is the proportion of all test results that are correct. It adds the correct answers from both disease groups together.
Likelihood Ratios
Likelihood ratios are among the most clinically useful measures because they directly update pre-test probability into post-test probability using Bayes' theorem. They do not change with disease prevalence in the tested population, making them portable across clinical settings. You can learn more about this foundational concept in the Bayes' Theorem guide.
LR+ = how much more likely a positive result is in sick vs. healthy people
LR− = how much more likely a negative result is in sick vs. healthy people
LR+ > 10 or LR− < 0.1 generally provides strong diagnostic evidence
An LR+ of 15 means a positive result is 15 times more likely in a person with the disease than in a person without it. Combined with the pre-test probability of disease in a given patient, the clinician can calculate a post-test probability that guides the management decision.
ROC Curve and AUC
Most quantitative tests (blood glucose, PSA, troponin) produce a continuous numerical result rather than a simple positive/negative answer. The clinician must choose a threshold: values above it count as positive, below as negative. Different thresholds produce different sensitivity-specificity pairs.
A Receiver Operating Characteristic (ROC) curve plots every possible threshold as a point: sensitivity (y-axis) against 1 minus specificity (x-axis), which equals the false positive rate. Connecting all points traces the full curve. A perfect test curves all the way to the top-left corner — 100% sensitivity with 100% specificity. A worthless test follows the diagonal, performing no better than chance.
| AUC Value | Interpretation | Clinical Context |
|---|---|---|
| 1.0 | Perfect discrimination | Theoretical only; no real test achieves this |
| 0.90 – 0.99 | Excellent | High-quality confirmatory tests; most molecular assays at peak performance |
| 0.80 – 0.89 | Good | Acceptable for clinical use; many well-validated biomarkers |
| 0.70 – 0.79 | Acceptable | Useful but imperfect; often requires clinical context to interpret |
| 0.60 – 0.69 | Poor | Limited standalone utility; may contribute to a multi-test panel |
| 0.50 – 0.59 | No useful discrimination | Not better than flipping a coin; do not use as a diagnostic basis |
| Below 0.50 | Worse than chance | Possible reversed scale; check how the test result is coded |
The Area Under the ROC Curve (AUC, also called the c-statistic) summarises the entire curve in a single number. An AUC of 0.85, for example, means there is an 85% probability that the test will rank a randomly chosen sick patient higher than a randomly chosen healthy patient.
The point on the ROC curve closest to the top-left corner is often selected as the optimal threshold, but the true optimum depends on relative costs. If false negatives are 5 times more harmful than false positives (as in cancer screening), the optimal point shifts upward along the curve, accepting more false positives to prevent more missed cases. Statistical software can calculate the threshold that minimises a weighted error function reflecting these clinical costs.
Disease Prevalence and Test Performance
One of the most counterintuitive facts in diagnostic testing is that the same test, applied to two populations with different disease prevalence, gives very different predictive values — even though its sensitivity and specificity stay exactly the same. Understanding this protects against a common and dangerous error in clinical reasoning. For more on how disease frequency is measured in populations, see the companion guide to incidence vs. prevalence.
Consider a COVID-19 lateral flow test with 95% sensitivity and 97% specificity. Applied to a high-risk hospital emergency department where 20% of patients have COVID-19, the PPV is roughly 88% — a positive result is highly likely to be genuine. The same test applied to routine airport screening where prevalence is 1% yields a PPV closer to 24% — three out of every four positive results are false positives, despite the same test characteristics.
| Prevalence Scenario | Sensitivity | Specificity | PPV | NPV |
|---|---|---|---|---|
| High prevalence (50%) — ICU setting | 95% | 97% | 97% | 95% |
| Moderate prevalence (20%) — Emergency dept | 95% | 97% | 88% | 99% |
| Low prevalence (5%) — General clinic | 95% | 97% | 63% | 99.7% |
| Very low prevalence (1%) — Airport screening | 95% | 97% | 24% | 99.9% |
This table makes a critical point: the same 95% sensitive, 97% specific test produces a PPV that varies from 24% to 97% depending on who is being tested. This is not a flaw in the test — it is a mathematical property of how rare events interact with imperfect measurements. Two core takeaways follow from this: test high-risk groups first (pre-test probability matters), and do not treat a positive test result in a low-prevalence screening program the same way you would treat one from a high-risk clinical context.
Worked Example: COVID-19 Rapid Antigen Test
Calculating all four core metrics from a lateral flow test study
Scenario: A lateral flow antigen test is evaluated in 1,000 people. 200 are confirmed COVID-positive by PCR (the gold standard). 800 do not have COVID.
Build the confusion matrix. Of the 200 true COVID cases, the lateral flow test correctly detects 185 (TP = 185) and misses 15 (FN = 15). Of the 800 healthy people, the test correctly clears 776 (TN = 776) and wrongly flags 24 as positive (FP = 24).
Calculate sensitivity. Sensitivity = TP / (TP + FN) = 185 / (185 + 15) = 185 / 200 = 0.925, or 92.5%. The test detects 92.5% of true COVID cases and misses 7.5%.
Calculate specificity. Specificity = TN / (TN + FP) = 776 / (776 + 24) = 776 / 800 = 0.97, or 97%. The test correctly clears 97% of disease-free people and produces 3% false positives.
Calculate PPV. PPV = TP / (TP + FP) = 185 / (185 + 24) = 185 / 209 = 0.885, or 88.5%. When this test returns positive, there is an 88.5% chance the person genuinely has COVID — given this 20% prevalence setting.
Calculate NPV. NPV = TN / (TN + FN) = 776 / (776 + 15) = 776 / 791 = 0.981, or 98.1%. When this test returns negative, there is a 98.1% chance the person genuinely does not have COVID.
Interpret likelihood ratios. LR+ = 0.925 / (1 − 0.97) = 0.925 / 0.03 = 30.8. A positive lateral flow result is about 31 times more likely in a COVID-positive person than a COVID-negative one — strong confirmatory evidence. LR− = (1 − 0.925) / 0.97 = 0.075 / 0.97 = 0.077. A negative result is 13 times less likely in a COVID-positive person — good but not perfect for ruling out.
✓ Results: Sensitivity 92.5%, Specificity 97%, PPV 88.5%, NPV 98.1%, LR+ 30.8, LR− 0.077. In a hospital setting with 20% COVID prevalence, positive results reliably indicate true infection. In an airport with 1% prevalence, PPV would drop to roughly 24% from the same test numbers.
Worked Example: Breast Cancer Screening with Mammography
Mammography is one of the most studied and debated screening tests in medicine. Its performance figures are widely discussed in clinical literature, and its sensitivity-specificity trade-off illustrates the challenges of population-level cancer screening.
Interpreting mammography performance and the false positive challenge
Establish test performance figures. Based on data reviewed by the National Cancer Institute, mammography has a sensitivity of approximately 80–87% for detecting breast cancer, with specificity around 85–90%. These figures vary by age, breast tissue density, and imaging technique.
Apply to a screening population. Consider 10,000 women aged 50 screened by mammography. Breast cancer prevalence in this age group is roughly 0.4% (4 per 1,000). So approximately 40 women have breast cancer and 9,960 do not.
Calculate the confusion matrix. With 85% sensitivity: TP = 40 × 0.85 = 34; FN = 6 missed cancers. With 87% specificity: TN = 9,960 × 0.87 = 8,665; FP = 1,295 false alarms.
Calculate PPV. PPV = 34 / (34 + 1,295) = 34 / 1,329 = 0.026, or 2.6%. A positive mammogram in this population means roughly a 1-in-38 chance of cancer. This is not a test failure — it is the mathematical consequence of low prevalence. This is why positive mammograms trigger biopsy rather than immediate surgery.
Interpret the clinical implications. Over 10 years of annual screening, a woman has roughly a 60% probability of receiving at least one false-positive mammogram. These generate callbacks, additional imaging, biopsies, anxiety, and healthcare costs. The U.S. Preventive Services Task Force weighs these harms against the mortality reduction from early detection when developing screening guidelines.
✓ Lesson: Low PPV in population screening does not mean the test is bad — it reflects the mathematics of low-prevalence testing. The clinical value of screening comes from the lives saved through early detection, weighed against the harms from false positives. Both numbers are needed to evaluate the program.
Worked Example: Diabetes Screening with Fasting Blood Glucose
How threshold choice shifts sensitivity and specificity for HbA1c testing
Effect of lowering the threshold to 6.0%. More patients are classified as positive. This catches more early-stage cases (higher sensitivity) but also flags more non-diabetic individuals as positive (lower specificity). The result is more diagnoses, more interventions, but also more unnecessary treatment for people who would not have progressed.
Effect of raising the threshold to 7.0%. Fewer patients are classified as positive. Only clear-cut cases are diagnosed (higher specificity), but borderline cases who could benefit from early intervention are missed (lower sensitivity). Late diagnosis means more complications — neuropathy, nephropathy, retinopathy — before treatment begins.
Using the ROC curve to choose the threshold. Researchers construct a ROC curve by plotting sensitivity vs. 1 minus specificity at every possible HbA1c cutoff. The area under the curve for HbA1c as a diabetes diagnostic test is approximately 0.93 — excellent discrimination. The clinically chosen cutoff of 6.5% represents a balance between catching early disease and avoiding excessive false positives in healthy people with naturally variable HbA1c levels.
The two-test approach in practice. Clinical guidelines often recommend a second confirmatory test when HbA1c falls in the 5.7–6.4% prediabetes range. This mirrors the high-sensitivity screen followed by a high-specificity confirmer strategy: cast a wide net first, then confirm before committing to a lifetime diagnosis.
✓ Lesson: For continuous tests, sensitivity and specificity are not fixed properties — they depend on the chosen threshold. ROC analysis maps the full range of possible trade-offs, and the clinically selected cutoff reflects the balance between missing early disease versus over-diagnosing healthy individuals.
Choosing Between High Sensitivity and High Specificity
Deciding which measure to prioritise is a clinical judgment, not a statistical one. The right answer depends on three factors: the seriousness of the disease, the consequences of a missed diagnosis, and the harm caused by a false positive.
Define the clinical context
Is this a first-line screening test applied to a broad population, or a confirmatory test following a positive screen? The role in the diagnostic pathway determines which metric to optimise.
Weigh the cost of each error
What happens if a true case is missed (false negative)? What happens if a healthy person is labelled sick (false positive)? The greater harm determines which error to minimise.
Choose the priority metric
If missing disease is more dangerous → prioritise sensitivity. If false positives cause more harm → prioritise specificity. Many pathways use both in sequence.
Consider prevalence
In low-prevalence settings, even highly specific tests generate many false positives relative to true positives. Pre-test probability estimation improves test interpretation.
Use a two-test strategy when possible
Screen with a sensitive test to avoid missing cases, then confirm with a specific test before taking irreversible action. This is the standard pathway in HIV testing, cancer screening, and infectious disease control.
Evaluate with AUC for quantitative tests
For tests with continuous output, use ROC analysis to see the full sensitivity-specificity landscape. AUC tells you overall discriminatory power before committing to any single threshold.
| Clinical Scenario | Priority | Reason | Example Tests |
|---|---|---|---|
| HIV screening in blood donors | Maximum sensitivity | One missed case infects others; false positives face confirmatory testing, not immediate harm | 4th-generation ELISA, nucleic acid testing |
| HIV confirmatory test after positive screen | Maximum specificity | A false positive means lifelong stigma and medication with serious side effects | Western blot, HIV RNA quantification |
| Meningitis rule-out in emergency | Maximum sensitivity | Missing meningitis is fatal; false positive leads to lumbar puncture and antibiotics — uncomfortable but survivable | Clinical criteria (jolt accentuation, neck stiffness), CT imaging |
| Biopsy referral in cancer screening | Balanced, leaning specific | Biopsy carries procedural risk, anxiety, and cost; false positives must be minimised while still catching true cases | PSA + additional biomarkers, core biopsy protocols |
| Population TB screening | High sensitivity | Undetected TB drives transmission; false positives proceed to sputum culture confirmation | Tuberculin skin test, IGRA assay |
| Surgical referral for appendicitis | Balanced to specific | Unnecessary surgery causes real harm; missing appendicitis risks perforation — clinical scoring guides the balance | Alvarado score, CT abdomen |
| Drug testing in elite sport | Maximum specificity | A false positive ends a career unjustly; confirmatory testing required before sanctions | WADA-accredited laboratory methods |
Common Misconceptions About Diagnostic Test Accuracy
The most common error is treating accuracy as the primary metric — it is misleading when disease is rare. A test that always says negative achieves 99% accuracy in a 1% prevalence population while having zero clinical utility. Other frequent mistakes include believing sensitivity and specificity change with prevalence (they do not), assuming high sensitivity means no false positives (it does not), and using a single test result to rule in or out disease without considering pre-test probability.
| Misconception | Why It Is Wrong | The Correct Understanding |
|---|---|---|
| A highly sensitive test has no false positives | Sensitivity only measures performance among sick people. It says nothing about what happens in healthy people — that is specificity's domain. A 99% sensitive test can still produce thousands of false positives if applied to a healthy population with low disease prevalence. | Sensitivity and specificity are independent measures of different populations. High sensitivity controls false negatives; high specificity controls false positives. Both are needed to assess the full picture of test performance. |
| Sensitivity and specificity change depending on who gets tested | These are intrinsic test properties calculated within each disease group separately. They remain constant regardless of how many sick or healthy people are in the tested population. PPV and NPV change with prevalence — not sensitivity and specificity. | Sensitivity = TP / (TP + FN) uses only disease-positive people. Specificity = TN / (TN + FP) uses only disease-negative people. The ratio of sick to healthy in the population does not affect either calculation. What changes is how useful the test is — reflected in PPV and NPV. |
| Overall accuracy is the best way to judge a diagnostic test | Accuracy = (TP + TN) / Total. In a population where 1% of people have the disease, a test that always says "negative" achieves 99% accuracy while detecting exactly zero cases. Accuracy is heavily influenced by the dominant group and hides how well the test performs for the rare condition. | For clinical evaluation, sensitivity and specificity tell you what a test does to each patient group. For rare diseases, PPV is often more informative. For overall discrimination power, the AUC is preferred. Accuracy is most meaningful when disease prevalence is close to 50%. |
| A positive result means the patient has the disease | A positive result means the test responded positively. In low-prevalence settings, the majority of positives can be false positives. Even a test with 97% specificity generates many false alarms when prevalence is below 1%. | A positive result shifts the probability of disease upward from the pre-test probability. By how much depends on the LR+ and the pre-test probability via Bayes' theorem. A positive result initiates a diagnostic process; it does not conclude one. |
| A negative result safely rules out disease | The probability of disease after a negative result (1 minus NPV) depends on the test's sensitivity and the pre-test probability. A 90% sensitive test still misses 10% of cases. In a high-prevalence setting, the absolute number of missed cases can be large enough to warrant follow-up. | A negative result from a high-sensitivity test in a low-prevalence population is strong evidence against disease. A negative result from a low-sensitivity test in a high-prevalence population is much weaker. Always combine the result with clinical context and pre-test probability. |
| The best test is the one with the highest AUC | AUC measures average performance across all possible thresholds, which may include thresholds never used clinically. A test with a slightly lower AUC may perform much better at the specific threshold relevant to the clinical decision. | AUC is a useful summary but the final assessment must compare performance at the clinically relevant threshold — the one actually used in practice. Two tests with similar AUC can differ substantially at any particular cutoff point. |
Diagnostic Test Metrics Cheat Sheet
| Metric | Formula | What It Answers | Affected by Prevalence? | Priority When |
|---|---|---|---|---|
| Sensitivity | TP / (TP + FN) | Of all truly sick patients, what proportion does the test detect? | No | Missing disease is dangerous |
| Specificity | TN / (TN + FP) | Of all truly healthy patients, what proportion does the test correctly clear? | No | False positives cause serious harm |
| PPV | TP / (TP + FP) | If the test is positive, what is the probability the patient is truly sick? | Yes — rises with prevalence | Clinician needs to act on a positive result |
| NPV | TN / (TN + FN) | If the test is negative, what is the probability the patient is truly healthy? | Yes — rises as prevalence falls | Clinician needs to act on a negative result |
| Accuracy | (TP + TN) / Total | What proportion of all tests are correct? | Yes — inflated when one group dominates | Disease prevalence near 50% (balanced datasets) |
| LR+ | Sensitivity / (1 − Specificity) | How much more likely is a positive result in a sick patient than a healthy one? | No | Updating pre-test probability after a positive result |
| LR− | (1 − Sensitivity) / Specificity | How much less likely is a negative result in a sick patient compared to a healthy one? | No | Updating pre-test probability after a negative result |
| AUC | Area under the ROC curve | Across all thresholds, how well does the test separate sick from healthy? | No | Comparing overall test quality for continuous measures |
| False Positive Rate | FP / (TN + FP) = 1 − Specificity | Of all healthy patients, what proportion does the test wrongly flag as sick? | No | Quantifying over-referral burden from screening programs |
| False Negative Rate | FN / (TP + FN) = 1 − Sensitivity | Of all sick patients, what proportion does the test miss? | No | Assessing the risk of delayed diagnosis |
| F1 Score | 2 × (PPV × Sensitivity) / (PPV + Sensitivity) | Harmonic mean of precision and recall — balances both error types | Yes (via PPV) | Machine learning evaluation with imbalanced datasets |
Diagnostic Testing Glossary
| Term | Definition | Context / Notes |
|---|---|---|
| Sensitivity | The proportion of people with the disease that a test correctly identifies as positive; the true positive rate | Calculated within the disease-positive group only; unaffected by prevalence; mnemonic SnNout |
| Specificity | The proportion of people without the disease that a test correctly identifies as negative; the true negative rate | Calculated within the disease-negative group only; unaffected by prevalence; mnemonic SpPin |
| True Positive (TP) | A test result that correctly identifies a person who has the disease as positive | Appears in the upper-left cell of the confusion matrix |
| True Negative (TN) | A test result that correctly identifies a person without the disease as negative | Appears in the lower-right cell of the confusion matrix |
| False Positive (FP) | A test result that incorrectly identifies a healthy person as having the disease; also called a Type I error in hypothesis testing | Can lead to unnecessary treatment, anxiety, and further invasive procedures |
| False Negative (FN) | A test result that incorrectly identifies a person with the disease as healthy; also called a Type II error | The most clinically dangerous error when delayed diagnosis causes harm; minimised by high sensitivity |
| Positive Predictive Value (PPV) | The probability that a person who tests positive actually has the disease; TP / (TP + FP) | Depends on prevalence; rises in high-prevalence populations; the clinically actionable positive result interpretation |
| Negative Predictive Value (NPV) | The probability that a person who tests negative genuinely does not have the disease; TN / (TN + FN) | Depends on prevalence; rises in low-prevalence populations; the clinically actionable negative result interpretation |
| Confusion Matrix | A 2×2 table summarising the four possible test outcomes: TP, FP, FN, and TN; the foundation for all diagnostic accuracy calculations | Also called a contingency table or error matrix; used identically in machine learning and clinical testing |
| ROC Curve | A graph plotting sensitivity (y-axis) against 1 minus specificity (x-axis) across all possible test thresholds, showing the full sensitivity-specificity trade-off | Used to select optimal thresholds and compare competing tests; a curve closer to the top-left corner indicates better performance |
| AUC (Area Under the Curve) | The area under the ROC curve; a single summary statistic of overall test discrimination ranging from 0.5 (no better than chance) to 1.0 (perfect) | Also called the c-statistic; represents the probability that the test ranks a randomly chosen sick patient above a randomly chosen healthy patient |
| Likelihood Ratio Positive (LR+) | Sensitivity divided by (1 minus specificity); indicates how much more likely a positive result is in a person with the disease compared to one without | LR+ greater than 10 provides strong evidence for the presence of disease; does not change with prevalence |
| Likelihood Ratio Negative (LR−) | (1 minus sensitivity) divided by specificity; indicates how much less likely a negative result is in a person with disease compared to one without | LR− less than 0.1 provides strong evidence against disease; does not change with prevalence |
| Pre-Test Probability | The clinician's estimated probability of disease before the test result is known, based on clinical history, risk factors, and population prevalence | Combined with likelihood ratios via Bayes' theorem to calculate post-test probability; context-dependent and patient-specific |
| Post-Test Probability | The revised probability of disease after incorporating the test result into the pre-test probability | Calculated using Fagan's nomogram or Bayes' theorem; the probability a clinician acts on when making treatment decisions |
| Gold Standard | The reference test accepted as the definitive method for establishing whether a patient truly has the disease; used to classify patients into disease-positive and disease-negative groups for test evaluation | In practice, no gold standard is perfect; imperfect reference standards can bias sensitivity and specificity estimates |
| Cut-off Value (Threshold) | The numerical value at which a continuous test result is classified as positive or negative; changing the threshold shifts the sensitivity-specificity trade-off | Optimal thresholds depend on the relative clinical costs of false positives and false negatives; ROC analysis reveals the full range |
| Screening Test | A test applied to asymptomatic populations to detect disease at an early stage; generally requires high sensitivity to avoid missing cases | Positive screening results typically require confirmatory testing before diagnosis or treatment begins |
| Confirmatory Test (Diagnostic Test) | A test applied after a positive screen to verify the presence of disease; generally requires high specificity to minimise false positives before treatment | Often more expensive, invasive, or technically demanding than screening tests |
| Prevalence | The proportion of a defined population that has a specific disease or condition at a given point or period in time | Directly affects PPV and NPV but not sensitivity and specificity; high-prevalence populations yield better PPV from the same test |
| Diagnostic Accuracy | The overall ability of a test to correctly identify patients with and without disease; often measured by sensitivity, specificity, and AUC together | No single number fully captures diagnostic accuracy; different metrics answer different clinical questions |
| SnNout | Clinical mnemonic: a highly Sensitive test with a Negative result helps rule Out disease | Applies because a sensitive test rarely misses true cases, so a negative result strongly suggests the disease is absent |
| SpPin | Clinical mnemonic: a highly Specific test with a Positive result helps rule In disease | Applies because a specific test rarely flags healthy people, so a positive result strongly suggests the disease is present |
Frequently Asked Questions
Key sources and further reading: WHO Diabetes Fact Sheet · NCI Mammography Fact Sheet · USPSTF Breast Cancer Screening Recommendation · CDC Diabetes Testing · WHO COVID-19 Resources · Statistics Fundamentals — Home · Incidence vs. Prevalence Guide · Bayes' Theorem Guide · Type I and Type II Errors · Altman DG, Bland JM. Diagnostic tests 1: sensitivity and specificity. BMJ. 1994;308:1552 · Deeks JJ, Altman DG. Diagnostic tests 4: likelihood ratios. BMJ. 2004;329:168–169 · Metz CE. Basic principles of ROC analysis. Semin Nucl Med. 1978;8(4):283–298 · Gordis L. Epidemiology (5th ed.). Elsevier, 2014