How to Evaluate Diagnostic Test Accuracy

By Kevin Fernando · Sep 17, 2026 · 9 min read

Diagnostic test accuracy is measured through sensitivity, specificity, and predictive values. Sensitivity and specificity are stable properties of the test itself. Positive and negative predictive values depend on disease prevalence in the population being tested, so the same test produces different predictive values in high-prevalence versus low-prevalence settings.

When you read a diagnostic study or a device submission, accuracy numbers appear everywhere. A test reports high accuracy, or strong sensitivity, or a positive predictive value. These numbers sound authoritative but they measure different things, and a high number in one metric can coexist with a useless test in your clinical setting.

The problem is that accuracy metrics interact with disease prevalence, how common the condition is in the population being tested. A test validated in hospitalized patients with a high disease rate may perform very differently when used in primary care, where disease is rare. Understanding which metrics are stable properties of the test itself and which ones shift with prevalence is the foundation for reading diagnostic studies honestly.

What you'll be able to do

After working through this guide, you will understand the four core accuracy metrics and what each one measures. You will recognize which metric answers which clinical question. You will spot when a test's reported accuracy is misleading, when a high overall accuracy number masks a test that cannot reliably identify disease in your population, or when predictive values are inflated because the study enrolled an unrealistic mix of diseased and healthy subjects. You will know how to extract the information you need from published studies and regulatory documents, and how to compare tests fairly.

Understand the four core accuracy metrics

All four metrics come from a 2×2 table of test results against true disease status. Every subject tested falls into one of four cells: true positive (test positive, disease present), false positive (test positive, disease absent), true negative (test negative, disease absent), or false negative (test negative, disease present). The metrics are ratios drawn from this table, and they answer different questions about test performance.

Sensitivity and specificity: what the test does in a known population

Sensitivity measures the proportion of positive test results among all subjects who truly have the disease. It answers: "If someone has the disease, how likely is the test to catch it?" Sensitivity is calculated as true positives divided by all truly positive subjects (true positives plus false negatives), multiplied by 100.

Specificity measures the proportion of negative test results among all subjects who truly do not have the disease. It answers: "If someone does not have the disease, how likely is the test to correctly say so?" Specificity is calculated as true negatives divided by all truly negative subjects (true negatives plus false positives), multiplied by 100.

Both metrics assume disease status is already known. You are measuring how well the test performs when you know who has the disease and who does not. Critically, sensitivity and specificity do not depend on how common the disease is in the population. A test with the same sensitivity will have that sensitivity whether the disease affects a small portion of people or a large portion. This stability is what makes these metrics useful for describing the test itself.

Positive and negative predictive value: what the test means for a real patient

Positive predictive value (PPV) measures the proportion of subjects with a positive test result who actually have the disease. It answers the clinical question: "My patient tested positive. What is the probability they actually have the disease?" This is what a clinician and a patient want to know.

Negative predictive value (NPV) answers: "My patient tested negative. What is the probability they do not have the disease?" A high NPV means a negative result is likely to be correct.

Unlike sensitivity and specificity, predictive values depend on disease prevalence, how common the disease is in the population being tested [F9, F10]. The same test (same sensitivity and specificity) will have very different predictive values in a high-prevalence population versus a low-prevalence one. This is not a flaw in the test; it is how probability works. When disease is rare, even a positive test is more likely to be a false positive than a true positive, so PPV drops. When disease is common, a negative test is more likely to be a false negative, so NPV drops.

Gather the test's reported performance data

Published diagnostic studies report accuracy metrics in the results section, usually in a table or in the text. Regulatory submissions for diagnostic devices include a performance section that states sensitivity, specificity, and often predictive values. Product documentation and clinical validation studies may report these metrics as well.

Look for a table labeled "diagnostic accuracy", "test performance", or "sensitivity and specificity". If the study reports only overall accuracy, or only sensitivity, or only specificity, that is a red flag, you do not have enough information to evaluate the test fairly. The study should report all four metrics, or at minimum sensitivity, specificity, and the prevalence of disease in the study population so you can calculate predictive values yourself.

Identify the study population and disease prevalence

Before you trust any accuracy number, you must know who was tested and how many had confirmed disease. Extract three pieces of information: the total number of subjects tested, the number with confirmed disease, and the number without disease. Disease prevalence is the number with disease divided by the total.

Find these numbers in the study methods (sample size, enrollment criteria) and results (how many subjects were positive, how many negative). If the study does not state these numbers explicitly, it should report them in a contingency table or in the text as "X subjects had disease and Y did not."

Check whether the study population matches your clinical context

A test validated in hospitalized patients with confirmed or suspected disease may perform very differently in primary care, where the disease is suspected but less common. The study population's disease prevalence may not match the prevalence in the setting where you will use the test.

Compare three things: the clinical setting (hospital, primary care, screening program), the enrollment criteria (symptomatic patients, asymptomatic screening, referred patients), and the disease prevalence in the study. If the study enrolled only symptomatic patients and you want to use the test for screening asymptomatic people, the study population does not match your context. If the study reports one disease prevalence but your patient population has a different prevalence, the predictive values you read in the paper will not apply to your patients. This mismatch is one of the most common reasons a test performs worse in real use than the published numbers suggest.

Calculate or verify the four metrics from the raw data

If the study provides a 2×2 contingency table with raw counts (true positives, false positives, true negatives, false negatives), you can verify the reported metrics or calculate missing ones.

Sensitivity = (true positives) / (true positives + false negatives), multiplied by 100.

Specificity = (true negatives) / (true negatives + false positives), multiplied by 100.

Positive predictive value = (true positives) / (true positives + false positives), multiplied by 100.

Negative predictive value = (true negatives) / (true negatives + false negatives), multiplied by 100.

If the study reports sensitivity and specificity but not predictive values, and you know the disease prevalence, you can calculate predictive values using the prevalence formulas. Positive predictive value can be computed from sensitivity, specificity, and prevalence [F14, F15]. Negative predictive value can be computed from sensitivity, specificity, and prevalence [F8, F12]. Plug in the numbers and work through the arithmetic to see what the predictive values should be in your population.

Recognize how prevalence shifts predictive values

Suppose a test has 95% sensitivity and 95% specificity. These numbers are fixed properties of the test. Now suppose you use the same test in two different populations.

In a high-prevalence setting, the positive predictive value is higher than in a low-prevalence setting. A positive test is more likely to be correct when disease is common. In a low-prevalence setting, the positive predictive value is lower than in a high-prevalence setting. A positive test is more likely to be false than true, even though the test itself has not changed.

This shift happens because when disease is rare, the denominator of false positives grows relative to true positives. Even a highly specific test will generate many false positives when it is applied to a large population with little disease. Conversely, when disease is common, the negative predictive value falls because false negatives become more common relative to true negatives.

This is why you cannot simply apply a test's published predictive values to your population unless your population's disease prevalence matches the study population's prevalence. The test's sensitivity and specificity remain stable, but the predictive values change.

Spot common accuracy reporting pitfalls

Several reporting practices obscure a test's real-world performance. Learning to recognize them will help you identify studies where the accuracy numbers are misleading or incomplete. Watch for these red flags when reading diagnostic studies or device submissions.

High accuracy with low predictive value

Accuracy is the overall proportion of correct results (true positives plus true negatives divided by all subjects). When disease is rare, most subjects do not have disease, so the test can be correct on most of them simply by calling them negative. A test that calls everyone negative will have high accuracy in a low-prevalence population even though it is useless for identifying disease.

Accuracy alone tells you almost nothing about whether the test works for your purpose. Always demand sensitivity, specificity, and predictive values, not just accuracy.

Unequal case-control study designs

Some diagnostic studies enroll equal numbers of diseased and non-diseased subjects to ensure enough power to detect differences. This is a valid research design for some purposes, but it artificially inflates predictive values because it does not reflect real-world disease prevalence.

If a study enrolls subjects with disease and subjects without disease in a 1:1 ratio, the reported positive and negative predictive values will be higher than they would be in a real population where disease is much rarer. When you apply the test to your patients, where disease prevalence is lower, the actual predictive values will be worse than the paper reports. The sensitivity and specificity remain valid because they do not depend on prevalence, but the predictive values do. If a study uses an unequal case-control design, calculate what the predictive values would be at your population's true disease prevalence before you decide whether the test is fit for use.

Decide whether the test is fit for your purpose

Match the test's metrics to the clinical decision you need to make. Different decisions prioritize different metrics.

If your goal is to rule out disease, to exclude the diagnosis and avoid unnecessary treatment, prioritize high sensitivity and high negative predictive value. You want the test to catch disease when it is present, so you do not send someone home who actually has it. A test with high sensitivity will miss few cases, but the acceptable miss rate depends on the consequences of missing disease and whether you have a follow-up test.

If your goal is to rule in disease, to confirm the diagnosis and justify starting treatment, prioritize high specificity and high positive predictive value. You want the test to be correct when it is positive, so you do not treat someone who does not have the disease. A test with high specificity will incorrectly flag few healthy people, but if those false positives are few in absolute number and the consequences of treatment are serious, that may be acceptable.

If your goal is screening, identifying disease in an asymptomatic population, you typically need both high sensitivity (to catch cases) and high specificity (to avoid false alarms that trigger unnecessary follow-up). But the balance depends on the disease, the treatment, and the consequences of missing or misidentifying disease. A screening test for a serious treatable disease may tolerate lower specificity; a screening test for a rare disease may require very high specificity to avoid overwhelming the system with false positives.

Before you adopt a test, confirm that its sensitivity, specificity, and predictive values are appropriate for your clinical decision, and that the study population's disease prevalence is close enough to your population's prevalence that the published predictive values will hold.

Do this automatically

Clinicians evaluating a diagnostic test must manually extract sensitivity, specificity, and predictive values from published studies, cross-reference these metrics against their patient population, and repeat this lookup process each time a new test or clinical scenario arises. Software can systematically organize and retrieve these accuracy metrics from diagnostic literature, allowing clinicians to quickly identify which performance measure answers their specific clinical question. Healthcare research runs this kind of analysis end to end and cites every figure back to its original source, across as many subjects as you need, with Mimir attaching a citation to every figure. It will not predict a trial's outcome or advise on anyone's treatment. Those decisions stay with clinicians and regulators, and deliberately so. Every plan includes it, including the free one.

Start your first report and run it on something you already know well.

This article is general information about healthcare and life-sciences topics. It is not medical advice, diagnosis, or treatment guidance, and it is not a substitute for a qualified professional. Any treatment decision should be discussed with a clinician.