EDUKATE SINGAPORE · GOLD STANDARD PRACTICAL CASE STUDY 193
A test detecting 95% of true cases does not mean 95% of its positive results are true.
A fictional screening test has 95% sensitivity and specificity but is used in a synthetic population where only 1% has the condition. The resulting positive predictive value is around 16%, a very different answer from the number on the promotional brochure. This case shows how to compute and communicate the distinction safely.
All conditions, participants and statistical counts are invented solely for teaching. This is public-health evidence-management education, not a medical test recommendation or an estimate of anyone’s personal risk.
The worked numbers at a glance
- Tested population: 10,000 hypothetical people.
- Condition prevalence: 1%, or 100 people.
- Sensitivity: 95%, producing 95 true positives and 5 false negatives.
- Specificity: 95%, producing 9,405 true negatives and 495 false positives.
- Total positive tests: 590.
- Positive predictive value: 95 / 590 ≈ 16.1%, not 95%.
Chapter navigation
- 1. A strong test sensitivity becomes a misleading promise to screened people
- 2. Sensitivity asks about people who have the condition
- 3. A 10,000-person worked example reveals the base-rate effect
- 4. CDC says predictive values depend on prevalence and intended population
- 5. NCI distinguishes positive predictive value from sensitivity
- 6. A positive screen and a confirmed diagnosis are different decisions
- 7. An accuracy headline can conceal the relevant false-positive burden
- 8. The original communication should be corrected before release
- 9. The acceptance test must include populations with different prevalence
- 10. Practice: 95 true positives and 495 false positives
1. A strong test sensitivity becomes a misleading promise to screened people
A fictional public-health research team evaluates a screening test for an invented condition called Condition Z. The programme’s planning document specifies 95% sensitivity and 95% specificity under its synthetic training assumptions. Management prepares educational materials explaining how people should interpret a positive result.
A communications draft declares that a positive screening result means there is a 95% chance the person has Condition Z. The analyst spots a problem: the figure 95% refers to sensitivity, or the proportion of people with the condition correctly identified as positive under the example. It is not the proportion of positive tests that represent true cases.
In a population where the condition is uncommon, even a test with high sensitivity and specificity can produce more false positives than true positives. The Gold Standard question is whether the statistical statement being communicated matches the direction of the underlying conditional probability and the population to which the test will be applied.
2. Sensitivity asks about people who have the condition
Sensitivity is the proportion of people who truly have the target condition who receive a positive test result under an appropriately defined validation process. Under the invented 95% sensitivity, ninety-five out of one hundred people with the condition would test positive while five would not, in the simplified expected-count model.
Specificity is the proportion of people without the condition who test negative. A hypothetical 95% specificity means that among one hundred people without the condition, about ninety-five test negative and five test positive despite not having the condition.
Neither quantity independently answers the question a person who has just tested positive may naturally ask: what fraction of positive results in this tested population come from people who truly have the condition? That is positive predictive value, which depends on the prevalence and relevant characteristics of the population as well as test performance.
3. A 10,000-person worked example reveals the base-rate effect
Imagine a fully synthetic population of 10,000 with true Condition Z status established for this teaching exercise. Exactly 1%, or 100 people, have the condition; 9,900 do not. Assume the test has 95% sensitivity and 95% specificity in this population under the model.
Of the 100 people with Condition Z, 95 test positive and 5 test negative. Of the 9,900 without the condition, 95%—that is 9,405—test negative, while 5%—that is 495—test positive. The total number of positive tests is 95 + 495 = 590.
Only 95 out of those 590 positives are true positives, yielding positive predictive value 95/590, approximately 16.1%. The other 495 positives are false positives in this synthetic example. The test can correctly detect 95% of people with the condition and still have most positives fail to correspond to actual Condition Z status in a low-prevalence population.
4. CDC says predictive values depend on prevalence and intended population
The Centers for Disease Control and Prevention’s Field Epidemiology Manual on Epidemiology–Laboratory Collaboration advises that laboratory results be interpreted in the context of the question being asked, the test’s sensitivity and specificity, disease prevalence and the population used to validate performance.
CDC’s Diagnostic Sensitivity and Specificity job aid explains true positives, false negatives, true negatives and false positives. These definitions are essential to distinguishing the fraction a test captures among true cases from the fraction of positives that are true cases.
The CDC guidance does not establish any screening performance for fictional Condition Z or mean that all tests with 95% sensitivity have a 16% predictive value. That particular result depends on the invented 1% prevalence and 95% specificity used here. A different population or test could yield another result.
5. NCI distinguishes positive predictive value from sensitivity
The US National Cancer Institute’s Cancer Screening Overview explicitly distinguishes sensitivity, specificity, positive predictive value and negative predictive value. It explains that predictive values depend on prevalence through Bayes’ rule and that positive predictive value generally declines when the screened population’s prevalence is lower, with other characteristics fixed.
NCI’s definition of positive predictive value asks about the likelihood a person with a positive result actually has the condition being tested. This is the reverse conditional question from sensitivity, which starts with known condition status.
These NCI publications address real cancer screening and clinical concepts; the fictional numbers in this case must not be read as cancer-test performance or personal disease risk. The Gold Standard applies the statistical distinction to responsible health-service communication without recommending any individual test or medical action.
6. A positive screen and a confirmed diagnosis are different decisions
The hypothetical screening service has a preliminary test event, a result-reporting step and a separate authorised clinical pathway for interpreting whether further assessment is appropriate. A positive screening result may warrant a specific follow-up under the actual medical guidance, but the test’s score is not automatically a completed diagnosis.
The service must not tell a synthetic patient that 95% sensitivity means a 95% personal chance of illness. Nor should it reassure a real person that their individual risk is exactly 16.1% based on an invented population model. Personal interpretation depends on the actual test, symptoms, clinical context and medical professionals’ assessment.
The Gold Standard separates screening detection, predictive interpretation, clinical review and communication. GS-R03 identifies legitimate decision authority while GS-R08 ensures a statistical model is not transformed into a false certainty about an individual.
7. An accuracy headline can conceal the relevant false-positive burden
Management may present overall accuracy as one number. Under the synthetic 10,000-person table, 95 true positives plus 9,405 true negatives make 9,500 correct classifications, or 95% overall accuracy. Yet a person who receives a positive result sees a very different pattern: only 95 of 590 positive results correspond to Condition Z.
The same 95% numerical headline can therefore describe accuracy, sensitivity or specificity without telling a positive recipient what that result means. Each measure has a different denominator. A public dashboard should name those denominators and the conditions under which the estimates were obtained.
False-positive results can create anxiety and lead to further evaluation under actual clinical pathways. Their significance should not be minimised or exaggerated using an anonymous percentage. GS-R01 identifies the service purpose, GS-R02 the performance boundary and GS-R08 the legitimate claims supported by a given statistic.
8. The original communication should be corrected before release
The programme’s medical and analytics owners hold the promotional draft and preserve the test specification, evaluation data, assumptions and calculation. They replace the unsupported 95% personal-probability claim with an explanation distinguishing the three statistical measures. A health professional should approve any public patient-facing clinical advice.
The immediate correction is not to replace ‘95%’ with ‘16%’ everywhere. The 16.1% result is derived from a controlled hypothetical population, not a universal fact about the actual service. A responsible message describes real data and limits, and explains the appropriate approved follow-up route without alarming or reassuring people through invented numbers.
The centre should avoid publishing any identifiable screening records to prove the arithmetic. The worked table can be illustrated entirely with fictional data while protecting the privacy and dignity of actual participants.
9. The acceptance test must include populations with different prevalence
The evaluation team creates synthetic ten-thousand-person tables for more than one plausible prevalence under a fixed hypothetical sensitivity and specificity. Reviewers calculate true positives, false positives and predictive values and confirm that the reported interpretation changes when the underlying tested population changes.
A second scenario gives a good model result from a population unlike the intended recipients. The team must identify what external validation is still needed rather than assume performance transports unchanged. A third changes the test cutoff, potentially affecting both sensitivity and specificity, and requires appropriate review.
GS-R07 protects the handoff from method evaluation to public language; GS-R12 handles changes to population, threshold or technology. A correctly calculated table for one population cannot certify another without supporting evidence.
10. Practice: 95 true positives and 495 false positives
Exercise. In a fictional population of 10,000, Condition Z prevalence is 1%, test sensitivity is 95% and specificity is 95%. Calculate how many test positive among those who have and do not have the condition, then compute positive predictive value.
One hundred people have the condition; 95 test positive and 5 test negative. Among 9,900 without it, 495 test positive and 9,405 test negative. Positive predictive value is 95/(95 + 495), or about 16.1%, not 95%.
The Gold Standard conclusion is that sensitivity and positive predictive value answer different conditional questions. A screening result needs an appropriate, clinically authorised explanation grounded in the real tested population and test characteristics, not a convenient but misleading percentage.
Questions about screening statistics
If a test is 95% sensitive, does positive mean 95% likelihood of disease?
No. Sensitivity begins with people who truly have the condition. Positive predictive value begins with people who tested positive and depends on prevalence and specificity too.
Why are there more false positives in the example?
Because 9,900 people do not have the condition. Five percent of that large group yields 495 positive results, compared with 95 true positives from the smaller group.
Is 16.1% a typical screening test predictive value?
No. It belongs only to the hypothetical 1% prevalence and 95% sensitivity and specificity used in this exercise.
Does a positive screen establish a clinical diagnosis?
Not automatically. Real follow-up and diagnostic decisions require the governing clinical pathway and qualified professionals.
Can predictive value change without changing test technology?
Yes. When the tested population’s prevalence or characteristics differ, predictive values may change even with unchanged assumed sensitivity and specificity.
What should public reports show?
Definitions and denominators for sensitivity, specificity, predictive value, prevalence, sample and study limitations, with patient-facing implications reviewed by the appropriate clinical authority.
Evidence and communication checks before a public screening claim
- Define the condition, intended population and accepted screening purpose.
- Provide support for sensitivity and specificity in populations relevant to intended use.
- State the assumed prevalence and distinguish it from individual clinical circumstances.
- Calculate the appropriate predictive values and check all table denominators.
- Keep the screening result separate from diagnosis and clinical follow-up decisions.
- Require competent medical and data review of any public patient-facing statement.
A table that makes the direction of probability visible
An instructor can draw a four-cell table with true status across the top and screening result down the side. The two positive cells contain 95 true positives and 495 false positives. The two negative cells contain 5 false negatives and 9,405 true negatives. Summing by disease status produces 100 and 9,900. Summing by test result produces 590 positives and 9,410 negatives. Students can then see that the same number 95 is part of different ratios depending on what population is used as the denominator.
Sensitivity begins with the 100 people known to have Condition Z and asks how many test positive. Positive predictive value begins with the 590 people who received a positive result and asks how many truly have Condition Z. The order of the words is not decoration: reversing the condition of the probability changes the answer. A person may understand 95 out of 100 perfectly but still answer the wrong question if the denominator is switched.
Why changing the population changes the answer
If the prevalence is much higher in a different synthetic population, a larger share of positive results may correspond to true cases with the same assumed test characteristics. If prevalence is lower, false positives among many unaffected individuals may dominate. That is why test results cannot be interpreted from sensitivity alone and why validation in the actual population matters. A screening programme that changes who it invites may change what a positive result can reasonably imply.
A careful health-service manager asks the analyst which estimate came from which population and whether its uncertainty was properly characterised. The clinician asks what the result means under the actual patient’s circumstances. Neither should transform the teaching table into a personal diagnosis. The Gold Standard links technical statistics to an appropriately authorised next decision, not to dramatic numbers selected for publicity.
The difference between a false alarm and a diagnostic conclusion
In the synthetic exercise, all 10,000 people have a known true condition status solely so that the arithmetic can be taught. A real screening programme often does not know every person’s true status at the instant of the first test. Establishing performance characteristics requires an appropriate reference method and study design; follow-up can be needed to evaluate an initial screening finding under the applicable clinical pathway. Calling a screening positive a ‘false alarm’ is appropriate in the teaching table only because its synthetic true status has been specified. In actual public communications, the phrase should be chosen carefully so people are neither unnecessarily frightened nor falsely reassured.
The centre also must not interpret the 495 fictional false positives as negligence or evidence that the technology is worthless. A test threshold may reflect trade-offs among sensitivity, specificity and the purposes of the programme. The appropriate balance depends on the condition, possible benefits, potential harms, downstream capacity and clinical guidance. An editorial case cannot determine the right cutoff for real people. Its limited lesson is that the consequences of a screening result cannot be described accurately by one sensitivity percentage without the relevant population and other performance information.
Comparing two populations with the same invented test
For a second classroom demonstration, keep the 95% sensitivity and 95% specificity but imagine a different group of 10,000 in which 10%—one thousand people—have Condition Z. The hypothetical test would correctly flag 950 of those one thousand and miss 50. Among 9,000 without it, five percent or 450 would test positive and 8,550 negative. The total positive count would be 1,400, making positive predictive value 950/1,400, approximately 67.9%. None of the test’s assumed sensitivity or specificity changed, yet the predictive interpretation changed substantially.
That contrast makes the role of prevalence visible. A manager cannot transfer the 16.1% predictive value from the 1%-prevalence example to a group with a different base rate and call the number universally correct. Nor can the higher example be used to claim that a particular individual has a 67.9% probability without an appropriate clinical model and personal context. The exercise illustrates an important property of conditional probability, not an approved test’s actual accuracy in either population.
Measurement accuracy and useful screening service are different promises
A test vendor may correctly report sensitivity and specificity from an appropriate validation study. A health programme may correctly record that every sample was processed and every result dispatched. Those are legitimate stages. The ultimate service still needs understandable communication, access to whatever medically appropriate follow-up is indicated, protection of sensitive records and a clear owner for unresolved results. A green laboratory-completion rate does not show whether the recipient understands the meaning of a preliminary positive result.
An organisation should also be cautious about the other side of screening: a negative result is not automatically a guarantee that a condition is absent. False negatives exist under the assumed sensitivity, and negative predictive value likewise depends on prevalence. The Gold Standard demands that neither direction of a statistical statement be inflated beyond its actual meaning. For teaching, it is enough to explain why the denominator changes; real screening counselling belongs to qualified clinical staff and applicable protocols.
A red-team review before the brochure goes public
A useful controlled editorial exercise gives a mock communications team three statements: ‘This test detects ninety-five percent of people with the condition under the selected validation assumptions’; ‘A positive test means a ninety-five percent chance of the condition’; and ‘In our fictional low-prevalence population, roughly sixteen percent of positive results are true cases’. The reviewer must distinguish the supported definition of sensitivity from an unsupported positive-result inference, then trace the illustrative predictive value to its 10,000-person table. The team should flag any claim whose denominator or source population cannot be reconstructed.
Next, change the invented prevalence while retaining the same sensitivity and specificity. A legitimate analytics process should calculate a new predictive value and label the new population clearly. A clinical owner then reviews any patient-facing language to ensure that the communication does not provide personal diagnosis from hypothetical numbers. This test requires no actual patients, samples or private records. Its success is a precise, reproducible public explanation of what a screening statistic does and does not support.
The Gold Standard result worth reporting
A bounded management finding is that sensitivity was misrepresented as positive predictive value in the original communications draft. The simulated population has ninety-five true positive results, four hundred and ninety-five false positives and five hundred and ninety total positives, supporting a positive predictive value of about 16.1%. No actual clinical result is established by that exercise. The responsible team should preserve the input assumptions, correct the wording, and demonstrate that future educational materials distinguish the conditional probabilities before release.
This is the same general evidence principle as the earlier Gold Standard case on laboratory nondetection: a technically accurate property of an assay can become misleading when it is described as evidence of a stronger different proposition. Responsible communication does not require hiding numerical performance; it requires naming the question each number truly answers, so patients, practitioners and policymakers can make their own legitimate decisions from sound evidence.
Official and technical source references
- CDC · Field Epidemiology Manual, Laboratory Interpretation
- CDC · Diagnostic Sensitivity and Specificity Job Aid
- NCI · Cancer Screening Overview and Predictive Values
- NCI · Positive Predictive Value Definition
- Gold Standard · Measurement and Performance Review
- Gold Standard · Evidence, Claims and Verification
- Gold Standard · Core Requirements and Evidence
- Case 128 · Laboratory Nondetect with High Reporting Limit
- Case 141 · AI-Writing Detector Misused as Proof
- Gold Standard Casebook and Decision Practice
The fictional numbers do not represent actual cancer, infectious disease or screening performance. Real clinical interpretation requires approved protocols and qualified health professionals. GS-R01–GS-R12 is eduKate’s educational management reference, not medical certification.