Screening, Diagnosis, Monitoring

Educational orientation

Expanded educational edition: 2026-09-10. This page is not a screening or care recommendation.

By the End of This Chapter

Calculate predictive value in an example and distinguish screening from confirmation.

Learning matrix and gaps in the short edition

Core Idea

Screening, diagnosis, and monitoring answer different questions. Screening looks for signs of disease in people without related symptoms, within a defined population and criteria. Diagnosis asks whether a specific person has a specific disease. Monitoring follows known disease, risk, response, or recurrence over time.

Confusing these tasks leads to bad reasoning. A useful screening test must balance benefit, false positives, false negatives, overdiagnosis, access, and population risk. A diagnostic test is interpreted in a clinical context.

Step-by-Step Explanation

1. The question comes before the test

Screening looks for signs of disease in a population without related symptoms, under a program and its criteria. Diagnostic investigation clarifies a suspicion. Monitoring follows an already known condition or trajectory. A positive screening result is not a diagnosis; test performance also depends on the population and intended use.

2. Earlier detection and benefit are not synonyms

Sensitivity asks how many people with the condition test positive; specificity asks how many without it test negative. Positive predictive value instead asks how many positive results actually have the condition. Beyond false results, overdiagnosis means detecting disease that would not cause problems during a person's lifetime. Moving diagnosis earlier can also increase measured survival without postponing death, called lead-time bias.

3. Change prevalence while keeping the test

Repeat the exercise with 1,000 people, but now 100 have the condition. Keeping sensitivity at 80% and specificity at 90% gives 80 true positives and 90 false positives among the 900 without the condition. Positive predictive value becomes 80 ÷ 170 ≈ 47.1%. The test did not change; the population's composition did. This comparison explains why a result from a case-enriched sample cannot be transferred directly to population screening. Both scenarios are hypothetical and evaluate no product.

Supporting reading: NCI PDQ: Cancer Screening Overview.

The Concept Illustrated

Among the 107 positive results● 8 true positives○ 99 false positives

011,000 people10 with the condition

0210 with disease8 positive; 2 negative

03990 without disease99 positive; 891 negative

04107 positive results8 true positives: 7.5%

Entirely hypothetical example. Each mark represents one of 107 positive results: 8 true positives and 99 false positives. These are not data from a real test.

Worked Example

Of 1,000 people, 10 have the condition. At 80% sensitivity, the test finds 8 and misses 2. Among 990 without the condition, 90% specificity produces 891 negatives and 99 false positives. There are therefore 107 positives but only 8 true positives: 8 ÷ 107 ≈ 7.5%. There is no contradiction: the group without disease was much larger. These hypothetical counts do not determine whether a program should be adopted.

In this hypothetical example	With condition	Without condition	Total
Positive test	8	99	107
Negative test	2	891	893
Total	10	990	1,000

Essential Vocabulary

• Sensitivity: proportion of people with the condition who test positive.
• Specificity: proportion of people without the condition who test negative.
• False positive: test suggests disease when it is absent.
• Overdiagnosis: detecting disease that would not have caused harm.

What This Does Not Mean

More testing is not automatically better. Test value depends on the question and population.

Builder/Data Lens

Model thresholds change false positives and false negatives. Report the intended use before reporting performance.

Check Your Understanding

Does a positive result with 90% specificity mean a 90% probability of disease?

Answer and reasoning

No. Specificity conditions on absence of disease; interpreting a positive result needs a different denominator and depends on prevalence. Confusing these probabilities changes the statistical question.

Optional Deeper Exploration

Lead-time example: diagnosis in year 2 and death in year 8 give six years of survival; diagnosis in year 5 with the same death gives three. That survival difference alone does not demonstrate reduced mortality.

Go Deeper

• Epidemiology and prevention
• Clinical oncology 101
• ML pitfalls

References

1. National Cancer Institute. Cancer Screening
2. National Cancer Institute. Tests and Procedures Used to Diagnose Cancer