Community medicine chapters
Screening tests
What screening is and the criteria for a good screening programme, how to calculate sensitivity, specificity and predictive values from a 2 × 2 table, how prevalence changes predictive values, reliability, and lead-time and length-time bias.
- Sensitivity = ability to detect those with disease (TP ÷ all diseased); specificity = ability to identify those without disease (TN ÷ all healthy); both are properties of the test.
- Predictive values depend on prevalence: when a disease is rare, even a good test has a low PPV (many false positives).
- Lead-time bias makes survival look longer only because diagnosis is earlier; length-time bias happens because screening picks up slow-growing disease more often.
On this page
Overview
Screening is the search for unrecognised disease or risk factors in apparently healthy people, using tests that can be applied quickly, to sort out those who probably have the disease from those who probably do not.
A screening test is not diagnostic. People who screen positive need a diagnostic test to confirm.
Screening is secondary prevention (see levels of prevention).
Criteria for a screening programme (Wilson and Jungner, WHO)
The disease:
- An important health problem.
- Has a recognisable latent or early stage.
- Its natural history is well understood.
The test:
- Suitable, simple, safe and acceptable to the public.
- Valid and reliable.
Treatment and services:
- An effective treatment exists, and treating early is better than treating late.
- Facilities for diagnosis and treatment are available.
- An agreed policy on whom to treat.
- The cost is balanced against overall health spending.
- Case finding is a continuing process.
Types of screening
- Mass screening: a whole population (now less favoured).
- High-risk (selective) screening: groups at high risk (for example diabetes screening in obese adults).
- Multiphasic screening: several tests at one time.
- Opportunistic screening (case finding): when a person attends for another reason.
The 2 × 2 table
| Disease present | Disease absent | Total | |
|---|---|---|---|
| Test positive | a (true positive) | b (false positive) | a + b |
| Test negative | c (false negative) | d (true negative) | c + d |
| Total | a + c | b + d |
Validity (accuracy)
Validity is the ability of a test to correctly separate those with and without disease.
| Measure | Formula | Meaning |
|---|---|---|
| Sensitivity | a ÷ (a + c) | Proportion of diseased people who test positive |
| Specificity | d ÷ (b + d) | Proportion of healthy people who test negative |
| False-negative rate | c ÷ (a + c) = 1 − sensitivity | Diseased people missed |
| False-positive rate | b ÷ (b + d) = 1 − specificity | Healthy people wrongly labelled |
| Accuracy | (a + d) ÷ total | Proportion correctly classified |
Sensitivity looks at the people with disease (column a + c). Specificity looks at the people without disease (column b + d).
SnNout: a highly Sensitive test, when Negative, rules the disease out. SpPin: a highly Specific test, when Positive, rules it in.
Which matters more?
- High sensitivity is needed when missing a case is dangerous and treatment is effective (screening blood donors for HIV and hepatitis; screening for phenylketonuria in newborns).
- High specificity is needed when a false positive would cause harm, costly tests or stigma (confirming HIV before telling a patient).
There is usually a trade-off: moving the cut-off to increase sensitivity lowers specificity, and vice versa. A ROC curve plots sensitivity against (1 − specificity) for different cut-offs.
Predictive values
Predictive values answer the patient's question: "I tested positive; what is the chance I have the disease?"
| Measure | Formula | Meaning |
|---|---|---|
| Positive predictive value (PPV) | a ÷ (a + b) | Of those testing positive, the proportion who truly have the disease |
| Negative predictive value (NPV) | d ÷ (c + d) | Of those testing negative, the proportion who truly do not have it |
Effect of prevalence
- Sensitivity and specificity do not change with prevalence.
- PPV rises as prevalence rises; NPV falls.
- In a low-prevalence population, most positives are false positives, even with a good test.
Prevalence ↑ → PPV ↑, NPV ↓. That is why screening is more efficient in high-risk groups.
1,000 people screened; prevalence 10% → 100 have the disease, 900 do not
Test sensitivity 90%, specificity 80%
TP = 90% of 100 = 90; FN = 100 − 90 = 10
TN = 80% of 900 = 720; FP = 900 − 720 = 180
PPV = 90 ÷ (90 + 180) = 90 ÷ 270 = 33.3%
NPV = 720 ÷ (720 + 10) = 720 ÷ 730 = 98.6%
Accuracy = (90 + 720) ÷ 1,000 = 81%
1,000 people; prevalence 1% → 10 diseased, 990 healthy
TP = 9, FN = 1; TN = 80% of 990 = 792, FP = 198
PPV = 9 ÷ (9 + 198) = 9 ÷ 207 = 4.3%
Same sensitivity and specificity, but PPV falls from 33% to about 4%
Likelihood ratios
- Positive likelihood ratio = sensitivity ÷ (1 − specificity). In the example above: 0.9 ÷ 0.2 = 4.5.
- Negative likelihood ratio = (1 − sensitivity) ÷ specificity = 0.1 ÷ 0.8 = 0.125.
Like sensitivity and specificity, likelihood ratios do not depend on prevalence.
Reliability (repeatability, precision)
Reliability is the ability of a test to give the same result when repeated on the same person under the same conditions. It depends on:
- Biological variation in the person (BP varies through the day).
- Observer variation: between observers (inter-observer) and the same observer at different times (intra-observer). Measured by the kappa statistic.
- Instrument (method) error.
A test can be reliable but not valid (consistently wrong, like a scale that always reads 2 kg too high). A valid test must also be reasonably reliable.
Combining tests
- Series (sequential) testing: a person is called positive only if both tests are positive → specificity increases, sensitivity falls (for example HIV confirmation).
- Parallel (simultaneous) testing: positive if either test is positive → sensitivity increases, specificity falls.
Biases in evaluating screening
| Bias | What happens |
|---|---|
| Lead-time bias | Screening finds disease earlier, so the time from diagnosis to death looks longer, even if death happens at the same time as it would have without screening |
| Length-time bias | Screening is more likely to detect slow-growing, less aggressive disease (which has a long detectable phase), so screen-detected cases seem to do better |
| Overdiagnosis | Screening finds disease that would never have caused symptoms or death (an extreme form of length-time bias) |
| Selection (volunteer) bias | People who choose to be screened are often healthier than those who do not |
Lead time is the period by which screening advances the diagnosis. Survival measured from diagnosis looks longer even when the patient does not live longer. Comparing mortality rates in randomised trials avoids this bias.
Expect: calculations of sensitivity, specificity, PPV and NPV from a 2 × 2 table, the effect of prevalence on predictive values, SnNout and SpPin, which property matters for a given screening purpose, validity vs reliability, series vs parallel testing, Wilson and Jungner criteria, and lead-time vs length-time bias.
Quick revision
- Sensitivity = a ÷ (a + c); specificity = d ÷ (b + d).
- PPV = a ÷ (a + b); NPV = d ÷ (c + d).
- Prevalence affects predictive values, not sensitivity or specificity.
- SnNout and SpPin.
- Series testing ↑ specificity; parallel testing ↑ sensitivity.
- Validity = accuracy; reliability = repeatability (kappa).
- Lead-time bias: earlier diagnosis, not longer life. Length-time bias: slow cancers found more often.
Written by The House Job from Excel Community Medicine and Park's Textbook of Preventive and Social Medicine (named in the UHS PGET syllabus).
FAQ
What is the sensitivity of a test?
The proportion of people with the disease who test positive: true positives ÷ (true positives + false negatives).
How does prevalence affect the positive predictive value?
When prevalence rises, the positive predictive value rises; when the disease is rare, most positive results are false positives.
What is lead-time bias?
The apparent increase in survival time caused only by diagnosing the disease earlier through screening, without changing the time of death.
What is the difference between validity and reliability?
Validity is whether the test measures what it should (accuracy); reliability is whether it gives the same result when repeated.