Learning outcomes
By the end of this lesson you should be able to: distinguish screening from diagnosis and state the ten classic criteria a screening programme must satisfy; compute sensitivity, specificity, predictive values and likelihood ratios from a 2x2 table; explain how prevalence changes the predictive values of a test while sensitivity and specificity stay fixed; describe the types of screening and the biases — lead-time, length-time and overdiagnosis — that make a screened population look better without living longer; and define surveillance, its types, and the sources on which disease surveillance in India runs.
Prerequisites
Lessons 5 and 6. Screening is evaluated with the 2x2 table and the measures of the epidemiological methods lesson, and the natural history and iceberg of disease from lesson 1 explain why it works at all.
The big idea
A screening test is a filter, not a verdict. Its value rests on one property of the disease: a long pre-clinical phase during which treatment still works — cancer of the cervix, hypertension, congenital hypothyroidism, phenylketonuria. Where the silent phase is long, a cheap filter applied to thousands will find the few diseases that matter in time to change their outcome; where the disease is already incurable at detection, the filter merely labels people earlier. The entire discipline of screening evaluation is the arithmetic of that trade: how many false alarms each programme buys, and whether it changes deaths rather than diagnoses.
| Screening test | Diagnostic test | |
|---|---|---|
| Population | Well people, no symptoms | People with symptoms or a positive screen |
| Aim | To sort for further testing | To establish the diagnosis and guide treatment |
| Acceptance | Applied to thousands; must be cheap, safe, acceptable | Applied to few; may be invasive and expensive |
| Result | 'At risk' or 'not at risk' | Disease present or absent |
| Consequence of a mistake | Anxiety, needless tests — and a false sense of security | Missed disease and inappropriate treatment |
The Wilson and Jungner criteria
In 1968 Wilson and Jungner set the standard that still decides whether a screening programme should exist. The examiners ask the list; the real world applies it.
| Criterion | What it requires |
|---|---|
| The condition is an important health problem | It is common, or serious, or both |
| Treatment for the recognised disease exists | An accepted, effective treatment is available |
| Facilities for diagnosis and treatment are available | The system can handle the generators |
| A recognisable latent or early symptomatic stage exists | There is a pre-clinical phase that treatment can benefit from |
| A suitable test or examination exists | Valid, safe and acceptable to the population |
| The test is acceptable to the population | Acceptable in cost, discomfort and culture |
| The natural history is adequately understood | Including how the disease develops |
| Agreement on whom to treat as a patient | A diagnostic boundary exists after the screen |
| The cost of case-finding is economically balanced | Cost is balanced against the benefit achieved |
| Case-finding is a continuing process | Not a one-off campaign |
The performance of a test
| Measure | Formula | The question it answers |
|---|---|---|
| Sensitivity | True positives / all diseased (a/(a+c)) | Of those with disease, what fraction will the test catch? |
| Specificity | True negatives / all well (d/(b+d)) | Of those well, what fraction will the test clear? |
| Positive predictive value | True positives / all positives (a/(a+b)) | Of those positive, what fraction truly have disease? |
| Negative predictive value | True negatives / all negatives (d/(c+d)) | Of those negative, what fraction truly have none? |
| Likelihood ratio positive | Sensitivity / (1 - specificity) | How much the odds of disease rise on a positive |
| Likelihood ratio negative | (1 - sensitivity) / specificity | How much the odds of disease fall on a negative |
A validation study screens 300 people and follows each to the reference standard: 100 are diseased and 200 are well; the screen is positive in 90 of the diseased and 20 of the well.
| Disease present | Disease absent | Total | |
|---|---|---|---|
| Screen positive | 90 (true positive) | 20 (false positive) | 110 |
| Screen negative | 10 (false negative) | 180 (true negative) | 190 |
| Total | 100 | 200 | 300 |
| Measure | Calculation | Result |
|---|---|---|
| Sensitivity | 90 / 100 | 90% |
| Specificity | 180 / 200 | 90% |
| Positive predictive value | 90 / 110 | 81.8% |
| Negative predictive value | 180 / 190 | 94.7% |
| Prevalence | 100 / 300 | 33.3% |
| Likelihood ratio positive | 0.90 / (1 - 0.90) | 9.0 |
NoteRepeat the same test where prevalence is 5% instead of 33.3%: sensitivity and specificity stay 90%, but the positive predictive value falls to 32% — roughly two of every three positive screens are false alarms. This is why a screening test that performs beautifully in a hospital (where prevalence is high) can become a false-alarm factory in a community camp, and why screening in low-prevalence populations must choose tests with high specificity.
Types of screening and its biases
| Type | Definition | Example |
|---|---|---|
| Mass screening | Screening applied to a whole population group | Neonatal screening for hypothyroidism and phenylketonuria |
| Multiple screening | Several tests at one visit | BP, blood sugar, haemoglobin and vision at a health camp |
| Targeted (high-risk) screening | Restricted to people above a risk | Mammography screening for women over 40 with a family history; HIV screening in antenatal women |
| Case-finding (opportunistic) | Screening who presents for another reason | Cervical screening at a family planning visit |
| Bias | What makes the screened group look better | The control |
|---|---|---|
| Lead-time bias | Disease is diagnosed earlier without any change in survival, so 'survival from diagnosis' lengthens | Compare mortality rates between populations, not survival from diagnosis |
| Length-time bias | Slow-growing, screen-detectable cases are over-represented; the aggressive cases are missed | A randomised trial of screening with mortality as the endpoint |
| Overdiagnosis | The screen finds disease that would never have caused symptoms or death | Autopsy series; cancer incidence that rises while mortality does not |
| Selection (healthy volunteer) bias | The people who enrol in screening are healthier to start with | Randomised invitation; compare intention-to-screen populations |
The single decisive judge of a screening programme is mortality from the disease in the screened population compared with an unscreened population — randomised and unbiased by the lead time or the enthusiasm. Cervical cytology screening reduced cervical cancer mortality in countries where coverage was high; mammography's effect is real but far smaller, because breast cancer in its pre-clinical phase is less often a disease that screening can call forward. This is the honest arithmetic behind a Wilson and Jungner criterion being a judgement, not a slogan.
Surveillance
| Type of surveillance | How it works | Strength and weakness |
|---|---|---|
| Passive | Health workers report routinely to the system | Cheap and continuous; under-reporting is the rule |
| Active | The programme goes and seeks cases | Complete data, expensive; used for elimination programmes |
| Sentinel | A chosen set of sites report carefully | Good quality data for those sites; not population-representative |
| Syndromic | Cases are reported by a syndrome, not a diagnosis | Fast and cheap; not specific |
In India, the Integrated Disease Surveillance Project (IDSP) collects syndromic, presumptive and laboratory-confirmed reports from reporting units through the S, P, L surveillance forms; a handful of notifiable diseases must be reported immediately (plague, cholera, polio, meningitis, dengue, SARS-class outbreaks), and the National Centre for Disease Control coordinates the response. Surveillance is also how non-communicable disease is now tracked — stroke and diabetes registries, risk-factor surveys — and it is the reason a new syndrome is noticed within days rather than years.
Common mistakes
Watch outQuoting a positive predictive value as a fixed property of the test. Sensitivity and specificity travel with the test; predictive values travel with the population, and they rise and fall with its prevalence. Always quote the prevalence alongside the predictive value, and expect the PPV to shrink in a low-prevalence camp.
Watch outTreating increased 'survival after diagnosis' as proof that screening saves lives. Lead-time bias and length-time bias can lengthen survival after diagnosis while no life is saved; only a mortality comparison by randomised invitation settles the question.
Watch outRunning a screening programme before facilities exist. The Wilson and Jungner criteria demand available diagnosis and treatment; a screen that finds cases the system cannot treat buys anxiety and costs credibility, and it fails the economic criterion as well.
Check your understanding
From the worked validation table, state the sensitivity, specificity and negative predictive value, and interpret each.
Why does the positive predictive value collapse when the same screening test moves from a hospital clinic to a community camp?
A district reports that five-year survival after breast cancer diagnosis rose after a screening programme. What two biases must be excluded before claiming a benefit?
Multiple choice
From the worked validation table, the positive predictive value is:
A screening programme for cervical cancer is planned. Which of the Wilson and Jungner criteria is fulfilled by the availability of colposcopy and treatment facilities?
Of all the sources of bias in screening evaluation, which is removed by comparing disease mortality in a randomly invited and an uninvited population?