Ask Mallard

Evaluating a screening programme

A screening evaluation asks whether systematically testing people without symptoms leads to fewer of them dying of the disease, or living better, than not testing them.
NHMRC EVIDENCE LEVEL · 2009Depends on how the study is designed · Screening interventionProvisional design classification. Study quality and certainty of findings require separate appraisal.
Why this level?The design family and its defining features have not been specified for this hierarchy. NHMRC 2009 source

The level follows how the programme was evaluated, and that is the substance of the page. A randomised screening trial is a randomised comparison; a before-and-after look at a programme that was simply rolled out is not, and lead-time and length bias make the second one look better than it is.

Where the NHMRC 2009 hierarchy names this design, it places it here:

Research questionLevelWhen
Screening interventionIIParticipants or clusters are randomly allocated to concurrent comparison groups.
Screening interventionIII-2An observational comparison uses concurrent controls. Statistical adjustment does not change this design level.
Screening interventionIVA single case series has post-test or pre-test/post-test outcomes, without a comparison group.

What question does this help answer?

Whether a screening programme should exist. That is a question about a population, about harms as well as benefits, and about a decision made on behalf of healthy people.

Why survival from diagnosis cannot answer it

Five-year survival rises whenever diagnosis moves earlier, whether or not anyone lives a day longer, and rises further whenever screening adds cases that would never have progressed. Across cancers over four decades, changes in five-year survival correlate with changes in INCIDENCE and hardly at all with changes in mortality. A screening study reporting improved survival among screen-detected cases has demonstrated that screening detects cases, and nothing more.

The three biases, concretely

  • LEAD TIME: diagnosis at 60 instead of 65, death at 70 either way. Survival goes from 5 years to 10 with no benefit at all.
  • LENGTH BIAS: a tumour that takes eight years to become symptomatic has eight years in which a screening round can find it; one that takes eight months has eight months. Screening therefore enriches for indolent disease.
  • OVERDIAGNOSIS: the extreme of length bias, where the disease would never have surfaced in the person's lifetime. Estimated at roughly a quarter of screen-detected breast cancers and a majority of PSA-detected prostate cancers.
  • Healthy screenee effect: people who attend screening differ systematically from those who do not, in ways that predict outcome independently.
  • Stage shift is a necessary condition for benefit and not evidence of it, since all three biases produce a stage shift too.

What lead time and length bias look like

Two screening biases on timelines. Lead time: with screening the disease is diagnosed earlier but the fatal outcome happens at the same time, so survival measured from diagnosis is longer with no benefit. Length bias: slow-growing disease stays detectable across several screening rounds while fast-growing disease falls between them, so screening finds mostly slow disease.Lead time: an earlier diagnosis, the same fatal outcomeWithout screeningFrom symptomsWith screeningBy screeningSurvival from diagnosis, without screeningSurvival from diagnosis, with screening: longer, and nothing changedTimeLength bias: slow disease is detectable for longer, so screening rounds find more of itDetectable before symptomsDetectableScreening roundsSlow-growing diseaseFast-growing diseaseTimeHow often the disease proves fatal across everyone invited to screening is the endpoint neither biascan move.Disease beginsDiagnosisFatal outcome
Conceptual diagram. It shows the structure of the idea, not data from any study.

What a defensible design looks like

Randomize people to be INVITED to screening or not, follow everyone including non-attenders, and compare disease-specific mortality by the group they were invited to. All-cause mortality is the more demanding endpoint and is usually underpowered, but it is the one immune to misattribution of cause of death. The unit of analysis is the invitation, not the test.

Harms are part of the answer, not a caveat

False positives lead to further investigation, anxiety and sometimes invasive procedures in people who were well. Overdiagnosis leads to treatment of disease that would never have mattered, with the full toxicity of that treatment. A screening evaluation that reports only the mortality benefit has reported half the result.

Why Mallard may recommend it

Mallard may point here when a description proposes early detection, a new screening test, or a comparison of screen-detected with symptomatic cases, and it will question survival-from-diagnosis as an endpoint.

What to watch out for

Three biases make an ineffective screening programme look effective, and all three operate even when nothing about the disease has changed. LEAD TIME moves the diagnosis earlier without moving death, so survival measured from diagnosis lengthens automatically. LENGTH BIAS means screening preferentially catches slow-growing disease, which had a better prognosis anyway. OVERDIAGNOSIS finds disease that would never have caused harm, adding survivors who were never at risk. The only endpoint immune to all three is DISEASE-SPECIFIC MORTALITY in the whole invited population.

Clinical example

Question: Should we screen for a condition with a new blood test? Design: Randomize invitation to screening, follow all invited and uninvited people, and compare disease-specific mortality in the whole invited group, reporting false positives, investigations and estimated overdiagnosis alongside it.

How to report it

No reporting guideline dedicated to the evaluation of a screening programme was found in a PubMed search on 2026-09-16. CONSORT covers a randomised screening trial and STARD covers the accuracy of the test itself, and neither answers the mortality question this design asks. Report disease-specific mortality in the whole invited group, with the overdiagnosis and lead-time work shown rather than asserted. A reporting guideline is not a design manual: use it to make the study transparent, not as a substitute for careful design. Check the target journal's own requirements as well, which this page cannot know.

Recent studies that used this
Searching PubMed…
Related Learn pages
Sources & further reading
  1. Overdiagnosis in cancer (opens in a new tab) · J Natl Cancer Inst · doi 10.1093/jnci/djq099
  2. Are increasing 5-year survival rates evidence of success against cancer? (opens in a new tab) · JAMA · doi 10.1001/jama.283.22.2975
  3. CONSORT 2025 statement: updated guideline for reporting randomised trials (opens in a new tab) · BMJ · doi 10.1136/bmj-2024-081123
  4. STARD 2015: an updated list of essential items for reporting diagnostic accuracy studies (opens in a new tab) · BMJ · doi 10.1136/bmj.h5527
Last source review: 2026-09-06
AI-generated educational content
Mallard uses reputable methodological and reporting sources, but this material may contain errors or omit important context. Verify important methodological decisions with the cited sources and, when appropriate, a qualified research methodologist or statistician.