Reading Cancer Survival Figures: What Five-Year Survival Can and Cannot Tell

A cancer survival figure describes a group of people diagnosed years ago, usually compared with the general population of the same age. It can rise because cancers are found earlier, or found more often, even when nobody lives longer, and it is not a forecast for any one person. Across 20 common cancers in the US over 45 years, the change in five-year survival bore almost no relation to the change in deaths^[1].

Survival numbers travel: into charity appeals, news stories, policy targets and clinic waiting rooms, often without the small print that gives them meaning. For someone newly diagnosed, or a family searching online, a single percentage can feel like a verdict. It is not one. This explainer sets out, using the definitions of the agencies and researchers who produce the figures, what the main survival measures are, how they can mislead, and why survival and mortality can tell different stories.

Survival, incidence and mortality answer three different questions

• Incidence counts new diagnoses in a population, usually as a rate per 100,000 people a year.
• Mortality counts deaths from the cancer in the whole population, usually per 100,000 people a year: for US thyroid cancer, an age-adjusted 0.5 per 100,000 in 2020–2024^[3].
• Survival follows only the people who have been diagnosed, and measures how many are alive, or alive after allowing for other causes of death, one, five or ten years later.

Survival has the number of diagnoses built into its denominator, so anything that changes who gets diagnosed, and when, changes survival, whether or not fewer people die. Mortality has the whole population as its denominator and is not inflated simply by finding more cancers, or finding them sooner. That is why researchers write that improved survival and increased early detection "do not prove that cancer screening tests save lives"^[2], and why the US National Cancer Institute says that falls in late-stage incidence "represent a better measure of decreased cancer mortality due to screening than do 5-year survival trends"^[4].

None of this makes survival a poor statistic. Five-year survival is "a valid measure for comparing cancer therapies in a randomized trial", where everyone enters at the same point^[1]. Population survival is also how health systems are compared, as the study on international survival describes. The care lies in knowing which survival is quoted, and what moves it.

Observed, relative and net survival measure different things

People with cancer, particularly older people, can die of other causes: heart disease, stroke, accidents. A survival measure has to decide what to do about those deaths, and the main measures differ precisely on that point.

An illustrative example, not from any dataset: if 600 of 1,000 people diagnosed are alive five years later, observed survival is 60%. If life tables say 80% of people of the same age and sex would have been alive anyway, relative survival is 60% ÷ 80% = 75%.

Each has weak points: relative survival assumes "independent competing causes of death", and death certificates sometimes record the site a cancer spread to rather than where it began^[6].

Why international comparisons use net survival, age-standardised

In 2012 the statisticians Maja Pohar Perme, Janez Stare and Jacques Estève noted that the standard relative survival estimators "do not provide information on cancer mortality that is independent of the national general population mortality", so are "not suitable for comparison between countries", and proposed a new net survival estimator^[8]. Global comparisons now rest on net survival: the CONCORD programme estimates five-year net survival "adjusted for background mortality in every country or region by age (single year), sex, and calendar year"^[9].

Headline figures are also age-standardised: weighted to a common age mix, the International Cancer Survival Standard, so that a country where people are diagnosed at older ages does not look worse for that reason alone. CONCORD-3 (37.5 million people diagnosed 2000–2014, 71 countries and territories) uses those weights^[10], as does NHS England, which notes that "age-standardisation allows comparisons between population groups and over time"^[5].

Net survival can exceed 100%. NHS England's latest release (people diagnosed 2018–2022, followed up to 2023) reports five-year net survival of 101.1% for men and 102.4% for women with basal cell carcinoma, a type of skin cancer^[11]. NHS England explains that above 100% means "people with cancer live longer than expected", which can happen in cancers such as melanoma "often linked to higher socioeconomic groups"^[5]. At the other end, five-year net survival for pancreatic cancer was 9.4% for men and 9.3% for women^[11].

Five-year survival describes people diagnosed years ago

Registry survival estimates "are usually dated measures of current-year survival, because of the time needed to observe survival and lag between available data and the current year"^[12]. The most recent English figures, published in April 2026, cover people diagnosed in 2018–2022^[11]; the CONCORD-3 global figures, published in 2018, cover people diagnosed up to 2014^[10].

Registries use three main ways of grouping people to get the most up-to-date answer they can^[12]:

• Cohort analysis "includes calendar years for which all cases have potential follow-up for the survival duration". It is the most intuitive, and the most out of date.
• Complete analysis includes everyone diagnosed in the most recent years, including some not yet followed for the full period, and so is more current.
• Period analysis "uses only the most recent interval survival estimate of cases diagnosed in different calendar years": a snapshot of recent survival experience.

Period analysis was introduced by Hermann Brenner and Olaf Gefeller in 1996 because traditional methods disclose changes in prognosis "with considerable delay", by analogy with demographers' period life tables^[13]. Tested on Finnish Cancer Registry data for 15 cancers diagnosed in 1983–1987, period estimates were close to the 10-year relative survival later observed (median difference 0.9 percentage points); cohort estimates available at the time were much lower (median 5.8 points, maximum 18.4)^[14]. When survival is improving, a cohort figure understates it. Anyone quoting a figure should say whose diagnosis years it describes and which method was used.

Finding cancer earlier can lengthen survival even when no one lives longer

Survival is counted from the date of diagnosis. Move that date earlier and survival gets longer, whether or not death is delayed. This is lead-time bias, which the NCI defines as "longer estimated survival of people with screen-identified cancers because the survival calculation includes the time preceding what would have been the clinical diagnosis of the cancer in the absence of screening"^[4].

An illustrative example with invented numbers: picture a group whose cancers would be diagnosed from symptoms at 67 and who die of the cancer at 70. Survival from diagnosis is three years, so none reach five years. Now a test finds the same cancers at 64, but nothing that follows changes the outcome, and death still comes at 70. Survival is now six years (70 − 64) and five-year survival has gone from 0% to 100%, with no change in deaths. Real screening can bring real benefits; this arithmetic is why survival alone cannot show them.

The second distortion is overdiagnosis: "the diagnosis of a 'cancer' that would otherwise not go on to cause symptoms or death", in the definition of H. Gilbert Welch and William Black. It needs "a silent disease reservoir and activities leading to its detection (particularly cancer screening)"^[15]. Overdiagnosed cancers, by definition, "have 100% cancer-specific survival"^[4]; adding them to the count raises survival while deaths stay where they were. From randomised trials, Welch and Black estimated that about 25% of mammographically detected breast cancers, 50% of lung cancers detected by chest x-ray and/or sputum tests, and 60% of PSA-detected prostate cancers were overdiagnosed^[15].

Thyroid cancer in the US shows the pattern in the data. Among adults in nine SEER areas, incidence "nearly tripled, from 4.9 to 14.3 per 100,000" between 1975 and 2009, almost entirely papillary cancers, while the death rate stayed at about 0.5 per 100,000^[16]. The authors concluded it was "not an epidemic of disease but rather an epidemic of diagnosis"^[16]. SEER five-year relative survival for thyroid cancer (SEER 8 registries) rose from 92.95% for people diagnosed in 1975 to 98.43% in 2018^[3].

Comparing lifetime risk of diagnosis in Australia in 2012 with 1982, adjusted for changes in risk factors, researchers estimated that 18% of all cancer diagnoses in women and 24% in men were overdiagnosed, about 11,000 and 18,000 cancers a year^[17]. These are modelled estimates, sensitive to their assumptions.

A related "healthy screenee bias" arises because "people who choose to be screened are often more health conscious than the general population"^[4]. The study on screening and the study on stage at diagnosis look at what earlier diagnosis can and cannot achieve.

Survival and mortality can move in different directions

If survival gains came only from better treatment, deaths should fall too. Welch, Lisa Schwartz and Steven Woloshin tested this using US SEER data for the 20 most common solid tumours, 1950–1995. Five-year survival rose for all 20, by between 3 percentage points (pancreatic cancer) and 50 points (prostate cancer). Death rates fell for 12 and rose for 8. The correlation between change in survival and change in mortality was essentially zero (Pearson r = .00), while the change in survival was positively correlated with the change in incidence (Pearson r = +.49)^[1]. The authors concluded that survival changes "appear primarily related to changing patterns of diagnosis"^[1].

That does not make survival gains illusory, but it makes them hard to read alone. US prostate cancer shows how tangled the picture can be. SEER's five-year relative survival (SEER 8 registries) rose from 67.68% for men diagnosed in 1975 to 98.65% in 2018, while incidence peaked at 224.93 per 100,000 in 1992. The age-adjusted death rate rose from 30.94 per 100,000 in 1975 to 39.74 in 1992, then fell to 18.95 by 2022^[18]. Survival alone cannot separate the parts played by treatment, earlier diagnosis and overdiagnosis; that needs mortality data and trials, as the prostate cancer study discusses.

Even professionals find this hard. In a randomised survey of 412 US primary care physicians (2010–2011), 69% recommended a hypothetical screening test described by five-year survival rising from 68% to 99%, against 23% for a test described by cancer deaths falling from 2 to 1.6 per 1,000 people. Almost half (47%) said that finding more cancers in screened people "proves that screening saves lives"^[2]. The authors' conclusion: "only reduced mortality in a randomized trial constitutes evidence of the benefit of screening"^[2].

A population figure is not a forecast for one person

NHS England states that its estimates "do not reflect the survival prospects for any individual cancer patient; they represent the net survival for all cancer patients in each area, in a given period of time, diagnosed with a specified type of cancer"^[5]. The NCI calls net survival a "policy-based statistic" and says the crude probability of death is better suited to "the impact of cancer diagnosis at an individual level"^[6].

A group figure cannot be read as one person's chances:

• It averages over very different situations. In the US, five-year relative survival for prostate cancer, people diagnosed 2016–2022, was 98.2% overall, but 100.0% when the cancer was found while localised and 40.1% when it had spread to distant organs^[18]. Individual circumstances vary further within each stage.
• It describes the past. A five-year figure is about people diagnosed at least five years ago, under the treatments and services of that time^[12].
• It is net of other causes, by design. Net survival imagines a world in which no one dies of anything else^[6].
• It may be age-standardised. Headline figures are weighted to a standard age mix so that groups and periods can be compared^[5]; they do not describe the ages of the people actually diagnosed.

"Five-year survival" is also a convention for comparing groups at a fixed point, not a threshold after which a person is safe.

What the evidence asks of us

Survival statistics are one of the few ways to compare how well health systems serve people with cancer. CONCORD was set up as "a metric of the effectiveness of health systems", and since 2017 the OECD has used its results "as the official benchmark of cancer survival"^[10]. Used carefully, they make the case for better services.

• Fund and protect population-based cancer registries. Every measure here depends on complete registration and reliable linkage to deaths. The CONCORD-3 authors call on governments to "recognise population-based cancer registries as key policy tools"^[10].
• Report mortality and stage alongside survival. Because survival can rise with more diagnosis alone, progress claims, including for screening, should be judged on mortality and falls in late-stage disease^[4][2].
• Monitor overdiagnosis. The Australian estimates led their authors to conclude that "rates of overdiagnosis need to be reduced and health services should monitor emerging areas of overdiagnosis"^[17]. People with overdiagnosed cancers carry the cost of treatment without benefit; that is a failure of the system, not of the person.
• Publish the small print with the number. Every survival figure should travel with its measure (observed, relative or net), its time point, the diagnosis years, the country or registry, whether it is age-standardised, and the stage if given.

Language: this piece says "people with cancer" and "people living with and after cancer", and describes the illness and its treatment plainly, without combat metaphors. Clinical terms such as "stage", "incidence" and "net survival" are used in the sense the cited source defines them.

Why we're publishing this

Almost every study in our cancer research library quotes survival. Charities use these numbers to argue for earlier diagnosis and fairer access; people living with and after cancer meet them at some of the hardest moments of their lives. Both deserve figures that are quoted accurately and explained honestly: neither inflated by diagnosis patterns nor read as a verdict on anyone.

How to read this data

Definitions come from the bodies that produce the statistics (NHS England, the US National Cancer Institute's SEER programme and Surveillance Research Program, CONCORD) and the methods papers behind them. Most worked data are from the US, where SEER publishes long linked series of incidence, mortality and survival, with some from England and Australia; the principles apply everywhere, but effect sizes differ by country and cancer. The Welch analysis covers 1950–1995 and the thyroid analysis 1975–2009; they establish how survival can move, not current magnitudes. Overdiagnosis estimates are modelled and depend on their assumptions. The two illustrative examples (relative survival and lead time) are arithmetic, not data. Figures from different registries and methods are not directly comparable, and none of them is a prediction for an individual.

Use this data

Free to cite with attribution to Better Health For Everyone. Charities, campaigners and journalists are welcome to reuse these figures and charts. For the underlying figures as a spreadsheet, get in touch.

Important: This article summarises published research for general information. It is not medical advice and cannot be used to diagnose, treat or predict anything about an individual. If you have concerns about your own or someone else's health, please speak to a qualified health professional.

Sources

1. Welch HG, Schwartz LM, Woloshin S. Are increasing 5-year survival rates evidence of success against cancer? JAMA, 2000;283(22):2975–2978. europepmc.org/article/MED/10865276
2. Wegwarth O, Schwartz LM, Woloshin S, Gaissmaier W, Gigerenzer G. Do physicians understand cancer screening statistics? A national survey of primary care physicians in the United States. Annals of Internal Medicine, 2012. europepmc.org/article/MED/22393129
3. National Cancer Institute, SEER Program. Cancer Stat Facts: Thyroid Cancer (incl. trend table). SEER, accessed November 2026. seer.cancer.gov/statfacts/html/thyro.html
4. National Cancer Institute. Cancer Screening Overview (PDQ®) – Health Professional Version, updated 16 October 2023. cancer.gov. cancer.gov/about-cancer/screening/hp-screening-overview-pdq
5. NHS England Digital. Cancer Survival in England, cancers diagnosed 2018 to 2022, followed up to 2023: Interpretation of these statistics. NHS England, 2026. digital.nhs.uk/…/interpretation-of-these-statistics
6. National Cancer Institute, Surveillance Research Program. Measures of Cancer Survival. surveillance.cancer.gov, accessed November 2026. surveillance.cancer.gov/survival/measures.html
7. National Cancer Institute, SEER Program. Cancer Survival Statistics. SEER, accessed November 2026. seer.cancer.gov/statistics/types/survival.html
8. Perme MP, Stare J, Estève J. On estimation in relative survival. Biometrics, 2012;68(1):113–120. europepmc.org/article/MED/21689081
9. Allemani C, Weir HK, Carreira H, et al.; CONCORD Working Group. Global surveillance of cancer survival 1995–2009: analysis of individual data for 25,676,887 patients from 279 population-based registries in 67 countries (CONCORD-2). The Lancet, 2015. europepmc.org/article/MED/25467588
10. Allemani C, Matsuda T, Di Carlo V, et al.; CONCORD Working Group. Global surveillance of trends in cancer survival 2000–14 (CONCORD-3). The Lancet, 2018. europepmc.org/article/MED/29395269
11. NHS England Digital. Cancer Survival in England, cancers diagnosed 2018 to 2022, followed up to 2023. NHS England, published 9 April 2026. digital.nhs.uk/…/cancers-diagnosed-2018-to-2022-followed-up-to-2023
12. National Cancer Institute, Surveillance Research Program. Cancer Survival Statistics: Cohort Definition Using Diagnosis Year. surveillance.cancer.gov, accessed November 2026. surveillance.cancer.gov/survival/cohort.html
13. Brenner H, Gefeller O. An alternative approach to monitoring cancer patient survival. Cancer, 1996. europepmc.org/article/MED/8909323
14. Brenner H, Hakulinen T. Up-to-date long-term survival curves of patients with cancer by period analysis. Journal of Clinical Oncology, 2002. europepmc.org/article/MED/11821467
15. Welch HG, Black WC. Overdiagnosis in cancer. Journal of the National Cancer Institute, 2010. europepmc.org/article/MED/20413742
16. Davies L, Welch HG. Current thyroid cancer trends in the United States. JAMA Otolaryngology–Head & Neck Surgery, 2014. europepmc.org/article/MED/24557566
17. Glasziou PP, Jones MA, Pathirana T, Barratt AL, Bell KJ. Estimating the magnitude of cancer overdiagnosis in Australia. Medical Journal of Australia, 2020. europepmc.org/article/MED/31858624
18. National Cancer Institute, SEER Program. Cancer Stat Facts: Prostate Cancer (incl. trend table). SEER, accessed November 2026. seer.cancer.gov/statfacts/html/prost.html