In 2007, a US presidential candidate who had survived prostate cancer told voters that his chance of surviving five years was 82% in the United States but only 44% in England. The figures sounded decisive. Later data made the contrast even starker — around 98% five-year survival in the US against 71% in Britain. Yet the proportion of men dying of prostate cancer was almost identical in the two countries: about 26 per 100,000 in the US and 27 per 100,000 in Britain. How can survival differ so much while deaths do not? The answer lies in three statistical illusions that make screening look better than it is: lead-time bias, length-time bias and overdiagnosis.
Lead-time bias: starting the clock earlier
Survival statistics count time from diagnosis. Screening moves the diagnosis earlier. If finding a cancer earlier does nothing to change when a person dies, survival still grows — simply because the clock started sooner. That extra time is the lead time.
Consider an example used by Gerd Gigerenzer and colleagues: a man whose cancer will kill him at age 70, whatever happens.
Diagnosis by screening
A screening test finds the tumour. The survival clock starts here. At 65 he is alive — he counts as a "five-year survivor".
Diagnosis without screening
Without screening, symptoms would have led to diagnosis now. The clock would start seven years later, and he would not reach five years of survival.
Death
He dies at the same age in both scenarios. Five-year survival rises from 0% to 100%. Not a single day of life has been gained.
Length-time bias: screening catches the slow ones
Tumours grow at very different speeds. A slow-growing cancer spends years in a silent, detectable phase; an aggressive one may go from undetectable to symptomatic between two screening rounds. A test repeated at intervals is therefore much more likely to catch the slow tumours — the ones with a better prognosis anyway. Screen-detected cancers look more benign than symptom-detected cancers not because screening made them so, but because screening preferentially found the benign ones. The aggressive tumours tend to surface as "interval cancers" between screens.
Overdiagnosis: cancers that would never have mattered
The most extreme form of length-time bias is overdiagnosis: finding a real cancer, confirmed under the microscope, that would never have caused symptoms or death during the person's lifetime. Every overdiagnosed person becomes a "cancer survivor", which inflates survival statistics further. Gigerenzer's illustration: if 440 of 1,000 people with progressive cancer survive five years, survival is 44%. Add 2,000 overdiagnosed people with non-progressive cancer, and survival rises to 81% — with exactly the same number of deaths.
South Korea's thyroid cancer "epidemic"
In South Korea, thyroid ultrasound was offered as an inexpensive add-on (about US$30–50) to the national cancer screening programme launched in 1999. The incidence of thyroid cancer rose about 15-fold from 1993. Thyroid cancer mortality stayed flat. The vast majority of the newly found tumours were small papillary cancers that would most likely never have caused harm. Treatment did cause harm: in an insurance-claims analysis of more than 15,000 Koreans who had surgery, 11% developed hypoparathyroidism (permanent calcium problems) and 2% vocal-cord paralysis.
Neuroblastoma screening in infants
Neuroblastoma, a childhood tumour, can be detected by a simple urine test. In Quebec, about 476,000 infants were screened between 1989 and 1994. Neuroblastoma was diagnosed about twice as often as in unscreened Ontario and Minnesota — almost entirely early-stage cases — but advanced disease and deaths were no less frequent. In Germany, around 1.5 million children were offered screening at about one year of age in six federal states. Screening did not reduce stage 4 disease or deaths from neuroblastoma; the investigators found substantial overdiagnosis, including tumours that would have regressed on their own. Three of 149 screen-detected children died of treatment complications. The trial concluded that general screening was not justified.
Prostate cancer
Prostate cancer is the classic case in adults. PSA screening does reduce prostate cancer deaths modestly in the large European trial, but it also finds many cancers that would never have threatened life. The trade-off is discussed in our article on the PSA screening controversy. The same logic applies to self-paid scans that search the whole body for abnormalities, as discussed in our review of whole-body MRI: more findings do not mean longer lives.
Why mortality is the right endpoint
All three biases inflate survival and detection rates. None of them affects mortality — the number of people who die of the disease per 100,000 population, regardless of when or whether they were diagnosed. If screening truly saves lives, deaths in a screened population must fall compared with an unscreened one. That is why the only convincing test of a screening programme is a randomised trial comparing deaths in people invited to screening with those not invited.
The difference is not theoretical. When Welch, Schwartz and Woloshin looked at the 20 most common solid tumours in the US, changes in five-year survival over time were entirely uncorrelated with changes in mortality. Rising survival rates tell you that diagnosis has moved earlier, not that patients live longer.
Ideally, trials should also report all-cause mortality, because harms of follow-up and treatment can offset a reduction in disease-specific deaths. In the mammography data summarised by the Harding Center from the Cochrane review, breast cancer deaths fell from about 5 to 4 per 1,000 women over about 11 years, but deaths from all cancers combined were 22 per 1,000 in both groups. Screening tests that have passed this test in randomised trials, such as colorectal cancer screening and low-dose CT for heavy smokers, earned their place by showing fewer deaths — not by showing better survival.
What this means for you
When a test or programme is promoted with survival rates or numbers of "cancers found early", ask:
- Has a randomised trial shown that this test reduces deaths from the disease — and ideally deaths overall?
- Out of 1,000 people like me who are screened, how many avoid death from the disease, and how many are overdiagnosed?
- What happens after a positive result — further tests, biopsies, surgery — and what are the risks?
- Is "early detection" being presented as proof of benefit on its own?
In Germany, many tests outside the statutory screening programmes are sold as self-pay "IGeL" services (individual health services). For these in particular, ask for mortality data before saying yes.
Bottom line
Longer survival and more early diagnoses do not prove that screening works; lead-time bias, length-time bias and overdiagnosis can produce both without saving a single life. Judge screening by reduced mortality in randomised trials, weighed against overdiagnosis and harms. Evidence: ★★★ High (for the existence and size of these biases).
References
- Gigerenzer G, Gaissmaier W, Kurz-Milcke E, Schwartz LM, Woloshin S. Helping doctors and patients make sense of health statistics. Psychol Sci Public Interest. 2007;8(2):53-96. doi:10.1111/j.1539-6053.2008.00033.x
- Welch HG, Schwartz LM, Woloshin S. Are increasing 5-year survival rates evidence of success against cancer? JAMA. 2000;283(22):2975-2978. doi:10.1001/jama.283.22.2975
- Ahn HS, Kim HJ, Welch HG. Korea's thyroid-cancer "epidemic" — screening and overdiagnosis. N Engl J Med. 2014;371:1765-1767.
- Schilling FH, Spix C, Berthold F, et al. Neuroblastoma screening at one year of age. N Engl J Med. 2002;346(14):1047-1053. doi:10.1056/NEJMoa012277
- Woods WG, et al. A population-based study of the usefulness of screening for neuroblastoma. Lancet. 1996;348:1682-1687.
- Gøtzsche PC, Jørgensen KJ. Screening for breast cancer with mammography. Cochrane Database Syst Rev. 2013;(6):CD001877.
This article is general information and not individual medical advice. Decisions about tests and treatment belong in a conversation with your own physician.