ENFR

Why survival misleads: lead-time bias and overdiagnosis

Screening moves diagnosis earlier and finds cancers that would never have mattered. Both push five-year survival up. Only one of them, sometimes, keeps anyone alive.

In a hurry

Five-year survival is measured from the day of diagnosis. Screening changes that day, and who gets a diagnosis at all, so it inflates survival in two ways that have nothing to do with anyone living longer: it starts the clock earlier (lead-time bias), and it labels people with slow cancers that would never have surfaced (overdiagnosis). Screening does sometimes save a life - when it catches a dangerous-but-treatable cancer in time - but that real benefit shows up in mortality, not in survival. Below you can screen a whole cohort and watch the two numbers come apart: survival soaring, deaths barely moving.

Three kinds of cancer in the barnyard

Screening is like trying to fence in animals of very different speeds, and the outcome depends entirely on which animal you caught. This picture, borrowed from H. Gilbert Welch, is the key to the whole page.

Turtles - going nowhere

So slow they would never cause symptoms or death. Screening finds them easily, because they sit still the longest. Every turtle caught is overdiagnosis: a diagnosis, a treatment, and its harms, with no death to prevent.

Rabbits - catchable

Dangerous, but slow enough that catching them before symptoms and treating actually prevents the death. This is the whole point of screening - and it is the only group where a life is genuinely saved.

Birds - already flown

So fast they have spread before any screen could catch them. Finding one early changes the date on the chart but not the date of death. Pure lead time: survival looks better, the patient does not live a day longer.

Try it: screen a whole cohort

Twelve patients, each with a tumour that appears at time zero. Every lifeline runs left to right over fifteen years and ends at a cross: red for a cancer death, grey for death from another cause. The small dot is when the tumour becomes detectable, the hollow circle when it would cause symptoms on its own, and the triangle is the day of diagnosis. Turn screening up and watch the diagnosis triangles jump left, the turtles light up as overdiagnosed, and a few rabbits have their red cross turned grey - a death prevented. Read the numbers above the chart as you go.

5-year survival
cancer deaths
overdiagnosed
lives saved
cancers diagnosed
tumour detectable would cause symptoms diagnosis cancer death death, other cause
Rows are grouped T (turtles), R (rabbits), B (birds). Cross colour is the cause of death: red cancer, grey other. As screening rises, five-year survival climbs steeply - fed by earlier diagnosis (birds and rabbits) and by overdiagnosed turtles who are guaranteed survivors - while cancer deaths fall only by the handful of rabbits caught in time. That gap, survival racing ahead of mortality, is the whole illusion.

What surprises people

Survival and mortality can point in opposite directions. Survival is measured among the diagnosed; mortality among everyone. Because screening changes who is diagnosed and when, survival can rise a lot while mortality barely moves - which is exactly what a useless screening programme also produces.
Overdiagnosis is not a false positive. The turtle is a real cancer under the microscope. That is why a more accurate test cannot fix it - a perfect test would find every harmless tumour that is genuinely there.
The only honest verdict is a randomized trial counting deaths. Invite half a population to screening, leave half alone, and compare how many die of the cancer - not how long the diagnosed survive. Every serious screening debate, mammography and PSA included, is fought on mortality.
More screening is not always more benefit. Past a point, extra intensity mostly adds turtles - overdiagnosis and treatment harm - while the rabbits are already being caught. The dose of screening has its own therapeutic window.
None of this means screening is useless. Rabbits are real, and catching them saves lives. It means you cannot read that benefit off a survival statistic - you have to look at whether fewer people died.

This exact confusion was once a campaign talking point - 82% US survival against 44% British - sold as proof of better care while the death rates were nearly the same:

The 82% survival myth

Sources

  • Welch, H. G., & Black, W. C. (2010). Overdiagnosis in cancer. Journal of the National Cancer Institute, 102(9), 605–613. https://doi.org/10.1093/jnci/djq099
  • Welch, H. G. (2011). Overdiagnosed: Making People Sick in the Pursuit of Health. Beacon Press. (Origin of the bird / rabbit / turtle metaphor.)
  • Gigerenzer, G., et al. (2007). Helping Doctors and Patients Make Sense of Health Statistics. Psychological science in the public interest, 8(2), 53–96. https://doi.org/10.1111/j.1539-6053.2008.00033.x

Common questions

What is lead-time bias?

Screening moves the moment of diagnosis earlier without changing when the patient dies. Five-year survival is counted from diagnosis, so starting that clock sooner inflates survival even if no one lives a day longer. The gain is in the measurement, not in the outcome.

What is overdiagnosis?

Overdiagnosis is finding a cancer that would never have caused symptoms or death in the person's lifetime. Screening is best at catching these slow, harmless tumours, and every one found is counted as a survivor, which pushes survival up while helping no one.

Why is mortality a better measure than survival for screening?

Survival is measured from diagnosis and is inflated by both lead-time bias and overdiagnosis. Mortality, deaths per population over time, is anchored to the whole group and is not moved by when or whether you label someone, so it reflects whether screening actually keeps people alive.

Stat Exam Pro turns screening, survival and the rest of medical statistics into exam-style questions with worked answers, in English and French.

Get it on iPhone