Confoundle · a reasoning trap

Lead-time bias

Survival is counted from the day you are diagnosed. So a test that finds a disease earlier automatically makes survival look longer, even if it changes nothing about when the disease kills you. You simply spend more of your life as a patient. That is why a screening programme can raise five-year survival dramatically while exactly the same number of people die. The number that cannot be gamed is deaths in the whole population, not survival among the diagnosed.

The rule

Finding a disease earlier stretches survival measured from diagnosis even when it postpones death by not a single day.

What it looks like

Screened patients survive five years after diagnosis. Unscreened ones survive two.The same cancer, growing at the same speed, treated the same way. One person had a scan that found it early. The other went to a doctor years later, when the first symptom appeared. Survival is counted from the day of diagnosis, which is how survival is almost always counted.

Both died on exactly the same day.Screening did not postpone anything. It moved the diagnosis three years earlier, so this person spent three more years knowing they had cancer. Counted from diagnosis that reads as three extra years of survival. Put both lives on the same calendar and they end at the same instant:

Why it works

Survival statistics start their clock on the day of diagnosis. That day is not a fact about the disease, it is a fact about when somebody looked. Move the looking earlier and you add time to the front of the measurement while changing nothing at the back. Everyone diagnosed early is guaranteed to clear the five-year mark more often, because they were handed a head start. Two other effects push the same way. A screening programme catches slow-growing disease far more often than fast-growing disease, simply because slow disease sits there longer waiting to be found, and slow disease has a better outlook anyway. And a sensitive enough test finds harmless abnormalities that would never have caused trouble, which are then counted as cured cancers. All three flatter survival without saving anyone. The only honest test is to take a whole population, invite half of it to be screened, and count deaths in everyone from the day of the invitation. Screening programmes that pass that test exist, which is exactly why the test is worth insisting on.

Source

Welch HG, Schwartz LM, Woloshin S. Are increasing 5-year survival rates evidence of success against cancer? JAMA. 2000;283(22):2975-2978.

See if it fools you

This page gives the answer away. The puzzle version shows you the same figures first and asks you to commit before the reveal.

Play this one