Five-year survival keeps rising. Nobody lives longer.

“Five-year survival” sounds like a fact about how long people live. It is really a fact about a stopwatch: the share of patients still alive five years after diagnosis. Move the moment of diagnosis and you move the number, without touching anyone’s lifespan — which is exactly what screening does.

Take the patient in the figure. Their cancer will end their life at 70 whatever anyone does. Left alone, it announces itself with symptoms at 67, and they “survive” three years — below the five-year bar, a statistic in the failure column. Now screen them, and the same tumour is spotted at 65. Nothing about the disease has changed and death still comes at 70, but the record now shows five years of survival. The patient has been converted, on paper, from a screening failure into a screening success. They have not gained a day. They have lost two years to knowing.

This is why comparing survival rates between countries, eras or hospitals is treacherous: whoever diagnoses earliest — not whoever treats best — wins the survival table.

That last point sank a famous political claim. In 2007 Rudy Giuliani’s campaign ran an advert boasting that a man’s chance of surviving prostate cancer was 82% in America but only 44% in England — proof, it said, of superior American medicine. The survival gap was real (about 98% versus 71% at five years). But the figure that actually counts, the death rate, was almost identical: roughly 26 American men per 100,000 died of prostate cancer against 27 in Britain. America wasn’t saving those men; its heavy PSA screening was simply diagnosing them years earlier, and often diagnosing cancers that would never have troubled them at all.

The second bias is subtler and, if anything, worse. A screen is a snapshot taken at intervals — every year or two. Whether it catches a tumour depends on how long that tumour sits in the window between “big enough to detect” and “bad enough to cause symptoms”. Slow, indolent cancers dawdle in that window for years, so a periodic scan is very likely to land on one. Fast, aggressive cancers tear through it in months and mostly surface as symptoms between scans.

So screening doesn’t sample cancers evenly — it preferentially harvests the slow, mild ones, precisely the cases with the best outlook to begin with. The screen-detected group then boasts wonderful survival, not because screening rescued anyone, but because it skimmed off the least dangerous disease and left the fast killers to the clinic. (This is the same length-biased sampling behind the inspection paradox — long things are easier to catch — only here the “long” thing is a tumour’s harmless spell.)

Push length-time bias to its limit and you reach the most troubling idea in screening: overdiagnosis. Some tumours grow so slowly — or not at all — that they would never have caused a symptom in the person’s remaining life. Screening finds them anyway. They are then treated, with surgery, radiation or worse, and every one is counted as a life saved. In truth nothing was saved; a healthy person was turned into a patient and given the risks of treatment for a disease that was never going to hurt them.

South Korea ran the natural experiment. After thyroid screening became cheap and routine in the late 1990s, the number of thyroid cancers diagnosed rose fifteen-fold in under two decades. A triumph, until you look at the death rate: it did not budge. Almost all the extra “cancers” were tiny papillary tumours that would have sat harmlessly for life, and tens of thousands of people had their thyroids removed for nothing. The survival statistics, meanwhile, looked spectacular — because a cancer that could never kill you is a cancer you are guaranteed to “survive”.

A test that finds harmless disease can only ever improve the survival rate. The person it “cured” was never in danger — but now carries a scar, a diagnosis and the risks of the treatment.

Ask for the death rate, not the survival rate. Survival counts from diagnosis and is corrupted by both biases; mortality — deaths per year in the whole population — is not. If screening truly works, fewer people die. If the boast is only about five-year survival, treat it as no evidence at all.

Distrust survival comparisons across places and times. A country, clinic or decade that screens more will always post better survival, even with identical medicine and identical deaths. That single confusion powered the prostate-cancer claim above.

Ask what happened to the number diagnosed. If diagnoses leap while deaths hold steady, you are almost certainly watching overdiagnosis, not a cure — the fingerprint South Korea left so clearly.

Trust randomised trials with a mortality endpoint. The only clean test is to randomise people to screening or not and count deaths years later. It is slow and expensive, which is why survival rates are quoted instead — they are available, flattering and, for judging screening, close to meaningless.

Don’t confuse this with the Will Rogers phenomenon. That one improves the averages of two groups by reclassifying patients between them; here, better detection inflates a single survival clock and cherry-picks the gentle cases. Cousins in the family of screening statistics that flatter without helping.