Knowledge Engine

How cancer survival can improve in every stage while no patient lives longer

·no read data

The idea

Will Rogers, the American humorist, is credited with a joke about the Dust Bowl migration: when Oklahomans moved to California, they raised the average intelligence of both states. The joke works because of a quiet bit of arithmetic. If you move someone who is below the average of the group they leave but above the average of the group they join, both averages go up, and nobody has changed at all.

In 1985 A. R. Feinstein, D. M. Sosin and C. K. Wells showed that this arithmetic was distorting cancer statistics. Writing in the New England Journal of Medicine, they compared lung cancer patients first treated in 1977 with a cohortcohort A group of people defined by a shared starting point, such as being diagnosed in the same years, and followed over time. treated at the same institutions between 1953 and 1964. The 1977 patients had better six-month survival overall and within each of the three main stages of the TNM system, which grades a cancer by tumour size, lymph node involvement and spread.

The catch was the scans. The 1977 patients had been through a battery of new imaging tests, which found metastasesmetastasis Cancer that has spread from where it started to another part of the body. that earlier doctors would never have seen. A patient who once would have been placed in a “good” stage was now placed in a “bad” one. These movers had a worse outlook than the patients who stayed in the good stage, so removing them raised that stage’s survival. They had a better outlook than the patients already in the bad stage, so adding them raised that stage’s survival too. In the authors’ words, survival rose in each group “without any change in individual outcomes.” The authors called this stage migrationstage migration When patients are moved into a different disease stage because of new tests or rules, rather than because their illness changed. and named it the Will Rogers phenomenon.

Why it might change a decision

The effect matters whenever survival is compared within stages across time, hospitals or countries, especially when one side uses more sensitive tests. A hospital with better scanners can look better at every stage than one with worse scanners, even if the two treat patients equally well.

That gives a specific, testable check. If stage-specific survival has improved, look at survival for all patients combined, regardless of stage. Moving people between categories cannot change the total. If every stage improves while the overall figure stays flat, migration is the leading suspect. If the overall figure improves too, something real may have happened, though earlier detection can raise it on its own by starting the clock sooner.

How solid is this?

The evidence here is two observational studies, not trials, but the core claim is unusual: it is arithmetic, not a hypothesis about biology. What the studies show is that the arithmetic operates in real data at a size worth caring about.

In the 1985 study, the decisive test was a reclassification. Using only the older kinds of diagnostic data for both cohorts, the authors found that the recent patients also entered the statistics at a slightly more favourable point in their illness, which they called a zero-time shift. When patients were grouped instead by symptom stages, which imaging does not change, the two cohorts had similar survival. The abstract does not report the stage-by-stage survival percentages, so the size of the shift has to be read from the full paper.

A 2005 study in the Journal of the National Cancer Institute, by P. C. Albertsen, J. A. Hanley and colleagues, found the same pattern from a different cause. Prostate cancer is graded by the Gleason scoreGleason score A pathologist's rating of how abnormal prostate cancer cells look under the microscope, with higher numbers meaning more aggressive cancer., a pathologist’s rating of how abnormal the tissue looks, and the way pathologists apply it changed between the early 1990s and the 2000s. The team took 1,858 men aged 75 or under, diagnosed in 1990 to 1992 and recorded in the Connecticut Tumor Registry, and had a pathologist who did not know the original scores reread the biopsy slides in 2002 to 2004. The mean score rose from 5.95 to 6.8. Standardised by the new grades, mortality was 1.50 deaths per 100 person-yearsperson-years A count that adds up how many years each person in a study was followed, so that people watched for different lengths of time can be combined., about 28 percent lower than the 2.08 implied by the old grades, though these were the same men with the same outcomes. The apparent improvement appeared in every Gleason category.

Both studies have limits. The prostate rereading relied on one pathologist, so the size of the drift reflects one person’s modern standard. The lung study compared cohorts decades apart, and treatment did change in that time; the symptom-stage comparison is what argues against treatment as the explanation. Neither study shows that every reported survival gain is an artefact, and the popular shorthand that “survival statistics are meaningless” goes well beyond them. They show that stage-specific comparisons can mislead when the staging itself has changed.

Sources

  • Feinstein AR, Sosin DM, Wells CK. “The Will Rogers phenomenon. Stage migration and new diagnostic techniques as a source of misleading statistics for survival in cancer.” New England Journal of Medicine, 1985;312(25):1604-1608. https://doi.org/10.1056/NEJM198506203122504 (PubMed 4000199: https://pubmed.ncbi.nlm.nih.gov/4000199/)
  • Albertsen PC, Hanley JA, Barrows GH, Penson DF, Kowalczyk PD, Sanders MM, Fine J. “Prostate cancer and the Will Rogers phenomenon.” Journal of the National Cancer Institute, 2005;97(17):1248-1253. https://doi.org/10.1093/jnci/dji248 (PubMed 16145045: https://pubmed.ncbi.nlm.nih.gov/16145045/)

Check yourself

Answers are checked in your browser. Nothing is recorded or sent anywhere.

  1. In Feinstein's lung cancer comparison, why did survival rise within each stage for the 1977 cohort?

  2. What happened when Albertsen and colleagues reread old prostate biopsy slides with modern grading?

  3. Which check most directly distinguishes a real improvement from stage migration?

  4. A reader says: 'Since both papers are observational, the Will Rogers phenomenon is unproven and might not exist.' How well does this match the evidence?

Questions without answers

  1. Hospital league tables often report survival by stage. What extra figure would you want published alongside them, and why?
  2. Can you think of a non-medical case, such as school grades or credit ratings, where tightening a category boundary could make every category look better?

Back to the archive