All 4 books by Edward Tufte now in
paperback editions, $100 for all 4
Visual Display of Quantitative Information
Beautiful EvidencePaper/printing = original clothbound books.
Only available through ET's Graphics Press:
catalog + shopping cart
All 4 clothbound books, autographed by the author $150
catalog + shopping cart
Edward Tufte e-books
Immediate download to any computer:
Visual and Statistical Thinking $2
The Cognitive Style of Powerpoint $2
Seeing Around + Feynman Diagrams $2
Data Analysis for Politics and Policy $2catalog + shopping cart
Edward Tufte one-day course,
Presenting Data and Information
Boston MA, October 23, 24, 25
Washington DC, November 6, 7
Bethesda MD, November 9
San Francisco CA, December 4, 5, 6
San Jose CA, December 8
Many new medical interventions look good when evaluated by survival time but fail when evaluated by mortality rates. See a fine article in The New York Times here, an article that would appear in no other newspaper.
Alvin Feinstein, my colleague at Yale, made the point for years that early diagnosis makes treatments look good because early diagnosis leads to longer survival times--but often with no improvement in mortality rates.
-- Edward Tufte
I think these two paragraphs should have been further up in the article:
Imagine two patients with lung cancer. Even if both die at age 70, a patient with cancer diagnosed by spiral CT screening at age 59 has a longer survival than one with cancer diagnosed because of symptoms (cough, weight loss and so on) at age 67. The first patient survives 11 years; the second 3 years. But both died at the same age. Survival is increased, but mortality is the same.
A second source of distortion results from overdiagnosis, when screening finds cancers that were never destined to progress and cause death. Overdiagnosis bias can also drastically inflate survival statistics, even if mortality is unchanged.
The author follows these excellent illustrations by saying "To understand why, you need to understand the definition of the two statistics" but never provides the definitions. If NYT provided definition links to Wikipedia, instead of their own search engine, this omission could have been nicely resolved. The reader's state of awareness goes from vague to clear, and back to vague. Instead of using the PGP model, the author has used a GPG model.[PGP: Particular—General—Particular. Frederick Mosteller, "Classroom and Platform Performance", The American Statistician, February 1980, Vol. 34, No.1, pp. 11-17.]
So many benign, slow-growing adrenal tumors are found on CT that some have taken to calling them incidentalomas. There is a similar observation about a form of breast cancer, ductal carcinoma in situ: many die with it, few die of it.
-- Niels Olson (email)
If a treatment is only symptomatic then early or late diagnosis may not matter at all. If the therapy treats disease progression, then early treatment may make a much larger difference on the survival time. Communicating this concept is often done with graphs that show a decline in status and the "catch-up" effect with a symptomatic therapy.
Cancer therapies often use biomarkers to link their drug to something critical (like survival time), but diseases of the central nervous system rarely have these markers.
-- Chris Pounds (email)
For an interesting analysis of non-cohort survival measures and distortions caused by changes in diagnosis numbers, see Samuel H. Preston, "Relations among Standard Epidemiologic Measures in a Population," American Journal of Epidemiology, 126 (1987): 336-45.
-- Sherman Dorn (email)
Can a Kindly Contributor find a link to this article?
RELATIONS AMONG STANDARD EPIDEMIOLOGIC MEASURES IN A POPULATION
SAMUEL H. PRESTON Population Studies Center, 3718 Locust Wallc/CR, University of Pennsylvania, Philadelphia, PA 19104. (Send reprint requests to Dr. Samuel H.Preston at this address.)
Recent developments In population mathematics apply to measurement issues in epidemiology. In particular, they demonstrate explicitly the relations that prevail among incidence, prevalence, case-fatality, mortality, and duration of illness in a population at a moment in time, rather than in a cohort of persons followed through time (or in a population artificially assumed to be stationary). They indicate explicitly how certain common indicators such as the ratio of deaths to new cases should be interpreted. They also suggest possible new strategies for estimating certain measures, but these would require some reorientation of current approaches to measurement.
-- Troy Goodson (email)