Specific guide to this web site for:
2. Medical Students
3. Science media writers
4. High School & College
1. Harvard led MI study
(J. of Amer. Coll.
4. Amer. J. of Cardio.
review of literature
6. Oat bran study
7. Pregnancy & Alcohol
8. Are Geminis
9. Columbia 'Miracle' Study
Limitations of Meta-Analyses
Large Randomized Clinical
Tale of Two Large
A statistical analysis in a major medical journal (Journal
of American College of Cardiology- JACC) which was so bad, that it may have led
to a subsequent improvement.
study in which this statistical problem occurred:
the Journal of American Cardiology (JACC), which is the
official journal for the society of board certified
cardiologists, there was an article with a major statistical
error. This involved a
study with the title of “Increased Myocardial Perfusion at
Rest and Diminished Perfusion Reserve in Patients With
Angiographic Normal Coronary Arteries” (Geltman EM, et al.
JACC 1990; 16:586-95). The
article illustrates a fundamental flaw in the use of
this statistical problem
way to make this fundamental statistical error more understandable is to note
the following analogy. A study examines two groups of people, one group from north of town (northerners) and
one group south of town (southerners). The northerners were then divided into
the tall northerners and the short northerners on the basis of height. The
researchers then inappropriately compared the tall northerner group to the
entire undivided southerner group.
Naturally, the tall northern subgroup who were
selected on the basis of being tall would on the average be taller than the
undivided total southern group.
Similarly, the tall northerner group would also be likely
to have a statistically greater value for a foot to waist measurement than the
average value for the undivided southerner group. However,
one could not say that because of this comparison, that there is a unique
subgroup that exists in the northerners of tall people that doesn’t exist in
the southerner group. This is fundamental error that the authors of this study
made. It was
perhaps less obvious because the measurements they used involved PET scanners
and cardiac measurements, but that doesn’t make this inappropriate statistical
analysis any less wrong than it is for the example given of people divided on
the basis of height.
types of statistical analysis are inherently invalid:
Analogy explaining the
statistical problem in the JACC (Journal of American Cardiology) article (click
Details of statistical problem in JACC article (click
See right column for
analogy of this statistical problem.
letter advocating an improvement in the Journal's review process was sent to
each of the editors:
This letter was sent to
each individual JACC editor
calling for a formal statistical review of each article prior
to publication to avoid comparable major statistical errors.
comments by one of JACC editors (click
here) Reply from one of the JACC editors
agreeing that this problem deserved further consideration.
improvement was later made in the journal's review process:
Announcement by JACC editors
that every manuscript that is strongly considered for
publication will be first reviewed for statistical issues.
analysis of statistical problem sent to the editors of JACC
influenced by this correspondence or not, ultimately, a change was made by JACC
and a statistical review was incorporated into their review process.