"R. Stegers" <[EMAIL PROTECTED]> wrote in message news:<a3kc0p$ae3$[EMAIL PROTECTED]>...
> I'm trying to understand some medical paper and they used both Log-rank and
> Mantel-Haenszel. Could anybody briefly explain what is measured by these
> specific tests? Why are they used (in general) en what a calculated value
> stands for.
>
> Thanks in advance,
>
> Ruud
you will often see the Mantel-Haenszel applied in epidemiological
studies where the investigators are testing for an asociation between
some form of "exposure" and a disease state, while controlling for one
of more other variables. You might see a case-control study in which
a group with a disease (some form of cancer, for example) and a
control group are studied in terms of their exposure to an agent.
often that agent may be considered toxic, but it could also be
considered protective, like fruits and vegetables. so we might think
the risk of cancer in a group exposed to fruits and vegetables would
be less than that in a group not exposed. however, since this is not
a randomized
study, one would want to control for other variables like smoking,
exercise, etc. their test statistic follows a chi-square.
in survival analysis, you will see the terms Mantel-Haenszel statistic
and logrank statistic used interchangeably. here a survival curve is
estimated, often using the kaplan-meier technique, for each of several
strata. for each stratum (perhaps two different drug treatments), the
number of units exhibiting the outcome of interest is determined for a
given unit of time. given the assumptions, the number expected in a
given time interval can also be calculated. again the chi-square
distribution is used to compare the number with the outcome to the
number expected. this is NOT the same expected frequency as in the
Pearson chi-square.
you said "briefly" but here is a lot more to this topic area.
hopefully this gives you a feeling for what is involved. there are a
bunch of books, try Parmar and Machin: Survival Analysis, Wiley 1995.
JJ Diamond
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at
http://jse.stat.ncsu.edu/
=================================================================
