I don't see why there HAS to be a fallacy in my logic.  There could
very well be a fallacy in the study.

Here's an exact quote from the abstract online:

" There were no statistical differences in OS, CSS, DM, or FFBF among men whose EBRT began
< 3, 3-6, 6-9, or > 9 months after diagnosis. This was also true at the median TTT of 3.1
months. A subgroup analysis was performed in which patients were stratified into low-,
intermediate- and high-risk groups based on pretreatment PSA, Gleason score and AJCC
T-stage. FFBF, and DM were calculated above and below the median TTT of 3.1 months. In
this analysis, there was no statistically significant difference in FFBF or DM within the
risk groups."
(http://www3.interscience.wiley.com/cgi-bin/abstract/110514439/ABSTRACT)

OS = overall survival
CSS = cause specific survival
DM = distant metastasis
FFBF = freedom from biochemical failure

Notice the "> 9 months" statement.  It doesn't say how much greater.

They need to account for this - at least with a hypothesis.  How
many months _would_ be required to see a difference?

Consider this proposition: In any case of fatal prostate cancer, there is
some point in time when a distant metastasis establishes itself
and the cancer turns fatal.  Before that point, the cancer may be
cured by radiation.  After that point it cannot.

From my layman's perspective, that proposition makes sense
to me.  If it's not, why not?  If it is right, then won't that point, for at
least _some_ patients, likely fall within the time period studied?

Here's another proposition: People are diagnosed at different
stages of disease.  One guy is diagnosed in plenty of time, another
is not diagnosed until it's too late.  Some patients are going to be
diagnosed somewhere right between those two points in time -
the point on one side of which it is not too late, and on the other side
of which it is.  Some guys are going to be diagnosed one month
before that point.

I.P., your explanation of cumulative change requiring longer than
3 months - or actually longer than "> 9 months" seems to me
possible, but not plausible.  It would require that all changes in the
development of the cancer be extraordinarily slow, and that the
"point" in time when a metastasis establishes itself be longer
than 3 or 9 months.

I have a friend who had a "normal" PSA test.  He didn't know the
number.  Two years later he had another one that showed a PSA of
245.  Two months after that it was 300.  Clearly he's got metastatic
cancer.  He probably didn't have it two years before.  It's easy to
imagine there was a nine month period during which his
cancer escaped the prostate and exploded into his body.

An alternative explanation might be that his cancer was metastatic
but slow growing at the time of his normal PSA reading - in which
case he was already beyond help.  But whatever the actual case
is, it seems to me that explanation is called for.  Putting out data
like this with no explanation for why it might be accurate seems to
me to be really irresponsible.

I'm not saying that what Andrews et. al. reported is false.  It might
be true.  If so there are many possible explanations - of which the
possibility that radiation does no good - is clearly one of them.  But
I'd like to see the data checked, the math checked, the study
repeated, and the objections raised above answered, before I'm
ready to accept it.

   Alan