I have a PhD in Social Research from the University of Michigan, and
worked for ten years as a biostatistician/ data analyst/ data manager at
Memorial Sloan Kettering Cancer Center in New York City. Our work used the
same proportional hazards general linear models Cox regressions that this
study seems to use. I did not look much at smoking issues but rather other
cancer inquiries, although one study of cervical esophageal cancer had
100% (small Ns) of decedents having been smokers. Ouch. I am a believer.
I then worked similarly for a few years in FDA submissions in one or
another US pharmaceutical company.
I have long been skeptical of ETS as a cause of mortality. I first
became a skeptic when I saw that an early report, early 1990s, probably a
meta analysis, moved the goalposts, from a p level of .05 to a p level of
.15. This was buried in the background of the report. Nothing since then
has been persuasive. Junk science.
I continued as a skeptic when I started to see that who was against
ETS: of course the financial interests, via litigation; and who in favor
of ETS, financial interests just as great, via tobacco-interest funding.
In the FDA world, all analysis is done double blinded, and
occasionally even triple blinded (the researchers don't know which is
brand x, and this analysis may even be farmed out, separate from the data
collection efforts). This is the gold standard, or the platinum iridium
standard for research, and from my FDA days, no study without double blind
has merit. That said, we must work with what analysis we have.
But this double blind criterion would silence many of the conflict of
interest complaints. It is indeed possible, I submit, to disguise the data
and the names of the variables, perhaps toothpaste preference in
toothpaste using spouses, and subcontract the work to a third party. When
I did physician-based outcome research in cancer intervention (which
doctors' patients had better survival?, I analyzed based on numbers, not
names, for example.) Note, there are differences between doctors, which we
then analyzed in terms of patient characteristics and surgical approaches.
I wish the authors had devoted more time to critiques of the meta
analysis data, the inappropriate combination of samples, and the differing
assumptions in combined studies. Alas, but this was not their report;
their report was their own data.
There are also important points to be made on percentage of followup.
The authors come in at 50%, over a lifetime cohort study. While at Sloan
Kettering, part of my job was followup on our cancer inpatients, to
monitor health, morbidity and mortality. Followup is a complicated
process, and cancer facilities were, in the 1980s, rated by percentage of
a mandated annual followup successfully done. We maintained a data file
called the Cancer Registry. Mortality was included as a followup end
point. Our methodology was essentially active letter writing, and data
mining within the then-new data integration capabilities in the emerging
PC world; I created such an internal network via 'sneaker-net,' and we
achieved 90% followup of our active denominator of former cancer
inpatients. Barely and exhaustingly. 50% is pretty good, and the authors
seem to have addressed self-selection or preferential access, but this is
a larger question. Obviously, a dead population over forty years will be
lost to followup more than a living population, but the authors use the
standard national death indices, as did we.
All this said, the data are statistically and intuitively overpoweringly
persuasive. It has always been unreasonable, that while smokers' pathology
is dose (smoking) related, with inhaled dosages themselves varying by
dosage (cigarettes smoked, per day and per year) in their pathogenesis,
the second hand inhalation is trivial compared to primary inhalation, and
yet there has been no inhaled dosage relationships reported in ETS
analysis.
Thus the true measure of 'exposure' is not the hours of co-presence
in the home with a smoker, but some measure of inhalation. This should be
measurable by analysis of exhaled air samples, or more modestly, by
comparing ambient air pathogens, in the home or elsewhere, with lung
exhalation from smokers themselves. The falloffs in presence of pathogens
would be in the order of hundreds or thousands, and these orders of
magnitude or exactly the orders of magnitude the EPA uses to discuss
environmental contaminant risks, orders of magnitude.
As for the criticisms that codomiciling with ever-smokers is
irrelevant, because of external ETS, this should randomize across the two
groups, and obscure differences, such that real differences between
codomiciling with smokers, vs nonsmokers, is even sharper.
The one writer who expresses concerns that twenty minutes a day of
ETS on mass transit is enough to cause cancer, is not to be taken
seriously. One doubts just what disease entity is so pathogenic as to be
mortal, with twenty minutes a day of exposure. Perhaps SARS, perhaps the
1920s influenza, perhaps bubonic plague. Cancer death is a cumulative not
immediate process. Were this 20-minutes and dead relationship to be so,
the world of smokers' neighbors would long have been depopulated. This is
clearly not so.
Anyway, the questions now move on to peer review, and data
reanalysis, and we may hope that science will accept the internal
discipline of inconvenient discrepant data and revised conclusions. BJM is
to be congratulated for providing this forum.
Rapid Response:
some thoughts
I have a PhD in Social Research from the University of Michigan, and
worked for ten years as a biostatistician/ data analyst/ data manager at
Memorial Sloan Kettering Cancer Center in New York City. Our work used the
same proportional hazards general linear models Cox regressions that this
study seems to use. I did not look much at smoking issues but rather other
cancer inquiries, although one study of cervical esophageal cancer had
100% (small Ns) of decedents having been smokers. Ouch. I am a believer.
I then worked similarly for a few years in FDA submissions in one or
another US pharmaceutical company.
I have long been skeptical of ETS as a cause of mortality. I first
became a skeptic when I saw that an early report, early 1990s, probably a
meta analysis, moved the goalposts, from a p level of .05 to a p level of
.15. This was buried in the background of the report. Nothing since then
has been persuasive. Junk science.
I continued as a skeptic when I started to see that who was against
ETS: of course the financial interests, via litigation; and who in favor
of ETS, financial interests just as great, via tobacco-interest funding.
In the FDA world, all analysis is done double blinded, and
occasionally even triple blinded (the researchers don't know which is
brand x, and this analysis may even be farmed out, separate from the data
collection efforts). This is the gold standard, or the platinum iridium
standard for research, and from my FDA days, no study without double blind
has merit. That said, we must work with what analysis we have.
But this double blind criterion would silence many of the conflict of
interest complaints. It is indeed possible, I submit, to disguise the data
and the names of the variables, perhaps toothpaste preference in
toothpaste using spouses, and subcontract the work to a third party. When
I did physician-based outcome research in cancer intervention (which
doctors' patients had better survival?, I analyzed based on numbers, not
names, for example.) Note, there are differences between doctors, which we
then analyzed in terms of patient characteristics and surgical approaches.
I wish the authors had devoted more time to critiques of the meta
analysis data, the inappropriate combination of samples, and the differing
assumptions in combined studies. Alas, but this was not their report;
their report was their own data.
There are also important points to be made on percentage of followup.
The authors come in at 50%, over a lifetime cohort study. While at Sloan
Kettering, part of my job was followup on our cancer inpatients, to
monitor health, morbidity and mortality. Followup is a complicated
process, and cancer facilities were, in the 1980s, rated by percentage of
a mandated annual followup successfully done. We maintained a data file
called the Cancer Registry. Mortality was included as a followup end
point. Our methodology was essentially active letter writing, and data
mining within the then-new data integration capabilities in the emerging
PC world; I created such an internal network via 'sneaker-net,' and we
achieved 90% followup of our active denominator of former cancer
inpatients. Barely and exhaustingly. 50% is pretty good, and the authors
seem to have addressed self-selection or preferential access, but this is
a larger question. Obviously, a dead population over forty years will be
lost to followup more than a living population, but the authors use the
standard national death indices, as did we.
All this said, the data are statistically and intuitively overpoweringly
persuasive. It has always been unreasonable, that while smokers' pathology
is dose (smoking) related, with inhaled dosages themselves varying by
dosage (cigarettes smoked, per day and per year) in their pathogenesis,
the second hand inhalation is trivial compared to primary inhalation, and
yet there has been no inhaled dosage relationships reported in ETS
analysis.
Thus the true measure of 'exposure' is not the hours of co-presence
in the home with a smoker, but some measure of inhalation. This should be
measurable by analysis of exhaled air samples, or more modestly, by
comparing ambient air pathogens, in the home or elsewhere, with lung
exhalation from smokers themselves. The falloffs in presence of pathogens
would be in the order of hundreds or thousands, and these orders of
magnitude or exactly the orders of magnitude the EPA uses to discuss
environmental contaminant risks, orders of magnitude.
As for the criticisms that codomiciling with ever-smokers is
irrelevant, because of external ETS, this should randomize across the two
groups, and obscure differences, such that real differences between
codomiciling with smokers, vs nonsmokers, is even sharper.
The one writer who expresses concerns that twenty minutes a day of
ETS on mass transit is enough to cause cancer, is not to be taken
seriously. One doubts just what disease entity is so pathogenic as to be
mortal, with twenty minutes a day of exposure. Perhaps SARS, perhaps the
1920s influenza, perhaps bubonic plague. Cancer death is a cumulative not
immediate process. Were this 20-minutes and dead relationship to be so,
the world of smokers' neighbors would long have been depopulated. This is
clearly not so.
Anyway, the questions now move on to peer review, and data
reanalysis, and we may hope that science will accept the internal
discipline of inconvenient discrepant data and revised conclusions. BJM is
to be congratulated for providing this forum.
Competing interests:
general skeptic over ETS
Competing interests: No competing interests