The only study that I can find that measures the effect of class of
travel on the chance of getting a DVT is this on (all the other
articles seem to refer back to it):

S Afr Med J. 2003 Jul;93(7):522-8.
The BEST study--a prospective study to compare business class versus
economy class air travel as a cause of thrombosis.

"
Only 434 subjects had a full venous duplex scan performed. None had
ultrasonic evidence of venous thrombosis. Nine passengers tested at
departure had elevated D-dimer levels and these volunteers were
excluded from further study. Seventy-four of the 899 passengers had
raised D-dimers on arrival. Twenty-two of 180 business class
passengers (12%) developed elevated D-dimers compared with 52 of 719
economy class passengers (7%). There was no significant association
between elevation of D-dimers and the class flown (odds ratio (OR)
0.61, p = 0.109).
"

I'd imagine that a huge amount of money, potentially, rests on this
evidence, so I'm surprised that this is the only study. I haven't got
Dr Jacobson's e-mail address at Wits, nor the whole text of the study
yet, but I'll follow these up to understand more detail and if there
was any special funding.

I find something curious about the above, though, maybe somebody with
some knowledge of statistics can help explain it.

1. If all air passengers were the same, and there was no bias caused
by the class of travel, then you'd expect, I'd have thought, to find
the same raised D-dimers (the proxy for potential DVT in the study) in
both populations. To me, finding 14% in on population and 7% in the
other would suggest that the first population was twice as likely to
suffer the effect.

2. Clearly the size of study is important. So, though the whole study
includes nearly 900 passengers, the study only examines 180 business
class passengers. So, these are less likely to be representative than
those not in business class.

3. Isn't it also likely that those flying in Business class will have
other characteristics that differ that might be significant in their
risk of DVT? Shouldn't these factors be exluded before a comparison is
made?

4. How, then, do they come to the conclusion that there is no effect?

A simple minded reading of the conclusion suggests to me that Business
Class might be more dangerous. However, if Business Class passengers
are older, fatter and maler than economy class passengers, all of
these being predisposing factors (some might even have been in
Business Class specifically because they were higher risk passengers
for other reasons) might, when excluded, show that Business Class is,
in fact, much safer.

How can this 2003 study be the only thing on which we can base our
conclusions?

Is it possible that airlines have tested this themselves (I'd find it
odd if this had not occurred to any airline) and decided not to
publish the results?

After all, if business class was much safer, then there would be
considerable pressure to increase the space in economy, rather than
put these passengers at a health risk - the conclusion of the study
removes this pressure. Alternatively, if it is actually the case that
business class is more dangerous, then fewer people would pay three
times the price for it - at least on health grounds.

I've got the full study now and it raises more questions than it asks.
The characteristics of the passengers in the two classes are below and
they seem to offer plenty of scope for influencing the results.

There is a greater proportion of taller, heavier masculine people in
Business class, and people in that class are older. Those in business
class also drank more though that may be connected with the other
characteristics.

There are more smokers in Economy - though smoking appears not to be
strongly connected to DVT.

It might also be significant that the study was sponsored by SAA and
the South African airports authority.

Unfortunately this study also excluded all high risk passengers, so we
don't know if they fare better in business class or not.

Characteristics Business class Economy class
Number 180 719
Age
Median (years) 47 43
Range (years) 28 - 78 18 - 81
Sex
Male 119 368
Female 61 351
Median height (cm) 176 173
Range (cm) 130 - 201 120 - 210
Median weight (kg) 82 76
Range (kg) 49 - 126 44 - 168
Median fluid intake (l) 1.0 1.0
Range (l) 0 - 3.25 0- 5.00
Mean alcohol intake (l) 0.250 0.175
Range (l) 0 - 2.38 0 - 1.5
Exercise (yes/no) 10/170 41/678
Percentage (yes) 6 6
Smokers (yes/no) 25/155 121/598
Percentage (yes) 14 17
Passengers (N (%))
Using aspirin 22 (12.2) 89 (12.4)
Using anti-inflammatories 4 (2.2) 26 (3.6)
Using sleeping tablets 24 (13.3) 94 (13.1)
Using OCs 9 (5) 72 (10)
Using HRT 12 (6.7) 71 (9.9)
Ultrasound (N %))
Immediate 77 (42.8) 310 (43.1)
Within 7 days 9 (5) 38 (5.3)

"Peter Brooks"  wrote in message
news:f26bce74-9036-46dd-9460-6ac9fb20b16d@a9g2000prl.googlegroups.com...
> The only study that I can find that measures the effect of class of
> travel on the chance of getting a DVT is this on (all the other
> articles seem to refer back to it):
>
> S Afr Med J. 2003 Jul;93(7):522-8.
> The BEST study--a prospective study to compare business class versus
> economy class air travel as a cause of thrombosis.
>
> "
> Only 434 subjects had a full venous duplex scan performed. None had
> ultrasonic evidence of venous thrombosis. Nine passengers tested at
> departure had elevated D-dimer levels and these volunteers were
> excluded from further study. Seventy-four of the 899 passengers had
> raised D-dimers on arrival. Twenty-two of 180 business class
> passengers (12%) developed elevated D-dimers compared with 52 of 719
> economy class passengers (7%). There was no significant association
> between elevation of D-dimers and the class flown (odds ratio (OR)
> 0.61, p = 0.109).
> "
>
> I'd imagine that a huge amount of money, potentially, rests on this
> evidence, so I'm surprised that this is the only study. I haven't got
> Dr Jacobson's e-mail address at Wits, nor the whole text of the study
> yet, but I'll follow these up to understand more detail and if there
> was any special funding.
>
> I find something curious about the above, though, maybe somebody with
> some knowledge of statistics can help explain it.
>
> 1. If all air passengers were the same, and there was no bias caused
> by the class of travel, then you'd expect, I'd have thought, to find
> the same raised D-dimers (the proxy for potential DVT in the study) in
> both populations. To me, finding 14% in on population and 7% in the
> other would suggest that the first population was twice as likely to
> suffer the effect.
>
> 2. Clearly the size of study is important. So, though the whole study
> includes nearly 900 passengers, the study only examines 180 business
> class passengers. So, these are less likely to be representative than
> those not in business class.
>
> 3. Isn't it also likely that those flying in Business class will have
> other characteristics that differ that might be significant in their
> risk of DVT? Shouldn't these factors be exluded before a comparison is
> made?
>
> 4. How, then, do they come to the conclusion that there is no effect?

Their conclusion is "There was no significant association between elevation
of D-dimers and the class flown (odds ratio (OR)> 0.61, p = 0.109)", which
means that, according to their calculations, the probability of the apparent
association being due to nothing more than chance is 10.9%; by convention, a 
result is only regarded as statistically significant if this probability is 
less than 5%.

It is not clear to me how they reached this conclusion.  According to the
figures given, the odds for business class passengers to have raised
D-dimers on arrival were found to be 22/158 and for economy class passengers
52/667.  The ratio between these is 1.79 or 0.56, depending on which way
round you take them, and  calculating the 95% confidence intervals as
described, for example, here
http://www.bmj.com/cgi/content/full/320/7247/1468 , gives intervals of
1.05-3.03 and 0.33-0.95 respectively.  Since these intervals do not include
the value 1, that seems to indicate p < 0.05.  Does the full text of the
study give any more information on how they arrived at their stated figures
for the odds ratio and p?

On May 25, 1:55 am, "Philip"  wrote:
> "Peter Brooks"  wrote in message
>
> news:f26bce74-9036-46dd-9460-6ac9fb20b16d@a9g2000prl.googlegroups.com...
>
>
>
> > The only study that I can find that measures the effect of class of
> > travel on the chance of getting a DVT is this on (all the other
> > articles seem to refer back to it):
>
> > S Afr Med J. 2003 Jul;93(7):522-8.
> > The BEST study--a prospective study to compare business class versus
> > economy class air travel as a cause of thrombosis.
>
> > "
> > Only 434 subjects had a full venous duplex scan performed. None had
> > ultrasonic evidence of venous thrombosis. Nine passengers tested at
> > departure had elevated D-dimer levels and these volunteers were
> > excluded from further study. Seventy-four of the 899 passengers had
> > raised D-dimers on arrival. Twenty-two of 180 business class
> > passengers (12%) developed elevated D-dimers compared with 52 of 719
> > economy class passengers (7%). There was no significant association
> > between elevation of D-dimers and the class flown (odds ratio (OR)
> > 0.61, p = 0.109).
> > "
>
> > I'd imagine that a huge amount of money, potentially, rests on this
> > evidence, so I'm surprised that this is the only study. I haven't got
> > Dr Jacobson's e-mail address at Wits, nor the whole text of the study
> > yet, but I'll follow these up to understand more detail and if there
> > was any special funding.
>
> > I find something curious about the above, though, maybe somebody with
> > some knowledge of statistics can help explain it.
>
> > 1. If all air passengers were the same, and there was no bias caused
> > by the class of travel, then you'd expect, I'd have thought, to find
> > the same raised D-dimers (the proxy for potential DVT in the study) in
> > both populations. To me, finding 14% in on population and 7% in the
> > other would suggest that the first population was twice as likely to
> > suffer the effect.
>
> > 2. Clearly the size of study is important. So, though the whole study
> > includes nearly 900 passengers, the study only examines 180 business
> > class passengers. So, these are less likely to be representative than
> > those not in business class.
>
> > 3. Isn't it also likely that those flying in Business class will have
> > other characteristics that differ that might be significant in their
> > risk of DVT? Shouldn't these factors be exluded before a comparison is
> > made?
>
> > 4. How, then, do they come to the conclusion that there is no effect?
>
> Their conclusion is "There was no significant association between elevation
> of D-dimers and the class flown (odds ratio (OR)> 0.61, p = 0.109)", which
> means that, according to their calculations, the probability of the apparent
> association being due to nothing more than chance is 10.9%; by convention, a
> result is only regarded as statistically significant if this probability is
> less than 5%.
>
> It is not clear to me how they reached this conclusion.  According to the
> figures given, the odds for business class passengers to have raised
> D-dimers on arrival were found to be 22/158 and for economy class passengers
> 52/667.  The ratio between these is 1.79 or 0.56, depending on which way> round you take them, and  calculating the 95% confidence intervals as
> described, for example, herehttp://www.bmj.com/cgi/content/full/320/7247/1468, gives intervals of
> 1.05-3.03 and 0.33-0.95 respectively.  Since these intervals do not include
> the value 1, that seems to indicate p < 0.05.  Does the full text of the> study give any more information on how they arrived at their stated figures
> for the odds ratio and p?
>
No, it doesn't, it simply declares that it is not significant.

On May 25, 1:55 am, "Philip"  wrote:
> "Peter Brooks"  wrote in message
>
> news:f26bce74-9036-46dd-9460-6ac9fb20b16d@a9g2000prl.googlegroups.com...
>
>
>
> > The only study that I can find that measures the effect of class of
> > travel on the chance of getting a DVT is this on (all the other
> > articles seem to refer back to it):
>
> > S Afr Med J. 2003 Jul;93(7):522-8.
> > The BEST study--a prospective study to compare business class versus
> > economy class air travel as a cause of thrombosis.
>
> > "
> > Only 434 subjects had a full venous duplex scan performed. None had
> > ultrasonic evidence of venous thrombosis. Nine passengers tested at
> > departure had elevated D-dimer levels and these volunteers were
> > excluded from further study. Seventy-four of the 899 passengers had
> > raised D-dimers on arrival. Twenty-two of 180 business class
> > passengers (12%) developed elevated D-dimers compared with 52 of 719
> > economy class passengers (7%). There was no significant association
> > between elevation of D-dimers and the class flown (odds ratio (OR)
> > 0.61, p = 0.109).
> > "
>
> > I'd imagine that a huge amount of money, potentially, rests on this
> > evidence, so I'm surprised that this is the only study. I haven't got
> > Dr Jacobson's e-mail address at Wits, nor the whole text of the study
> > yet, but I'll follow these up to understand more detail and if there
> > was any special funding.
>
> > I find something curious about the above, though, maybe somebody with
> > some knowledge of statistics can help explain it.
>
> > 1. If all air passengers were the same, and there was no bias caused
> > by the class of travel, then you'd expect, I'd have thought, to find
> > the same raised D-dimers (the proxy for potential DVT in the study) in
> > both populations. To me, finding 14% in on population and 7% in the
> > other would suggest that the first population was twice as likely to
> > suffer the effect.
>
> > 2. Clearly the size of study is important. So, though the whole study
> > includes nearly 900 passengers, the study only examines 180 business
> > class passengers. So, these are less likely to be representative than
> > those not in business class.
>
> > 3. Isn't it also likely that those flying in Business class will have
> > other characteristics that differ that might be significant in their
> > risk of DVT? Shouldn't these factors be exluded before a comparison is
> > made?
>
> > 4. How, then, do they come to the conclusion that there is no effect?
>
> Their conclusion is "There was no significant association between elevation
> of D-dimers and the class flown (odds ratio (OR)> 0.61, p = 0.109)", which
> means that, according to their calculations, the probability of the apparent
> association being due to nothing more than chance is 10.9%; by convention, a
> result is only regarded as statistically significant if this probability is
> less than 5%.
>
> It is not clear to me how they reached this conclusion.  According to the
> figures given, the odds for business class passengers to have raised
> D-dimers on arrival were found to be 22/158 and for economy class passengers
> 52/667.  The ratio between these is 1.79 or 0.56, depending on which way> round you take them, and  calculating the 95% confidence intervals as
> described, for example, herehttp://www.bmj.com/cgi/content/full/320/7247/1468, gives intervals of
> 1.05-3.03 and 0.33-0.95 respectively.  Since these intervals do not include
> the value 1, that seems to indicate p < 0.05.  Does the full text of the> study give any more information on how they arrived at their stated figures
> for the odds ratio and p?
>
I have been in touch with the author of the study who put me in touch
with the statistician that he used who had this comment:

"
Difficult to get hand on precise output, my suspicion is that the
problem may be with our reporting and that the non-significant p-value
is associated with the adjusted OR of 0.61 from a logistic regression.
I could actually trace an unadjusted OR of 0.61(approx), i.e. only
class was entered into a logistic regression, and the latter was based
on 873 observations with p=0.07. Adjusted for age and sex p=0.088.
There is a later dataset but I do not have stored output
"

He is happy to discuss this with me further, but I'm not sure what to
make of this. Any suggestions?

On 26 May, 19:39, Peter Brooks  wrote:
> On May 25, 1:55 am, "Philip"  wrote:
>
>
>
> > "Peter Brooks"  wrote in message
>
> >news:f26bce74-9036-46dd-9460-6ac9fb20b16d@a9g2000prl.googlegroups.com...
>
> > > The only study that I can find that measures the effect of class of
> > > travel on the chance of getting a DVT is this on (all the other
> > > articles seem to refer back to it):
>
> > > S Afr Med J. 2003 Jul;93(7):522-8.
> > > The BEST study--a prospective study to compare business class versus
> > > economy class air travel as a cause of thrombosis.
>
> > > "
> > > Only 434 subjects had a full venous duplex scan performed. None had
> > > ultrasonic evidence of venous thrombosis. Nine passengers tested at
> > > departure had elevated D-dimer levels and these volunteers were
> > > excluded from further study. Seventy-four of the 899 passengers had
> > > raised D-dimers on arrival. Twenty-two of 180 business class
> > > passengers (12%) developed elevated D-dimers compared with 52 of 719
> > > economy class passengers (7%). There was no significant association
> > > between elevation of D-dimers and the class flown (odds ratio (OR)
> > > 0.61, p = 0.109).
> > > "
>
> > > I'd imagine that a huge amount of money, potentially, rests on this
> > > evidence, so I'm surprised that this is the only study. I haven't got
> > > Dr Jacobson's e-mail address at Wits, nor the whole text of the study
> > > yet, but I'll follow these up to understand more detail and if there
> > > was any special funding.
>
> > > I find something curious about the above, though, maybe somebody with
> > > some knowledge of statistics can help explain it.
>
> > > 1. If all air passengers were the same, and there was no bias caused
> > > by the class of travel, then you'd expect, I'd have thought, to find
> > > the same raised D-dimers (the proxy for potential DVT in the study) in> > > both populations. To me, finding 14% in on population and 7% in the
> > > other would suggest that the first population was twice as likely to
> > > suffer the effect.
>
> > > 2. Clearly the size of study is important. So, though the whole study
> > > includes nearly 900 passengers, the study only examines 180 business
> > > class passengers. So, these are less likely to be representative than
> > > those not in business class.
>
> > > 3. Isn't it also likely that those flying in Business class will have
> > > other characteristics that differ that might be significant in their
> > > risk of DVT? Shouldn't these factors be exluded before a comparison is> > > made?
>
> > > 4. How, then, do they come to the conclusion that there is no effect?
>
> > Their conclusion is "There was no significant association between elevation
> > of D-dimers and the class flown (odds ratio (OR)> 0.61, p = 0.109)", which
> > means that, according to their calculations, the probability of the apparent
> > association being due to nothing more than chance is 10.9%; by convention, a
> > result is only regarded as statistically significant if this probability is
> > less than 5%.
>
> > It is not clear to me how they reached this conclusion.  According to the
> > figures given, the odds for business class passengers to have raised
> > D-dimers on arrival were found to be 22/158 and for economy class passengers
> > 52/667.  The ratio between these is 1.79 or 0.56, depending on which way
> > round you take them, and  calculating the 95% confidence intervals as
> > described, for example, herehttp://www.bmj.com/cgi/content/full/320/7247/1468, gives intervals of
> > 1.05-3.03 and 0.33-0.95 respectively.  Since these intervals do not include
> > the value 1, that seems to indicate p < 0.05.  Does the full text of the
> > study give any more information on how they arrived at their stated figures
> > for the odds ratio and p?
>
> I have been in touch with the author of the study who put me in touch
> with the statistician that he used who had this comment:
>
> "
> Difficult to get hand on precise output, my suspicion is that the
> problem may be with our reporting and that the non-significant p-value
> is associated with the adjusted OR of 0.61 from a logistic regression.
> I could actually trace an unadjusted OR of 0.61(approx), i.e. only
> class was entered into a logistic regression, and the latter was based
> on 873 observations with p=0.07. Adjusted for age and sex p=0.088.
> There is a later dataset but I do not have stored output
> "
>
> He is happy to discuss this with me further, but I'm not sure what to
> make of this. Any suggestions?- Hide quoted text -
>
> - Show quoted text -


After controlling for other variables,  it seems the observed
difference is not statistically significant.  Anyway,  it is in the
wrong direction (contrary to prediction).  So,  travel economy.

Dave

On May 26, 9:41 pm, Dave Smith  wrote:
>
> > "
> > Difficult to get hand on precise output, my suspicion is that the
> > problem may be with our reporting and that the non-significant p-value
> > is associated with the adjusted OR of 0.61 from a logistic regression.
> > I could actually trace an unadjusted OR of 0.61(approx), i.e. only
> > class was entered into a logistic regression, and the latter was based
> > on 873 observations with p=0.07. Adjusted for age and sex p=0.088.
> > There is a later dataset but I do not have stored output
> > "
>
> > He is happy to discuss this with me further, but I'm not sure what to
> > make of this. Any suggestions?- Hide quoted text -
>
> > - Show quoted text -
>
> After controlling for other variables,  it seems the observed
> difference is not statistically significant.  Anyway,  it is in the
> wrong direction (contrary to prediction).  So,  travel economy.
>
Well, yes, that would seem to be the case. I'd have thought, though,
that, we need to know the null hypothesis before we can know if it is
significant. I agree it seems to be the wrong direction - that's part
of what is interesting, it was funded by an airline and this 'wrong
direction' is reported to be insignificant - might that not itself be
significant? I can't actually see hoards of Business class passengers
flooding the economy section for their health, but we would need to
know if it would be sensible for them to do that - or if, rather, they
should just drink less, loose weight, have a few inches cut of their
height and get a castration [oh, dear, there's no data on eunuchs, so
that might not help].

If the result of 14% of Business Class passengers showing some proxy
event vs 7% of the Economy class passengers is inany way significant.
It is not clear to me, from the paper, what the null hypothesis
actually is. My thoughts, before seeing the study, would be that, if
anything, Business Class passengers (particularly when those at
particular risk - who may be travelling in Business Class for that
reason - have been excluded) would be expected to have fewer proxy
events than Economy - that is my null hypothesis would be something
like an expectation that, say, 5 rather than 22 (3.5% rather than 14%)
of the Business class passengers would have a proxy event. That would
mean, if that were the null hypothesis, that the actual result is very
surprising 14% instead of 3.5% would be a very significant difference
- and would suggest that, for some reason, Business class is more
dangerous than Economy class.

The paper mentions the adjustment for age and sex, which seems to me
not to be an easy one either - the sex ratio in economy class is close
to the standard 1:1, whilst that in business class is strongly male
dominated - now the paper makes it clear that OC and HRT use
(exclusively female, I would have thought, at the time, even though
there is some male HRT these days) increase the risk of a proxy event,
so this must mean that there have been a fair number of female proxy
events in the study, so females, being underrepresented in Business
class, might cause an over-correction downwards in the significance. I
don't know, this is just what the study leads me to wonder about.

Unfortunately none of this affects me directly. I'm a 'high-risk'
passenger as my grandmother died of phlebitis and I had a DVT back in
'92 - and so I'd have been excluded from the BEST study. The court is
out on whether Business class helps those that are high risk, or not.

For anybody who isn't utterly bored by this, I have put a text copy of
the study below.

Also, I'd be interested in any comments on another thing that confuses
me. The study uses a logistical regression to estimate the affect of
the various risk factors (http://en.wikipedia.org/wiki/
Logistic_regression). I can't see the logic in using it in this study.
The growth curve on which the logistical regression is based shows
tails at both ends - which make sense, as in the wikipaedia article,
for population growth. I don't see why the affect of age, sex, class
of travel or pretty well anything else would be related to the
development of thrombi in this way. Why would somebody use this
regression method for this?


"
ORIGINAL ARTICLES ORIGINAL ARTICLES
The BEST study — a prospective study to compare
business class versus economy class air travel as a cause
of thrombosis

Barry F Jacobson, Marion Münster, Alberto Smith, Kevin G Burnand,
Andrew Carter, A Talib O Abdool-Carrim, Elizabeth
Marcos, Piet J Becker, Timothy Rogers, Dirk le Roux, Jennifer LCalvert-
Evers, Marietha J Nel, Robyn Brackin, Martin Veller

Background. As many as 10% of airline passengers travelling
without prophylaxis for long distances may develop a venous
thrombosis. There is, however, no evidence that economy
class travellers are at increased risk of thrombosis.

Objectives. A suitably powered prospective study, based on
the incidence of deep-vein thrombosis (DVT) reported in
previous studies on long-haul flights, was designed to
determine the incidence of positive venous duplex scans and
D-dimer elevations in low and intermediate-risk passengers,
comparing passengers travelling in business and economy
class.

Patients/methods. Eight hundred and ninety-nine passengers
were recruited (180 travelling business class and 719
travelling economy). D-dimers were measured before and
after the flight. A value greater than 500 ng/ml was accepted
as abnormal. A thrombophilia screen was conducted which
included the factor V Leiden mutation, the prothombin
20210Amutation, protein C and S levels, antithrombin levels,
and anticardiolipin antibodies immunoglobulin G (IgG) and
immunoglobulin M (IgM). On arrival, lower limb
compression ultrasonography of the deep veins was
performed. Logistical regression analysis was used to
determine the risk factors related to abnormally high D-dimer
levels.

Results. Only 434 subjects had a full venous duplex scan

performed. None had ultrasonic evidence of venous
thrombosis. Nine passengers tested at departure had elevated
D-dimer levels and these volunteers were excluded from
further study. Seventy-four of the 899 passengers had raised
D-dimers on arrival. Twenty-two of 180 business class
passengers (12%) developed elevated D-dimers compared
with 52 of 719 economy class passengers (7%). There was no
significant association between elevation of D-dimers and the
class flown (odds ratio (OR) 0.61, p = 0.109). The factor V
Leiden mutation, factor VIII levels and the use of aspirin
were, however, associated with raised D-dimers (OR 3.36,
p = 0.024; OR 1.01, p = 0.014; and OR 2.04, p = 0.038,
respectively). Five hundred and five passengers were
contacted within 6 months and none reported any symptoms
of a clinical thrombosis or pulmonary embolus.

Conclusion. The incidence of ultrasonically proven DVT is
much lower than previously reported. However, more than
10% of all passengers developed raised D-dimers, which were
unrelated to the class flown. A rise in D-dimers is associated
with an inherent risk of thrombosis and/or thrombophilia,
demonstrates activation of both the coagulation and
fibrinolytic systems during long-haul flights, and may
indicate the development of small thrombi.

S Afr Med J 2003: 93: 522-528.

522
Departments of Haematology and Surgery, University of the
Witwatersrand, and
National Health Laboratory Service, Johannesburg

Barry F Jacobson, FRCS, PhD

Department of Haematology, University of the Witwatersrand, and
National Health
Laboratory Service, Johannesburg

Marion Münster, MMed
Robyn Brackin, BSc

Division of Surgery and Anaesthesia, Guy’s, King’s and St Thomas’s
School of
Medicine, UK

Alberto Smith, PhD
Kevin G Burnand, FRCS, MS

Department of Chemical Pathology, University of the Witwatersrand and
National
Health Laboratory Service, Johannesburg

Andrew Carter, FCPath

Department of Surgery, University of the Witwatersrand, Johannesburg

A Talib O Abdool-Carrim, FRCS
Elizabeth Marcos, Dip Clin Tech
Timothy Rogers, FCS
Dirk le Roux, FCS
Jennifer LCalvert-Evers, PhD
Maritha JNel, MSc
Martin Veller, FCS

Biostatistics Unit, Medical Research Council, Pretoria

Piet JBecker, PhD

July 2003, Vol. 93, No. 7 SAMJ


ORIGINAL ARTICLES ORIGINAL ARTICLES
The association between venous thromboembolism (VTE) and
air travel has received considerable attention because of several
recent media and scientific reports of fatal pulmonary emboli
developing in passengers, either during flights or shortly after
disembarking. Although not confined to flying, travel-
associated thrombosis is commonly referred to as the ‘economy
class syndrome’. The term evolved from the premise that the
underlying aetiology of VTE is venous stasis secondary to
relative immobility; a consequence of being confined to seating
with limited legroom. Flight-simulation studies have shown
that venous stasis is a consequence of being seated in an
aircraft for prolonged periods of time.1 A number of studies
have provided evidence that there is a link between long-haul
air travel and VTE,2-6 although it is accepted that this risk has
not been adequately quantified.7 As yet, no study has
demonstrated that passengers travelling in economy class are
more predisposed to VTE than those travelling in business
class or first class, where there is more legroom and seats
recline.

The BEST study (Business class versus Economy class
Syndrome as a cause of Thrombosis) was designed to be the
first large prospective study to compare the incidence of VTE
in low- and intermediate-risk8 economy class and business
class passengers. All passengers with any identifiable
predisposition to VTE were excluded. The intention was to
determine the incidence of thrombosis using a strategy
combining D-dimer measurement on departure and arrival,
and a full evaluation of all the deep veins by compression
ultrasound of all participating passengers on arrival. The study
was also designed to evaluate factors such as fluid and alcohol
intake, exercise, smoking status and medication.

Although recent media reports have dramatised the
‘economy class syndrome’, a World Health Organisation
(WHO) report7 suggests that the actual incidence of travel-
associated VTE might be much lower than previously
estimated. In an attempt to quantify the true incidence
accurately we decided to use D-dimers as well as compression
ultrasonography (CUS). CUS is an excellent technique to assess
venous thrombosis above the knee. Unfortunately, the
sensitivity of this technique diminishes markedly below the
knee.9 The D-dimer, a specific degradation product of crosslinked
fibrin, is a well-established highly sensitive test for the
detection of VTE.10,11 We therefore decided to use positive D-
dimers (• 500 ng/ml) as an indicator to identify passengers
with travel-induced thrombotic activity.

As the risk of VTE is known to increase with duration of
travel,6 the London-to-Johannesburg route was chosen. It has a
flight duration of approximately 11 hours and is therefore a
‘high-risk’ flight, with the added advantage of not having
major time zone changes as a potential confounding variable.

Materials and methods

Recruitment

Passengers booked on two daily non-stop scheduled flights
from London’s Heathrow Airport to Johannesburg
International Airport, between April 2002 and June 2002, were
approached for inclusion in the study at the South African
Airways check-in desk. An estimate of the number of
passengers required to evaluate the increased risk of
thrombosis caused by long-haul flight was derived from two
other prospective studies3,5 which reported an incidence of
between 3% and 10%. The aim was to enrol approximately
1 000 passengers, as this would give a greater than 90% power
to detect a lower incidence of thrombosis than that previously
reported. There was an active attempt to recruit equal numbers
of economy class and business class passengers. Financial
constraints limited the duration of the study, which was
terminated before equivalent numbers of business class
passengers could be recruited.

Exclusion criteria were a known history of VTE, current
warfarin and/or low-molecular-weight heparin usage,
pregnancy, surgery within the last 6 weeks and the intention to
use medical support stockings in-flight. All passengers who
agreed to participate were required to sign a consent form,
complete pre- and post-flight questionnaires, and undertake
pre- and post-flight blood sampling. Ultrasonography of the
lower limbs on disembarking was encouraged but optional. All
participants who completed the study were allotted an extra
5 000 South African Airways Voyager Air Miles (frequent flyer)
for the flight.

Questionnaire

A questionnaire documenting the class flown, age, gender,
height and weight, was given to passengers on departure. They
were asked to record their in-flight fluid intake (non-alcoholic
and alcoholic) and amount of in-flight exercise performed.
Smoking status, use of aspirin, anti-inflammatory agents,
sleeping tablets, oral contraceptives (OC) and hormone
replacement therapy (HRT) were also documented.

Phlebotomy

Ven e p u n t u re was performed after check-in and immediately on
disembarkation. The phlebotomists were instructed not to make
m o re than two attempts at venepuncture in an attempt to ensure
volunteer compliance and not impact negatively on future
volunteer re c ruitment. Blood was taken from each volunteer
passenger using a 21-gauge needle and the Becton Dickinson
Vacutainer system. Pre- and post-flight blood samples collected
included one 5 ml ethylenediamine tetra-acetic acid (EDTA),
one 10 ml serum separation table (SST) and two 5 ml sodium
citrate standard Becton Dickenson Vacutainer tubes.

July 2003, Vol. 93, No. 7 SAMJ


523

ORIGINAL ARTICLES ORIGINAL ARTICLES
524
Specimen preparation

The blood specimens collected in 0.105M (3.2%) trisodium
citrate (9 volumes blood: 1 volume citrate) tubes were
centrifuged at 2 500 g for 10 minutes. Specimens collected preflight
at Heathrow Airport were centrifuged, plasma separated,
aliquoted into plastic tubes and frozen at –20°C within 30
minutes of collection. The pre-flight specimens were
transferred from Heathrow Airport to St Thomas’s Hospital,
London, each evening for storage at –20°C for 1 month and
subsequently at –70°C. On completion of the study all plasma
specimens were couriered on dry ice to the National Health
Laboratory Service Laboratories at Johannesburg Hospital
(South Africa) for testing. The post-flight specimens were
transferred from the Johannesburg International Airport to the
Johannesburg Hospital Laboratories for processing after the
arrival of each flight. The maximum time between blood
collection and processing and/or freezing was 3 hours. The
buffy coat was separated from the EDTA specimen post platelet
count analysis and stored at –70°C until DNAextraction was
performed for factor V Leiden and prothrombin 20210Agene
mutation analysis.

Specimen analysis

All specimens were analysed by the National Health
Laboratory Service. The post-flight EDTA specimens
underwent platelet count analysis on a Coulter GENS within 4
hours of collection. The pre-flight trisodium citrate specimens
underwent D-dimer testing. Analysis of the post-flight
trisodium citrate specimens included assays for D-dimers,
fibrinogen, factor VIII (FVIII), protein C (PC), protein S (PS)
and antithrombin (AT). Anticardiolipin testing was performed
on blood collected in the SST tubes.

The international normalised ratio (INR), activated partial
thromboplastin time (aPTT) and D-dimer analyses were
performed on an Automated Coagulation Laboratory (ACL)
Futura (Instrumentation Laboratories, Milan) using
Thromborel R (Dade Behring — ISI 1.01), ILaPTT SPliquid
reagent (Instrumentation Laboratories) and ILD-dimer kit
(Instrumentation Laboratories) reagents respectively. The latter
kit has been validated in our laboratory against the Gold D-
dimer enzyme-linked immunosorbent (ELISA) assay. Any
value • 500 ng/ml was considered positive.

The FVIII, fibrinogen, PC and AT were analysed on an ACL
9000 (Instrumentation Laboratories). The FVIII was assayed
using FVIII-deficient plasma from Dade Behring and ILaPTT
SP reagent. The control plasma used was standard human
plasma (SHP) from Dade Behring. The fibrinogen, PC and AT
tests were all performed using standard Instrumentation
Laboratories reagents. The PS testing was performed on an
ACL3000 using the Instrumentation Laboratories Functional
Protein S kit. The anticardiolipins were assayed with the
QACAimmunoglobulin G (IgG) and immunoglobulin M (IgM)

ELISAkits from Cheshire Diagnostics Limited.
Factor V Leiden and prothrombin mutations were analysed
using a multiplex assay and real-time PCR. Genomic DNAwas
extracted from peripheral blood anticoagulated with EDTA
using the method previously described.12 The multiplex PCR
used contained two unlabelled primer sets and two pairs of
fluorescent labelled probes. A 306-bp fragment of the Factor V
gene was amplified using unlabelled primers and labelled
probes previously reported.13 A 291-bp fragment of the
prothrombin gene was also amplified using unlabelled primers
and labelled probes previously described.14 PCR reactions were
performed using the LightCycler. Fluorescence was measured
continuously during the melting curve analysis.

Lower limb CUS

After venepuncture had been carried out at Johannesbur g
International Airport, passengers were asked to stay for CUS of
the deep venous system of both legs. The paired venae
commitantes of the calf, the soleal, popliteal and the femoral
veins were all insonated. This was done by five experienced
ultrasound technicians (employees of Tecmed South Africa
under Toshiba Japan) using Toshiba Nemio 20 ultrasound
machines with 7211 MHz probes.

Passengers without connecting flights who were unwilling to
wait were asked to attend the Milpark Hospital for an
ultrasound assessment within 1 week of disembarkation. A
single senior ultrasonographer performed these delayed
ultrasounds using a Hewlett Packard instrument with a 7210
MHz probe.

A second senior ultrasonographer and a vascular surgeon
randomly checked all ultrasound examinations. The same
vascular surgeon reviewed all the ultrasound films of those
passengers who were shown to have positive D-dimers.

Follow-up

Attempts were made to contact all passengers who supplied
telephone contact details 6 months after this study flight in
order to ascertain whether there were any clinical sequelae
suggestive of VTE because of the flight studied.

Statistical analysis

A logistical regression analysis was performed using Stata
version 7 statistical software to assess risk factors that
contributed to a positive D-dimer result ( • 500 ng/ml). Only
subjects with complete datasets were included in the analysis.
In the logistical regression analysis D-dimer positivity was
initially regressed against all the characteristics in Tables Iand
II, after which those characteristics with high p-values and
odds ratios (ORs) close to 1 were removed. In the final analysis
class travelled, presence of factor V Leiden mutation, FVIII
levels, aspirin use, age, sex, height in 10 cm units, weight in 10

July 2003, Vol. 93, No. 7 SAMJ


ORIGINAL ARTICLES ORIGINAL ARTICLES
kg units and smoking status were regressed against D-dimer
positivity for 778 passengers who had complete data.

Ethics approval

The study was approved by the Ethics Committees for Human
Research from both the University of the Witwatersrand and St
Thomas’s Hospital College.

Results

Passenger demographics

These are summarised in Tables I and III. Business class and
first class passengers were analysed together as the majority of
passengers who flew in business class had a flat bed. There was a

4:1 ratio of economy class to business class passengers re c ru i t e
d
despite an attempt to obtain equivalent numbers. There was a
similar number of men and women in the economy class gro u p
but a 2:1 ratio of men to women in business class. While the height,
weight and fluid intake were similar in both classes, the mean
alcohol intake was 66% higher in business class. Only 6% of all
passengers in both classes exercised during the flight.
Table I. Passenger demographic data
Characteristics Business class Economy class
Number 180 719
Age
Median (years) 47 43
Range (years) 28 - 78 18 - 81
Sex
Male 119 368
Female 61 351
Median height (cm) 176 173
Range (cm) 130 - 201 120 - 210
Median weight (kg) 82 76
Range (kg) 49 - 126 44 - 168
Median fluid intake (l) 1.0 1.0
Range (l) 0 - 3.25 0- 5.00
Mean alcohol intake (l) 0.250 0.175
Range (l) 0 - 2.38 0 - 1.5
Exercise (yes/no) 10/170 41/678
Percentage (yes) 6 6
Smokers (yes/no) 25/155 121/598
Percentage (yes) 14 17
Passengers (N (%))
Using aspirin 22 (12.2) 89 (12.4)
Using anti-inflammatories 4 (2.2) 26 (3.6)
Using sleeping tablets 24 (13.3) 94 (13.1)
Using OCs 9 (5) 72 (10)
Using HRT 12 (6.7) 71 (9.9)
Ultrasound (N %))
Immediate 77 (42.8) 310 (43.1)
Within 7 days 9 (5) 38 (5.3)
OCs = oral contraceptives; HRT = hormone replacement therapy.
Table II. Thrombotic screen results
Total
Total number Percentage
number positive/ abnormal
Test tested deficient results
Factor V Leiden 844 31 3.7
Prothrombin 20210A 844 22 2.6
Protein C (IU/dl) 873 18 2
(normal range 70 - 160)
Antithrombin (IU/dl) 873 5 0.6
(normal range 76 - 125)
Protein S (IU/dl) 871 79 9
(normal range 60 - 140)
ACLAIgG GPLU/ml 817 9 1.1
(normal range < 16)
ACLAIgM MPLU/ml 817 12 1.5
(Normal range < 14)
Number
Test of tests Mean Range
Factor VIII (%) 899 107.5 33 - 293
(normal range 50 - 150)
Fibrinogen (g/l) 873 2.94 0.66 - 6.58
(normal range 2 - 4)
Platelet count (x 109/l) 702 271 127 - 497
(normal range 140 - 400)
D-dimers

Nine of the 491 samples taken at departure from Heathrow had
D-dimer levels higher than 500 ng/ml, 1% from business class
and 2% from economy class (Table III). These 9 passengers
were excluded from further analysis. Seventy-four of 899
samples taken on arrival had elevated D-dimers (Table III).
Twenty-two of 180 business class passengers had elevated
levels compared with 52/719 in economy class (7%). This
difference was not significant (OR 0.61, p = 0.109).

Logistical regression analysis on 784 passengers with
complete datasets produced significant positive correlations
between elevated D-dimers and the presence of the factor V
Leiden mutation (OR 3.36, p = 0.024), FVIII levels (OR 1.01,
p = 0.014) and the use of aspirin (OR 2.04, p = 0.038) (Table IV).

A separate logistical regression analysis on D-dimers was
performed on the subset of female passengers who were taking

Table III. D-dimer results per class
Business class Economy class
passengers passengers
D-dimers > 500 ng/ml 22/180 (12) 52/719 (7)
on disembarking (N (%))
D-dimers > 500 ng/ml 1/104 (1) 8/387 (2)
on embarking (N (%))
525
July 2003, Vol. 93, No. 7 SAMJ


ORIGINAL ARTICLES ORIGINAL ARTICLES
526
Table IV. Regression analysis of factors that have a statistically
significant association with positive D-dimers in both sexes
95% confidence
Odds ratio P-value interval
Factor V Leiden 3.36 0.024 1.17 - 9.63
Factor VIII 1.01 0.014 1.00 - 1.01
Aspirin 2.04 0.038 1.04 - 3.99
Table V. Regression analysis of potential hormonal factors
contributing to the development of positive D-dimers in females
95%
confidence
Odds ratio P-value interval
Hormone replacement 2.15 0.083 0.91 - 5.10
therapy
Oral contraceptive 1.23 0.707 0.42 - 3.62
HRT or OCs (Table V). There were 382 female passengers with
complete datasets available for this assessment. The OR of 2.15
for HRT might be clinically significant even though the p-value
is 0.083.

Lower limb CUS

Three hundred and eighty-seven passengers had CUS on
arrival and 47 passengers had it performed 1 week after
arrival. None of the 434 passengers who were assessed had
evidence of venous thrombosis of the lower limbs. Two
passengers had equivocal scans on arrival. Their femoral veins
were compressible but appeared to have diminished flow. Both
of these passengers had repeat examinations within 2 days,
which were normal, and both had D-dimers within the normal
range.

Follow-up

Telephonic contact details were available for 627 of the 899
passengers, of whom we were able to contact 505 (81%) within
6 months of the study flight. Sixty-four of 74 passengers with
positive post-flight D-dimers were contacted. None of the
passengers contacted reported any signs or symptoms of a
delayed thrombotic event pertaining to the study flight. One
passenger reported that he developed a clinical deep-vein
thrombosis (DVT) after a long flight that he had made six
flights after the study flight.

Discussion

There is considerable anecdotal evidence that DVT can follow
prolonged airline flight and other forms of travel. This has
been reinforced by more recent trials and surveys. 2-6,15 Two

studies have suggested a high prevalence of the condition in
high-risk passengers,3,5 although cockpit crew do not appear to
be affected.16 There is, however, no evidence that the economy
class syndrome exists, although it is now part of popular
mythology.

Our study specifically excluded passengers with a history of
VTE, recent surgery, and current pregnancy. No passenger used
compression stockings. The study therefore focused on
passengers at low or intermediate risk of developing
thrombosis. This is the first study that has specifically
examined the association between markers of thrombosis (Ddimer)
and class of travel.

In contrast to what we expected, there was no statistical
relationship between flight class and the development of
elevated D-dimers. We had hoped for a more even distribution
between the number of business and economy class
passengers, but volunteers in economy class outnumbered
those in business class. This discrepancy was probably caused
by a limited number of business class seats per flight and the
fact that many of the business class passengers either checked
in late or did not wish to be delayed on disembarking. The
similarity in the percentage of passengers with raised D-dimers
in both classes might have been the result of immobility rather
than the cramped seating as only 6% of passengers reported
exercising during the flight.

Although fewer pre-flight D-dimer assays were performed
compared with post-flight specimens due to an entire batch of
specimens becoming activated during transportation and being
discarded, over 7% of our passengers developed raised D-
dimers post-flight. This contrasts with the study by Scurr et al.3
where no D-dimer elevation was found, perhaps because blood
collection was delayed. The NZATT study17 has also recently
reported elevated D-dimers in passengers following long-haul
flights.

In contrast to previous reports none of the passengers in this
study who agreed to be examined had ultrasonic evidence of
lower limb thrombosis.3,5,17 Unfortunately many of the
passengers declined CUS examination at the end of their long
flight. Only 47% of business class and 48% of the economy
class passengers agreed to be examined and only 51% of the
passengers who had positive D-dimers had ultrasound scans
performed. Although the overall percentage of passengers who
underwent lower limb CUS was low, there was no bias
between the positive and negative D-dimer subgroups. The
majority of the ultrasound scans were performed on arrival
and this may have resulted in developing thrombi being
missed. However, none of the 47 late ultrasounds showed
evidence of thrombi. Furthermore, approximately 90% of
passengers with elevated D-dimers were contacted at 6 months
and none reported any symptoms suggestive of VTE.

The sensitivity of CUS for the detection of thrombi in the
lower limb is excellent above the knee, but its utility in the

July 2003, Vol. 93, No. 7 SAMJ


ORIGINAL ARTICLES
527
July 2003, Vol. 93, No. 7 SAMJ
detection of calf vein thrombi is far less impressive.9 In
contrast, D-dimer measurement is a highly sensitive test but
lacks specificity.18 The clinical significance of elevated D-dimers
is difficult to assess. Studies have shown that the D-dimer test
is a useful first-step investigation in patients with clinical signs
and symptoms suggestive of DVT or pulmonary embolism.10,18,19
The very high negative predictive value has led to the adoption
of this strategy in clinical practice, principally to minimise
unnecessary patient exposure to radiological investigations,
which are both costly and not without morbidity. It is
particularly useful in assessing non-surgical patients as
positive D-dimers are almost always observed postoperatively.
The D-dimer is an end product of plasmin digestion of cross-
linked fibrin, which in turn is generated by thrombin.20 The D-
dimer is therefore a marker of thrombin generation signifying
that the coagulation cascade must have been triggered. The
level of elevation of D-dimers is dependent on the mass and
age of the thrombus and hence D-dimers may be raised
without radiological evidence of thrombosis at the time of
testing in patients presenting with symptoms suggestive of
thrombosis.18,21 Bernardi et al.21 confirmed this concept in their
study in which 6% of subjects with positive D-dimers at
presentation subsequently developed DVTs despite an initial
normal ultrasound study.
It must be accepted that the D-dimer is at best a surrogate
marker of thrombosis and may be elevated in the absence of
thrombus. It is interesting, however, that factor V Leiden (the
major inherited risk factor for venous thrombosis) was
identified as an independent risk factor in passengers with
elevated D-dimers. The incidence of 3.7% heterozygous
individuals in this study is comparable with general
population studies in the Western world.22 The question of
whether pre-flight knowledge of factor V Leiden genotype
would influence the implementation of in-flight prophylaxis
requires a prospective study.
The prothrombin gene 20210Amutation was identified in
2.6% of passengers but surprisingly was not a predictor of
positive D-dimers. The other routine tests of hereditary
thrombophilic states, i.e. protein C, protein S and antithrombin
deficiencies, likewise did not prove to contribute to positive D-
dimers. The number of passengers with low protein S levels
was surprisingly high and did not correlate with the use of
oestrogen products. The presence of anticardiolipin antibodies
was not related to positive D-dimers. Although statistically
significant, it is difficult to draw any definite conclusion from
the finding of the correlation between elevated factor VIII
levels and positive D-dimers in the absence of an elevated odds
ratio (OR 1.01). The very tight 95% confidence interval (1.00 -
1.01) is, however, intriguing. Factor VIII becomes elevated
during stress and the stress of flying possibly contributes to an
underlying thrombophilic state. However, an elevated factor
VIII level consequent to subclinical thrombosis as suggested by
positive D-dimers cannot be excluded.
There was a relationship between HRT and elevated D-
dimers. This observation, although not statistically significant,
suggests that women using this medication should be
particularly cautious when travelling long distances. Positive
D-dimers were, however, not correlated with OC usage. In
light of the well-documented association between OC usage
and VTE, this observation cannot be explained.23,24 In particular
there was no association with age.
Smoking status was not related to the development of
positive D-dimers. Smoking is a potent cause of arterial
thrombosis but there is not much evidence linking it to venous
thrombosis. Although one might have expected that passengers
who were taller or heavier than the norm, and therefore subject
to greater in-flight space restriction, would have a higher
incidence of elevated D-dimers, this was not borne out by the
statistical analysis.
An assessment of advice previously offered to passengers,
viz. to increase fluid intake, limit alcohol intake and actively
perform the exercises recommended in the in-flight magazine,
revealed that none of them could be supported by the results of
our study. The use of sleeping tablets was expected to be a
potential factor contributing to the tendency to thrombosis by
decreasing in-flight movement. No association was noted.
Aspirin, but not other anti-inflammatory agent usage, was
statistically significantly associated with positive D-dimers.
Aspirin has been well documented to prevent arterial
thrombosis but the evidence for its preventing venous
thrombosis is far less convincing.25,26 It is unlikely that it causes
thrombosis. A more plausible theory is either that it caused
gastritis, which falsely elevated the D-dimers, or that the
passengers who used aspirin comprised a group of high-risk
individuals. The NZATT study17 has now reported that aspirin
does not prevent travel-associated venous thromboses.
In conclusion, in the largest prospective study to date, we
have shown that the incidence of clinical and/or CUS-
documented lower limb venous thrombosis is much lower than
has been previously reported as it occuured in none of the
passengers studied. No passenger reported symptoms of DVT
or pulmonary embolism within 6 months of the study flight.
This may in part be explained by differences in passenger
demographics as high-risk passengers were excluded from this
study. We have also demonstrated that a significant number of
passengers developed elevated D-dimer levels during flight.
This was unrelated to flight class. The large prospective study
proposed by the WHO may shed further light on the true risk
of air travel-associated thrombosis.
Major sponsors of the study were: South African Airways —
donation of Voyager Air Miles for study participants and return
travel for the South African research team to London; SSL
International — accommodation in London and local travel costs,
phlebotomy service in London and majority of laboratory
investigations; Tecmed South Africa and Toshiba Japan — loan of
ultrasound equipment and operators; Aventis South Africa —
ORIGINAL ARTICLES
527
July 2003, Vol. 93, No. 7 SAMJ
detection of calf vein thrombi is far less impressive.9 In
contrast, D-dimer measurement is a highly sensitive test but
lacks specificity.18 The clinical significance of elevated D-dimers
is difficult to assess. Studies have shown that the D-dimer test
is a useful first-step investigation in patients with clinical signs
and symptoms suggestive of DVT or pulmonary embolism.10,18,19
The very high negative predictive value has led to the adoption
of this strategy in clinical practice, principally to minimise
unnecessary patient exposure to radiological investigations,
which are both costly and not without morbidity. It is
particularly useful in assessing non-surgical patients as
positive D-dimers are almost always observed postoperatively.
The D-dimer is an end product of plasmin digestion of cross-
linked fibrin, which in turn is generated by thrombin.20 The D-
dimer is therefore a marker of thrombin generation signifying
that the coagulation cascade must have been triggered. The
level of elevation of D-dimers is dependent on the mass and
age of the thrombus and hence D-dimers may be raised
without radiological evidence of thrombosis at the time of
testing in patients presenting with symptoms suggestive of
thrombosis.18,21 Bernardi et al.21 confirmed this concept in their
study in which 6% of subjects with positive D-dimers at
presentation subsequently developed DVTs despite an initial
normal ultrasound study.
It must be accepted that the D-dimer is at best a surrogate
marker of thrombosis and may be elevated in the absence of
thrombus. It is interesting, however, that factor V Leiden (the
major inherited risk factor for venous thrombosis) was
identified as an independent risk factor in passengers with
elevated D-dimers. The incidence of 3.7% heterozygous
individuals in this study is comparable with general
population studies in the Western world.22 The question of
whether pre-flight knowledge of factor V Leiden genotype
would influence the implementation of in-flight prophylaxis
requires a prospective study.
The prothrombin gene 20210Amutation was identified in
2.6% of passengers but surprisingly was not a predictor of
positive D-dimers. The other routine tests of hereditary
thrombophilic states, i.e. protein C, protein S and antithrombin
deficiencies, likewise did not prove to contribute to positive D-
dimers. The number of passengers with low protein S levels
was surprisingly high and did not correlate with the use of
oestrogen products. The presence of anticardiolipin antibodies
was not related to positive D-dimers. Although statistically
significant, it is difficult to draw any definite conclusion from
the finding of the correlation between elevated factor VIII
levels and positive D-dimers in the absence of an elevated odds
ratio (OR 1.01). The very tight 95% confidence interval (1.00 -
1.01) is, however, intriguing. Factor VIII becomes elevated
during stress and the stress of flying possibly contributes to an
underlying thrombophilic state. However, an elevated factor
VIII level consequent to subclinical thrombosis as suggested by
positive D-dimers cannot be excluded.
There was a relationship between HRT and elevated D-
dimers. This observation, although not statistically significant,
suggests that women using this medication should be
particularly cautious when travelling long distances. Positive
D-dimers were, however, not correlated with OC usage. In
light of the well-documented association between OC usage
and VTE, this observation cannot be explained.23,24 In particular
there was no association with age.
Smoking status was not related to the development of
positive D-dimers. Smoking is a potent cause of arterial
thrombosis but there is not much evidence linking it to venous
thrombosis. Although one might have expected that passengers
who were taller or heavier than the norm, and therefore subject
to greater in-flight space restriction, would have a higher
incidence of elevated D-dimers, this was not borne out by the
statistical analysis.
An assessment of advice previously offered to passengers,
viz. to increase fluid intake, limit alcohol intake and actively
perform the exercises recommended in the in-flight magazine,
revealed that none of them could be supported by the results of
our study. The use of sleeping tablets was expected to be a
potential factor contributing to the tendency to thrombosis by
decreasing in-flight movement. No association was noted.
Aspirin, but not other anti-inflammatory agent usage, was
statistically significantly associated with positive D-dimers.
Aspirin has been well documented to prevent arterial
thrombosis but the evidence for its preventing venous
thrombosis is far less convincing.25,26 It is unlikely that it causes
thrombosis. A more plausible theory is either that it caused
gastritis, which falsely elevated the D-dimers, or that the
passengers who used aspirin comprised a group of high-risk
individuals. The NZATT study17 has now reported that aspirin
does not prevent travel-associated venous thromboses.
In conclusion, in the largest prospective study to date, we
have shown that the incidence of clinical and/or CUS-
documented lower limb venous thrombosis is much lower than
has been previously reported as it occuured in none of the
passengers studied. No passenger reported symptoms of DVT
or pulmonary embolism within 6 months of the study flight.
This may in part be explained by differences in passenger
demographics as high-risk passengers were excluded from this
study. We have also demonstrated that a significant number of
passengers developed elevated D-dimer levels during flight.
This was unrelated to flight class. The large prospective study
proposed by the WHO may shed further light on the true risk
of air travel-associated thrombosis.
Major sponsors of the study were: South African Airways —
donation of Voyager Air Miles for study participants and return
travel for the South African research team to London; SSL
International — accommodation in London and local travel costs,
phlebotomy service in London and majority of laboratory
investigations; Tecmed South Africa and Toshiba Japan — loan of
ultrasound equipment and operators; Aventis South Africa —

ORIGINAL ARTICLES ORIGINAL ARTICLES
phlebotomy service in Johannesburg; and National Health
Laboratory Service (South Africa) — paid leave for South African
research team and laboratory investigations.

Other sponsors were: Abbott Diagnostics — loan of haematology
analyser plus reagents; ILEX Medical South Africa — laboratory
reagents; and Airports Company South Africa — logistical support
and accommodation at Johannesburg International Airport.

The authors acknowledge Naseem Cassim for data management,
Peter Cleaton-Jones for ethical advice, Pierre Du Rand for data
collection, Dave Eastham for logistical support, Deborah Jacobson
for data collection and logistical support, Hilda Jacobson for
critical
review and acronym design, Maurizio Mariano for legal advice and
support, Barry Mendelow for critical review, Sandra Ryan for data
collection, logistical support and typing, and Gwynn Stevens for
PCR analysis.

References

1.
Landgraf H, Vanselow B, Schulze-Huermann D, Mulmann MV, Bergau L.
Economy class
syndrome: rheology, fluid balance and lower leg edema during a
simulated 12-hour long
distance flight. Aviat Space Environ Med 1994; 65: 930-935.
2.
Ferrari E, Chevallier T, Chapelier A, Baudouy M. Travel as a risk
factor for venous
thromboembolic disease: a case control study. Chest 1999; 115:
440-444.
3.
Scurr JH, Machin SJ, Bailey-King S, Mackie IJ, McDonald S, Coleridge
Smith PD. Frequency
and prevention of symptomless deep-vein thrombosis in long-haul
flights: a randomised
trial. Lancet 2001; 357: 1485-1489.
4.
Kraaijenhagen RA, Haverkamp D, Koopman MMW, Prandoni P, Piovella F,
Büller H. Travel
and the risk of venous thrombosis. Lancet 2000; 356: 1492-1493.
5.
Belcaro G, Gerouklas G, Nicolaides AN, Myers KA, Winford M. Venous
thromboembolism
from air travel:the LONFLIT study. Angiology 2001; 52: 369-374.
6.
Lapostolle F, Surget V, Borron S,et al. Severe pulmonary embolism
associated with air travel.
N Engl JMed 2001; 345: 779-783.
7.
Mendis S, Yach D, Alwin A. Air travel and venous thromboembolism. Bull
World Health
Organ 2002; 80: 403-406.
8.
THRIFT Consensus Group. Risk of and prophylaxis for venous
thromboembolism in hospital
patients. BMJ 1992; 305: 567-574.
9.
Michiels JJ, Kasbergen H, Oudega R, et al. Exclusion and diagnosis of
deep vein thrombosis
in outpatients by sequential non-invasive tools. Int Angiol 2002; 21:
9-19.
10.
Bounameaux H, de Moerloose P, Perrier A, Reber G. Plasma measurement
of D-dimer as a
diagnostic aid in suspected venous thromboembolism: an overview.
Thromb Haemost 1994; 71:
1-6.
11.
Janssen MC, Wollersheim H, Verbruggen B, Novakova IR. Rapid D-dimer
assays to exclude
deep venous thrombosis and pulmonary embolism: current status and new
developments.
Semin Thromb Hemost 1998; 24: 393-400.
12.
Dubreuil Lastrucci RM, Dawson DA, Bowden JH, Munster M. Development of
a simple
multiplex polymerase chain reaction for simultaneous detection of
factor V Leiden and
prothrombin 202010Amutations. Mol Diagn 1999; 4: 247-250.
13.
von Ahsen N, Oellerich M, Schutz E. A method for homogeneous colour-
compensated
genotyping of factor V (G1691A) and methylenetetrahydrofolate
reductase (C677T)
mutations using real-time multiplex fluorescence PCR. Clin Biochem
2000; 33: 535-539.
14.
von Ahsen N, Schutz E, Armstrong VW, Oellerich M. Rapid detection of
prothrombotic
mutations of prothrombin (G20210A), factor V(G1691A), and
methylenetetrahydrofolate
reductase (C677T) by real-time fluorescence PCR with the LightCycler
TM (Technical Brief).
Clin Chem 1999; 44: 694-696.
15.
Eklof B, Kistner RL, Masuda EM, Sonntag BV, Wong HP. Venous
thromboembolism in
association with prolonged air travel. Dermatol Surg 1996; 22:
637-641.
16.
Jacobson BF, Philippides M, Malherbe M, Becker P. Risk factors for
deep vein thrombosis in
short haul cockpit crews: a prospective study. Aviat Space Environ Med
2002; 73: 481-484.
17.
Hughes R, Hill S, Hopkins R, et al. The incidence of travellers
thrombosis in low to moderate
risk air travellers:an interim analysis of the NZATT study.
Proceedings of the Fourth Pacific
Vascular Symposium on Venous Disease, 12-16 November 2002, Hawaii,
USA.
18.
Freyburger G, Trillaud H, Labrouche S, et al. D-dimer strategy in
thrombosis exclusion.
Thromb Haemost 1998; 79: 32-37.
19.
Heit JA, Minor TA, Andrews JC, Larson DR, Li H, Nichols WL.
Determinants of plasma
fibrin D-dimer sensitivity for acute pulmonary embolism as defined by
pulmonary
angiography. Arch Pathol Lab Med 1999; 123: 235-240.
20.
Gaffney PJ, Brasher M, Lord K, et al. Fibrin subunits in venous and
arterial
thromboembolism. Cardiovasc Res 1976; 10: 421-426.
21.
Bernardi E, Prandoni P, Lensing AWA, et al. D-dimer testing as an
adjunct to ultrasonography
in patients with clinically suspected deep vein thrombosis:prospective
cohort study. BMJ
1998; 317: 1037-1040.
22.
Rees DC. The population genetics of factor V Leiden (Arg 506Gln). Br J
Haematol 1996; 95:
579-586.
23.
World Health Organisation Collaborative Study of Cardiovascular
Disease and Steroid
Hormone Contraception. Venous thromboembolic disease and combined oral
contraceptives:
results of international multicentre case-control study. Lancet 1995;
346: 1575-1582.
24.
Rosendaal FR, Helmerhorst FM, Vandenbroucke JP. Female hormones and
thrombosis.
Arterioscler Thromb Vasc Biol 2002; 22: 201-210.
25.
Mangano DT for the Multicenter Study of Perioperative Ischemia
Research Group. Aspirin
and mortality from coronary bypass surgery. N Engl JMed 2002; 347:
1309-1317.
26.
Samama MM. Advances and perspectives in the prevention of venous
thromboembolic
disease. Arch Mal Coeur Vaiss 2001; 94: suppl, 1313-1317.
Accepted 17 May 2003.

528
July 2003, Vol. 93, No. 7 SAMJ



"

"Peter Brooks"  wrote in message 
news:d69cdce7-dd3b-46d0-898d-962857e6a908@a1g2000hsb.googlegroups.com...
> For anybody who isn't utterly bored by this, I have put a text copy of
> the study below.
>
> Also, I'd be interested in any comments on another thing that confuses
> me. The study uses a logistical regression to estimate the affect of
> the various risk factors (http://en.wikipedia.org/wiki/
> Logistic_regression). I can't see the logic in using it in this study.
> The growth curve on which the logistical regression is based shows
> tails at both ends - which make sense, as in the wikipaedia article,
> for population growth. I don't see why the affect of age, sex, class
> of travel or pretty well anything else would be related to the
> development of thrombi in this way. Why would somebody use this
> regression method for this?

As far as I can gather, logistic regression is considered appropriate for 
modelling the probability of binary outcomes.  A number of texts point out 
that if a linear model is used, there is a limit to how far it can be 
extrapolated before it gives "impossible" probabilities (greater than 1 or 
less than 0).  It seems to me that what you refer to as the tails of the 
logistic function are inevitable for any function which is strictly 
increasing but constrained between 0 and 1.

On May 29, 2:56 am, "Philip"  wrote:
> "Peter Brooks"  wrote in message
>
> news:d69cdce7-dd3b-46d0-898d-962857e6a908@a1g2000hsb.googlegroups.com...
>
> > For anybody who isn't utterly bored by this, I have put a text copy of
> > the study below.
>
> > Also, I'd be interested in any comments on another thing that confuses
> > me. The study uses a logistical regression to estimate the affect of
> > the various risk factors (http://en.wikipedia.org/wiki/
> > Logistic_regression). I can't see the logic in using it in this study.
> > The growth curve on which the logistical regression is based shows
> > tails at both ends - which make sense, as in the wikipaedia article,
> > for population growth. I don't see why the affect of age, sex, class
> > of travel or pretty well anything else would be related to the
> > development of thrombi in this way. Why would somebody use this
> > regression method for this?
>
> As far as I can gather, logistic regression is considered appropriate for
> modelling the probability of binary outcomes.  A number of texts point out
> that if a linear model is used, there is a limit to how far it can be
> extrapolated before it gives "impossible" probabilities (greater than 1 or> less than 0).  It seems to me that what you refer to as the tails of the> logistic function are inevitable for any function which is strictly
> increasing but constrained between 0 and 1.
>
Yes, I understand that, but I do think that it is not an ideal
situation and that there is likely to be some distortions as a result.

No matter, though. I think that the real flaw to the study is that it
treats the two populations in Business and Economy class as if they
were randomly selected groups to whom different treatments were
applied. This assumption is completely false. I have suggested that a
way to get a slightly more realistic result (there is no way to avoid
the fact that various factors such a wealth, which have many
unmeasurable effects, will skew the results whatever analysis is done)
would be to reanalyse the data taking this into account. My suggestion
has been to select a partner in Economy for each person in Buisness
class that is as close as possible in all measured demographic and
health values (apart from having a DVT of course). The stats can then
be applied to these two groups (an passenger for whom there is no
close match can be excluded) and a more realistic measure of what the
effect of Business class on D-dimer elevation can be established. I'm
hoping to get the data to carry out this analysis.

On 29 May, 04:48, Peter Brooks  wrote:
> On May 29, 2:56 am, "Philip"  wrote:
>
>
>
> > "Peter Brooks"  wrote in message
>
> >news:d69cdce7-dd3b-46d0-898d-962857e6a908@a1g2000hsb.googlegroups.com...
>
> > > For anybody who isn't utterly bored by this, I have put a text copy of> > > the study below.
>
> > > Also, I'd be interested in any comments on another thing that confuses> > > me. The study uses a logistical regression to estimate the affect of
> > > the various risk factors (http://en.wikipedia.org/wiki/
> > > Logistic_regression). I can't see the logic in using it in this study.> > > The growth curve on which the logistical regression is based shows
> > > tails at both ends - which make sense, as in the wikipaedia article,
> > > for population growth. I don't see why the affect of age, sex, class
> > > of travel or pretty well anything else would be related to the
> > > development of thrombi in this way. Why would somebody use this
> > > regression method for this?
>
> > As far as I can gather, logistic regression is considered appropriate for
> > modelling the probability of binary outcomes.  A number of texts point out
> > that if a linear model is used, there is a limit to how far it can be
> > extrapolated before it gives "impossible" probabilities (greater than 1 or
> > less than 0).  It seems to me that what you refer to as the tails of the
> > logistic function are inevitable for any function which is strictly
> > increasing but constrained between 0 and 1.
>
> Yes, I understand that, but I do think that it is not an ideal
> situation and that there is likely to be some distortions as a result.
>
> No matter, though. I think that the real flaw to the study is that it
> treats the two populations in Business and Economy class as if they
> were randomly selected groups to whom different treatments were
> applied. This assumption is completely false. I have suggested that a
> way to get a slightly more realistic result (there is no way to avoid
> the fact that various factors such a wealth, which have many
> unmeasurable effects, will skew the results whatever analysis is done)
> would be to reanalyse the data taking this into account. My suggestion
> has been to select a partner in Economy for each person in Buisness
> class that is as close as possible in all measured demographic and
> health values (apart from having a DVT of course). The stats can then
> be applied to these two groups (an passenger for whom there is no
> close match can be excluded) and a more realistic measure of what the
> effect of Business class on D-dimer elevation can be established. I'm
> hoping to get the data to carry out this analysis.- Hide quoted text -
>
> - Show quoted text -

I doubt whether such an analysis will help.  In the absence of random
allocation to different treatment groups,  there will be doubt whether
like is being compared with like.  Why is matching cases better than
using other methods of statistical control?

Dave

On May 29, 10:38 am, Dave Smith  wrote:
>
>
> I doubt whether such an analysis will help.  In the absence of random
> allocation to different treatment groups,  there will be doubt whether
> like is being compared with like.  Why is matching cases better than
> using other methods of statistical control?
>
I agree completely that it is not going to give a definitive answer!

I believe that the advantage of matching cases is that you have people
who, apart from the Business Class related matters, are likely to have
a similar level of risk. There is evidence for this being true -
smoking, for example, doesn't affect your risk of venous thrombosis,
while height, weight, sex, HRT do. So, if you have these, and other
known and measured matters, controlled, you are left with only the
Business Class related factors - that are not all to do with travel,
as I said, wealth would be part of it.

As far as I can see this is an improvement on the current testing
because you don't have the large number of people with quite unrelated
risk profiles weighting the sample - the probable reason why Business
Class appears more dangerous.

The ideal would, of course, to arrange a series of flights where you
did randomly allocate people to the two classes - though you'd need to
be sure that the crew didn't know in case their treatment of
passengers was part of the effect.

Philip wrote:
> "Peter Brooks"  wrote in message
> news:d69cdce7-dd3b-46d0-898d-962857e6a908@a1g2000hsb.googlegroups.com...
> > For anybody who isn't utterly bored by this, I have put a text copy of
> > the study below.
> >
> > Also, I'd be interested in any comments on another thing that confuses
> > me. The study uses a logistical regression to estimate the affect of
> > the various risk factors (http://en.wikipedia.org/wiki/
> > Logistic_regression). I can't see the logic in using it in this study.
> > The growth curve on which the logistical regression is based shows
> > tails at both ends - which make sense, as in the wikipaedia article,
> > for population growth. I don't see why the affect of age, sex, class
> > of travel or pretty well anything else would be related to the
> > development of thrombi in this way. Why would somebody use this
> > regression method for this?
>
> As far as I can gather, logistic regression is considered appropriate for
> modelling the probability of binary outcomes.  A number of texts point out
> that if a linear model is used, there is a limit to how far it can be
> extrapolated before it gives "impossible" probabilities (greater than 1 or
> less than 0).  It seems to me that what you refer to as the tails of the
> logistic function are inevitable for any function which is strictly
> increasing but constrained between 0 and 1.

My understanding is that the logit transformation G(x) used in
Logistic regression ranges beween -$B!g(B and +$B!g(B, whereas the original $B&P(B(x)
is constrained between zero and one. This is one of the reasons for
using Logistic regression.

Lance

On May 29, 5:45 pm, Lance  wrote:
>
>
> My understanding is that the logit transformation G(x) used in
> Logistic regression ranges beween -$B!g(B and +$B!g(B, whereas the original $B&P(B(x)
> is constrained between zero and one. This is one of the reasons for
> using Logistic regression.
>
Yes, that's true. There are other distributions, though, that can be
constrained between nought and one. I just happens that this is an
easy one to compute. It resembles the top half of a hysteresis curve.
I don't disagree that it is convenient, useful and possibly the best
to be had at the moment, but I can't believe that it is optimal. Maybe
I'm wrong and there is a proof that it is and that the top and bottom
tails don't distort the data in any meaningful way - if so, I'd be
interested to learn that.

Peter Brooks wrote:
> On May 29, 5:45 pm, Lance  wrote:
> >
> >
> > My understanding is that the logit transformation G(x) used in
> > Logistic regression ranges beween -$B!g(B and $B!g(B, whereas the original $B&P(B(x)
> > is constrained between zero and one. This is one of the reasons for
> > using Logistic regression.
> >
> Yes, that's true. There are other distributions, though, that can be
> constrained between nought and one. I just happens that this is an
> easy one to compute. It resembles the top half of a hysteresis curve.
> I don't disagree that it is convenient, useful and possibly the best
> to be had at the moment, but I can't believe that it is optimal. Maybe
> I'm wrong and there is a proof that it is and that the top and bottom
> tails don't distort the data in any meaningful way - if so, I'd be
> interested to learn that.

My understanding is that a binary yes/no dependent variable is being
modeled with the logistic function. So what one is interested in is
the inflection point (the 50% yes, 50% no point). The tails are not
the main point of interest in fitting the ciurve. The regression tries
to fit the curve so that inflection point is correctly placed. Try
reading a book on logistic regression (or IRT in psychometrics) to see
how it works.

Lance

On May 29, 10:48 pm, Lance  wrote:
> Peter Brooks wrote:
> > On May 29, 5:45 pm, Lance  wrote:
>
> > > My understanding is that the logit transformation G(x) used in
> > > Logistic regression ranges beween -$B!g(B and +$B!g(B, whereas the original $B&P(B(x)
> > > is constrained between zero and one. This is one of the reasons for
> > > using Logistic regression.
>
> > Yes, that's true. There are other distributions, though, that can be
> > constrained between nought and one. I just happens that this is an
> > easy one to compute. It resembles the top half of a hysteresis curve.
> > I don't disagree that it is convenient, useful and possibly the best
> > to be had at the moment, but I can't believe that it is optimal. Maybe
> > I'm wrong and there is a proof that it is and that the top and bottom
> > tails don't distort the data in any meaningful way - if so, I'd be
> > interested to learn that.
>
> My understanding is that a binary yes/no dependent variable is being
> modeled with the logistic function. So what one is interested in is
> the inflection point (the 50% yes, 50% no point). The tails are not
> the main point of interest in fitting the ciurve. The regression tries
> to fit the curve so that inflection point is correctly placed. Try
> reading a book on logistic regression (or IRT in psychometrics) to see
> how it works.
>
Yes, that makes sense, I agree. But why go to all that trouble if that
is all you want? If you simply make anything less then the mean or
median of the date 0 and anything greater 1 there you are with much
less calculation and hassle.

Why go such a roundabout way to achieve something simple? My
assumption was that there must be more to it than just that...

Peter Brooks wrote:
> On May 29, 10:48 pm, Lance  wrote:
> > Peter Brooks wrote:
> > > On May 29, 5:45 pm, Lance  wrote:
> >
> > > > My understanding is that the logit transformation G(x) used in
> > > > Logistic regression ranges beween -$B!g(B and $B!g(B, whereas the original $B&P(B(x)
> > > > is constrained between zero and one. This is one of the reasons for
> > > > using Logistic regression.
> >
> > > Yes, that's true. There are other distributions, though, that can be
> > > constrained between nought and one. I just happens that this is an
> > > easy one to compute. It resembles the top half of a hysteresis curve.
> > > I don't disagree that it is convenient, useful and possibly the best
> > > to be had at the moment, but I can't believe that it is optimal. Maybe> > > I'm wrong and there is a proof that it is and that the top and bottom
> > > tails don't distort the data in any meaningful way - if so, I'd be
> > > interested to learn that.
> >
> > My understanding is that a binary yes/no dependent variable is being
> > modeled with the logistic function. So what one is interested in is
> > the inflection point (the 50% yes, 50% no point). The tails are not
> > the main point of interest in fitting the ciurve. The regression tries
> > to fit the curve so that inflection point is correctly placed. Try
> > reading a book on logistic regression (or IRT in psychometrics) to see
> > how it works.
> >
> Yes, that makes sense, I agree. But why go to all that trouble if that
> is all you want? If you simply make anything less then the mean or
> median of the date 0 and anything greater 1 there you are with much
> less calculation and hassle.
>
> Why go such a roundabout way to achieve something simple? My
> assumption was that there must be more to it than just that...

I guess you should read a book on the subject. I recommend

Alan Agresti, "An introduction to categorical data analysis" (second
edition).

Lance

On May 31, 12:10 am, Lance  wrote:
>
>
> I guess you should read a book on the subject. I recommend
>
> Alan Agresti, "An introduction to categorical data analysis" (second
> edition).
>
I might get it out of the library one day, it's a bit expensive for my
casual interest. There are a few papers on the internet discussing
this. Here's a comment from one on just this question:

"
Why use logistic regression rather than ordinary linear regression?

When I was in graduate school, people didn't use logistic regression
with a binary DV. They just used ordinary linear regression instead.
Statisticians won the day, however, and now most psychologists use
logistic regression with a binary DV for the following reasons:

If you use linear regression, the predicted values will become greater
than one and less than zero if you move far enough on the X-axis. Such
values are theoretically inadmissible.
One of the assumptions of regression is that the variance of Y is
constant across values of X (homoscedasticity). This cannot be the
case with a binary variable, because the variance is PQ. When 50
percent of the people are 1s, then the variance is .25, its maximum
value. As we move to more extreme values, the variance decreases. When
P=.10, the variance is .1*.9 = .09, so as P approaches 1 or zero, the
variance approaches zero.
The significance testing of the b weights rest upon the assumption
that errors of prediction (Y-Y') are normally distributed. Because Y
only takes the values 0 and 1, this assumption is pretty hard to
justify, even approximately. Therefore, the tests of the regression
weights are suspect if you use linear regression with a binary DV.
"

I don't see any of the above as a deeply convincing case.

On May 29, 3:20 am, Peter Brooks 
wrote:smoking, for example, doesn't affect your risk of venous
thrombosis, <<

That doesn't seem right.
Logic tells this is wrong.
Cigarette smoking causes polycythemia / hyperviscosity and
hyperviscosity / polycythemia and thrombosis are linked / higher rate
of thrombosis.

Who loves ya.
Tom


Jesus Was A Vegetarian!
http://jesuswasavegetarian.7h.com


Man Is A Herbivore!
http://tinyurl.com/a3cc3


 DEAD PEOPLE WALKING
 http://tinyurl.com/zk9fk



> On May 29, 10:38 am, Dave Smith  wrote:
>
> > I doubt whether such an analysis will help.  In the absence of random
> > allocation to different treatment groups,  there will be doubt whether> > like is being compared with like.  Why is matching cases better than
> > using other methods of statistical control?
>
> I agree completely that it is not going to give a definitive answer!
>
> I believe that the advantage of matching cases is that you have people
> who, apart from the Business Class related matters, are likely to have
> a similar level of risk. There is evidence for this being true -
> smoking, for example, doesn't affect your risk of venous thrombosis,
> while height, weight, sex, HRT do. So, if you have these, and other
> known and measured matters, controlled, you are left with only the
> Business Class related factors - that are not all to do with travel,
> as I said, wealth would be part of it.
>
> As far as I can see this is an improvement on the current testing
> because you don't have the large number of people with quite unrelated
> risk profiles weighting the sample - the probable reason why Business
> Class appears more dangerous.
>
> The ideal would, of course, to arrange a series of flights where you
> did randomly allocate people to the two classes - though you'd need to
> be sure that the crew didn't know in case their treatment of
> passengers was part of the effect.

On May 31, 7:46 am, "ironjust...@aol.com" <ironjust...@aol.comwrote:>
cigarette polycythemia / hyperviscosity <<

AccessMedicine - Content If left uncontrolled, erythrocytosis can lead
to thrombosis involving vital organs such as the liver, heart, brain,
or lungs. ...
www.accessmedicine.com/content.aspx?aid=2865889
------------------------
The American Journal of Gastroenterology
Volume 97, Issue 5 , May 2002, Pages 1264-1265


Letter to the editor
Heavy cigarette smoking induces hypoxic polycythemia (erythrocytosis)
and hyperuricemia in chronic hepatitis C patients with reversal of
clinical symptoms and laboratory parameters with therapeutic
phlebotomy


Abdel-Rahman El-Zayadi M.D.a, , Osaima Selim M.D.b, Hassan Hamdy
M.D.a,
Ahmed El-Tawil M.D.c and Hanaa Moustafa M.D.d


a Department of Tropical Medicine, Ain Shams University, Cairo, Egypt
b Department of Clinical Pathology, Ain Shams University, Cairo,
Egypt
c Department of Pathology, Ain Shams University, Cairo, Egypt
d Department of Tropical Medicine, National Liver Institute,
Menoufeya,
Egypt


Received 21 November 2001;  accepted 3 December 2001.  Available
online
1 May 2002.


 Reprint requests and correspondence: Abdel-Rahman El-Zayadi, M.D.,
Cairo Liver Center, 5, El Gergawy Street, Dokki-Giza, Cairo, Egypt


The American Journal of Gastroenterology
Volume 97, Issue 5 , May 2002, Pages 1264-1265

doi:10.1016/S0002-9270(02)04074-1
Copyright © 2002 Am. Coll. of Gastroenterology. Published by Elsevier
Science Inc.

Who loves ya.
Tom


Jesus Was A Vegetarian!
http://jesuswasavegetarian.7h.com


Man Is A Herbivore!
http://tinyurl.com/a3cc3


 DEAD PEOPLE WALKING
 http://tinyurl.com/zk9fk


> On May 29, 3:20 am, Peter Brooks 
> wrote:smoking, for example, doesn't affect your risk of venous
> thrombosis, <<
>
> That doesn't seem right.
> Logic tells this is wrong.
> Cigarette smoking causes polycythemia / hyperviscosity and
> hyperviscosity / polycythemia and thrombosis are linked / higher rate
> of thrombosis.
>
> Who loves ya.
> Tom
>
> Jesus Was A Vegetarian!http://jesuswasavegetarian.7h.com
>
> Man Is A Herbivore!http://tinyurl.com/a3cc3
>
>  DEAD PEOPLE WALKING
>  http://tinyurl.com/zk9fk
>
>
>
> > On May 29, 10:38 am, Dave Smith  wrote:
>
> > > I doubt whether such an analysis will help.  In the absence of random
> > > allocation to different treatment groups,  there will be doubt whether> > > like is being compared with like.  Why is matching cases better than
> > > using other methods of statistical control?
>
> > I agree completely that it is not going to give a definitive answer!
>
> > I believe that the advantage of matching cases is that you have people
> > who, apart from the Business Class related matters, are likely to have
> > a similar level of risk. There is evidence for this being true -
> > smoking, for example, doesn't affect your risk of venous thrombosis,
> > while height, weight, sex, HRT do. So, if you have these, and other
> > known and measured matters, controlled, you are left with only the
> > Business Class related factors - that are not all to do with travel,
> > as I said, wealth would be part of it.
>
> > As far as I can see this is an improvement on the current testing
> > because you don't have the large number of people with quite unrelated
> > risk profiles weighting the sample - the probable reason why Business
> > Class appears more dangerous.
>
> > The ideal would, of course, to arrange a series of flights where you
> > did randomly allocate people to the two classes - though you'd need to
> > be sure that the crew didn't know in case their treatment of
> > passengers was part of the effect.- Hide quoted text -
>
> - Show quoted text -

On May 31, 4:46 pm, "ironjust...@aol.com"  wrote:
> On May 29, 3:20 am, Peter Brooks 
> wrote:smoking, for example, doesn't affect your risk of venous
> thrombosis, <<
>
> That doesn't seem right.
> Logic tells this is wrong.
> Cigarette smoking causes polycythemia / hyperviscosity and
> hyperviscosity / polycythemia and thrombosis are linked / higher rate
> of thrombosis.
>
Thrombosis, yes, but not venous thrombosis, arterial thrombosis.

On May 31, 8:11 am, Peter Brooks 
wrote:Thrombosis, yes, but not venous thrombosis, arterial thrombosis.
<<

These guys seem to show the treatment of polycythemia leads to a
decrease in venous thrombosis ..

http://doctor.medscape.com/viewarticle/496544
Furthermore, in the only randomized controlled clinical trial to date,
in patients older than age 60 years, hydroxyurea was found to be more
effective than aspirin in preventing transient ischemic attacks,
digital microvascular ischemia, and venous thrombosis.[30]

 These data indicate that idiopathic erythrocytosis is not rare and
that phlebotomy alone is both necessary and sufficient to prevent
thrombosis in such patients.


Who loves ya.
Tom


Jesus Was A Vegetarian!
http://jesuswasavegetarian.7h.com


Man Is A Herbivore!
http://tinyurl.com/a3cc3


 DEAD PEOPLE WALKING
 http://tinyurl.com/zk9fk


> On May 31, 4:46 pm, "ironjust...@aol.com"  wrote:> On May 29, 3:20 am, Peter Brooks 
> > wrote:smoking, for example, doesn't affect your risk of venous
> > thrombosis, <<
>
> > That doesn't seem right.
> > Logic tells this is wrong.
> > Cigarette smoking causes polycythemia / hyperviscosity and
> > hyperviscosity / polycythemia and thrombosis are linked / higher rate
> > of thrombosis.
>
>

On May 31, 6:40 pm, "ironjust...@aol.com"  wrote:
>
> Furthermore, in the only randomized controlled clinical trial to date,
> in patients older than age 60 years, hydroxyurea was found to be more
> effective than aspirin in preventing transient ischemic attacks,
> digital microvascular ischemia, and venous thrombosis.[30]
>
Most opinion is against the use of asprin for venous thrombosis - it
can cause more problems flying than not taking it - so I'm not
surprised about that part of the study.

"Peter Brooks"  wrote in message 
news:d5d28505-6517-4ced-a3db-b3aa28b59e42@m44g2000hsc.googlegroups.com...
>On May 31, 12:10 am, Lance  wrote:
>>
>>
>> I guess you should read a book on the subject. I recommend
>>
>> Alan Agresti, "An introduction to categorical data analysis" (second
>> edition).
>>
>I might get it out of the library one day, it's a bit expensive for my
>casual interest. There are a few papers on the internet discussing
>this. Here's a comment from one on just this question:
>
>"
>Why use logistic regression rather than ordinary linear regression?
>
>When I was in graduate school, people didn't use logistic regression
>with a binary DV. They just used ordinary linear regression instead.
>Statisticians won the day, however, and now most psychologists use
>logistic regression with a binary DV for the following reasons:
>
>If you use linear regression, the predicted values will become greater
>than one and less than zero if you move far enough on the X-axis. Such
>values are theoretically inadmissible.
>One of the assumptions of regression is that the variance of Y is
>constant across values of X (homoscedasticity). This cannot be the
>case with a binary variable, because the variance is PQ. When 50
>percent of the people are 1s, then the variance is .25, its maximum
>value. As we move to more extreme values, the variance decreases. When
>P=.10, the variance is .1*.9 = .09, so as P approaches 1 or zero, the
>variance approaches zero.
>The significance testing of the b weights rest upon the assumption
>that errors of prediction (Y-Y') are normally distributed. Because Y
>only takes the values 0 and 1, this assumption is pretty hard to
>justify, even approximately. Therefore, the tests of the regression
>weights are suspect if you use linear regression with a binary DV.
>"
>
>I don't see any of the above as a deeply convincing case.

Surely a statistical analysis that isn't based on flawed assumptions is 
preferable to one that is.

On Jun 1, 6:07 am, "Philip"  wrote:
> "Peter Brooks"  wrote in message
>
> news:d5d28505-6517-4ced-a3db-b3aa28b59e42@m44g2000hsc.googlegroups.com...
>
>
>
> >On May 31, 12:10 am, Lance  wrote:
>
> >> I guess you should read a book on the subject. I recommend
>
> >> Alan Agresti, "An introduction to categorical data analysis" (second
> >> edition).
>
> >I might get it out of the library one day, it's a bit expensive for my
> >casual interest. There are a few papers on the internet discussing
> >this. Here's a comment from one on just this question:
>
> >"
> >Why use logistic regression rather than ordinary linear regression?
>
> >When I was in graduate school, people didn't use logistic regression
> >with a binary DV. They just used ordinary linear regression instead.
> >Statisticians won the day, however, and now most psychologists use
> >logistic regression with a binary DV for the following reasons:
>
> >If you use linear regression, the predicted values will become greater
> >than one and less than zero if you move far enough on the X-axis. Such
> >values are theoretically inadmissible.
> >One of the assumptions of regression is that the variance of Y is
> >constant across values of X (homoscedasticity). This cannot be the
> >case with a binary variable, because the variance is PQ. When 50
> >percent of the people are 1s, then the variance is .25, its maximum
> >value. As we move to more extreme values, the variance decreases. When
> >P=.10, the variance is .1*.9 = .09, so as P approaches 1 or zero, the> >variance approaches zero.
> >The significance testing of the b weights rest upon the assumption
> >that errors of prediction (Y-Y') are normally distributed. Because Y
> >only takes the values 0 and 1, this assumption is pretty hard to
> >justify, even approximately. Therefore, the tests of the regression
> >weights are suspect if you use linear regression with a binary DV.
> >"
>
> >I don't see any of the above as a deeply convincing case.
>
> Surely a statistical analysis that isn't based on flawed assumptions is
> preferable to one that is.
>
Yes, of course, in general, that must be the case - particularly if
the assumptions make a difference! I don't think that

It, though, the assumptions don't make a difference and the one based
on them is quicker and easier to calculate, then there is an argument
for using the one based on the flawed assumptions.

On May 31, 10:35 am, Peter Brooks 
wrote:hydroxyurea was found to be more
effective than aspirin in preventing transient ischemic attacks,
 digital microvascular ischemia, and venous thrombosis.[30]
 Most opinion is against the use of asprin for venous thrombosis - it
 can cause more problems flying than not taking it - so I'm not
surprised about that part of the study.<<

That would give credence to the **erythrocytosis / hypobaric hypoxia /
altitude induced polycythemia IN the pathogenesis of the thrombosis
due