Legislation for smoke-free workplaces and health of bar workers in Ireland: before and after study
BMJ 2005; 331 doi: https://doi.org/10.1136/bmj.38636.499225.55 (Published 10 November 2005) Cite this as: BMJ 2005;331:1117
All rapid responses
Editor - The introduction of a comprehensive smoke-free law in all
indoor workplaces in Ireland – including bars and restaurants – has led to
substantial reduction in exposure to second-hand tobacco and to
respiratory symptoms of bar workers.1 Such legislation is also supported
by the large majority of the population (i.e., 90.4% in Italy where a
similar law was introduced in 2005) and, if anything, favourably affects
the business of bars and restaurants (9.6% of the population in Italy
reported to go more frequently, and 7.4% less frequently to bars and
restaurants after the tobacco ban).2,3
No such ban of smoking in indoor public places has yet been adopted
in Switzerland. Thus, we considered data of a survey on smoking conducted
in September 2005 by the DOXA, a branch of the Gallup International
Association, on a sample of 2000 subjects aged 15 to 74, representative of
the Swiss adult population in terms of age, sex, and geographic area. The
data were collected by ad hoc trained interviewers using a structured
computer–assisted telephone interview.
In the adult Swiss population, 22.3% were current smokers (23.6% of
men, 21.1% of women), and 19.1% former smokers (22.3% of men, 16.2% of
women). Overall, 76.8% of the Swiss population was favourable to a total
ban of smoking in all indoor public places, 62.2% in all workplaces
(including private ones), and 64.1% in bars and restaurants. Among
smokers, corresponding proportions were 61.9%, 49.5% and 34.1%, and among
ex-smokers 79.6%, 62.6% and 68.4%.
This survey confirms that support for smoke-free policies is
widespread in the populations of many European countries, and stresses
therefore the urgency for adequate political intervention in this vital
public health issue.4
Fabio Levi, MD
Cancer Epidemiology Unit and Cancer Registries of Vaud and Neuchâtel,
Institut universitaire de médecine sociale et préventive, Bugnon 17, 1005
Lausanne, Switzerland fabio.levi@chuv.ch
Paolo Colombo, DSc
Istituto DOXA, Gallup International Association, 20144 Milano, Italy
paolo.colombo@doxa.it
Carlo La Vecchia, MD
Istituto di Ricerche Farmacologiche ‘Mario Negri’ and Università degli
Studi di Milano, 20157 Milano, Italy
lavecchia@marionegri.it
References
1.Allwright S, Paul G, Greiner B, Mullally BJ, Pursell L, Kelly A,
Bonner B, D’Eath M, McConnell B, McLaughlin JP, O’Donovan D, O’Kane E,
Perry IJ. Legislation for smoke-free workplaces and health of bar workers
in Ireland: before and after study. BMJ 2005:331:1117-20.
2.La Vecchia C, Garattini S, Colombo P, Scarpino V. Attitudes towards
smoking regulations in Italy. Lancet 2001;358:245.
3.Gallus S, Zuccaro P, Colombo P, Apolone G, Pacifici R, Garattini S,
La Vecchia C. Effects of new smoking regulations in Italy. Ann Oncol
(doi:10.1093/annonc/mdj070)
4.Gray N. National and international nicotine dependence. Ann Oncol
2005;16:681-2.
Competing interests:
None declared
Competing interests: No competing interests
In Table 3, the median change in hours exposed outside work (domestic
or social) during past 7 days was given as –1.25 with 95% confidence
interval –4 to –1.5. The point estimate is outside the confidence
interval. What should it be?
Competing interests:
None declared
Competing interests: No competing interests
It seems to become common practice in the medical society to counter serious
arguments against the outcomes or methodologies of studies with ad-hominem
attacks. When opposing views do not fit in the 'modern' medical ideology, the
man is attacked and not the ball.
We saw this happening when Enstrom and Kabat's study was published in BMJ.
Dr. Michael Siegel, a respected anti-smoking researcher who started to attack
the manipulations of the anti-smoking movement in
his blog, never gets answers
to the questions he raises but instead gets attacked as a 'front for tobacco
companies'.
And this time again, Mr. Piette's only defense against the disclosure of a
real flaw in this study doesn't consist of any counter argument but an attack on
the source of the criticism, of whom he even doesn't know the scientific
background.
The criticism from the opponent is qualified as an 'opinion' while obviously
his own personal interest (considering his own background as a 'health promotion
& education' professor) doesn't represent an opinion at all. I know that in
Belgium they also call this 'arrogance'.
Back to my criticism of this study: it IS 'sound science' until the
researchers jump into a conclusion that does not cover the outcome of the study.
All scientific studies following the scientific method start with an H0
hypothesis that describes the opposite of what we expect a study to find. In
this case this H0 would have read: "removing environmental tobacco smoke ban
does not influence the amount of cotinine in a human body". What this study
proofs (alternative hypothesis) is that it DOES influence the cotinine level.
That's all. The word 'protect' in the conclusion implies a kind of damage though.
If we would have compared groups of people that were eating a lot of potatoes,
tomatoes, aubergines and other Solanaceae with a group of people who never eat
these vegetables we also would have found a lower cotinine level. Is the group
of non-Solanaceae eaters then 'protected' against these vegetables?
A next question one should raise with this study is whether a cotinine level of 20.4
nmol/l is more dangerous than a level of 5.1 nmol/l. Where is the calibration
line that represents real danger? Nobody knows, but probably it's very much
higher than any of these two figures.
It is a good case that BMJ allows criticism from outside the medical society as
it may counter the inbreed-effect of a closed community. Ideology may be
blinding for people living inside a community. And, according to psychological
studies, blinding causes overreactiveness to opposing views. I consider the
latest reaction as an example of such an effect. On the other hand, I cannot
believe that a fancy study like this can be published in a respected journal
like BMJ...
Competing interests:
None declared
Competing interests: No competing interests
Wiel Maessen's comment was actually a fairly decent appraisal of this
study. Basically the investigators discovered two things:
1) When you ban smoking in a place there's not as much smoke in the
air so people in that place don't breathe in as much smoke. Somehow, even
without a grant to do a study, I think I could conjecture that people
standing on a desert island in the middle of the Pacific ocean would
generally have less car exhaust to breathe in than people in midtown
Manhattan.
and
2) If you remove smoke from the air, the kinds of irritations that
people complain about in smoky places are reduced, particularly if the
people questioned about it may have been more self selected than the
average of the whole pool toward favorable ban feelings.
The study says nothing about long term health, and says nothing about
whether levels of nicotine, cotinine, or other substances were harmful in
any way past immediate symptomologies of irritation. Additionally the
study pool was quite likely to be biased in favor of bar staff without
hostility toward the smoking ban, and indeed in favor of bar staff who
were sympathetic to it and far more likely to report symptomatic
improvements than the average.
There have now been several studies carried out similar to this, all
to the best of my knowledge sharing roughly the same defect of self-
selection and likely bias in reporting of symptomology and all carried out
with the researchers probably knowing fairly well beforehand what results
to expect. The value of the studies seems far more political than
medical. I realize this sounds like a very harsh assessment and that
researchers must take grant money where they can find it, but the
continued production of studies like this ilustrates the problem that
occurs when special interest groups have disproportionate influence in
controlling the funding and thereby the direction and determination of
what counts as important in medical research.
Michael J. McFadden,
Author of "Dissecting Antismokers' Brains",
www.Antibrains.com
Competing interests:
I do not now, nor have I ever, had any financial connections to or interest in the tobacco industry other than as a customer. I am the author of a book titled "Dissecting Antismokers' Brains" and am active in a number of Free-Choice groups although I have no financial connections to any of them other than as an occasional contributor.
Competing interests: No competing interests
The article by Shane Allwright and colleagues has sound scientific
basis and protocol. On the contrary, the answer by Wiel Maessen, is based
on emotion from this President of Forces Netherlands, Holland
(http://www.forces-nl.org/), a pro-smoking internet site.
Is it the role of the BMJ to publish such type of comments? Or is it an
ethical question to accept opinion as well as science?
Competing interests:
None declared
Competing interests: No competing interests
Playing on the naive by waving exposure measurements that show a
reduction in cotinine levels -- a byproduct of nicotine -- as evidence
that Ireland has started to save lives is indefensible. When one says
something believed to be "bad" has been reduced then it surely sounds like
an improvement. However, in the scientific world there are two thoughts
in this respect: 'The dose makes the poison' and 'exposure does not equal
harm'.
When you want to lower cotinine levels, you can as well forbid to eat
tomatoes as they also generate cotinine in the body.
No matter how you slice it, this is an exercise in trying to convince
everyone that if the water level we're standing in is reduced from two
feet to one, we've been saved from drowning.
Competing interests:
None declared
Competing interests: No competing interests
Not a typographical error
EDITOR-With regard to our recent paper,1 we thank Professor Bland2
for pointing out that the reported point estimate for the median
difference between baseline and follow-up (for the Republic of Ireland
responders) for the variable “hours exposed outside work (domestic or
social) during there last 7 days” (Table 3 long / Table 1 short version)
did not lie within the associated confidence interval. The median
difference was given as –1.25 hours with a confidence interval of–4 to
–1.5 hours. The possibility of a typographical error was implied. In
fact, in one sense at least, both sets of figures are correct.
The confidence interval for a median may be calculated in several
different ways, depending on certain assumptions. Consideration of both
sample size and the distribution of the observations should influence the
choice of method. The variable in question is highly unusual. The data
are not normally distributed although they display a measure of symmetry.
The distribution is heavy tailed and there is also a greater than expected
concentration of values at zero. The median is –1.25 with an
interquartile range of –5.75 to 0.0. The observations are rounded to
whole digits (with only 42 distinct values out of the 136 responses) and
heavily tied (e.g. 38 ties at zero, 11 ties at – 4 and 11 ties at –5).
Given the question posed at baseline and at follow-up, it is perhaps not
surprising that many observations would indeed be tied and especially that
a value of zero would dominate.
To illustrate the practical problems in providing a reasonable
confidence interval for the median of these data, a number of recognized
methods have been employed. The resulting confidence intervals are shown
below.
Normal approximation to Binomial: -4 to 0
Binomial (default ‘centile’ command in Stata (v9, 2005, Stata Corp):
-3.9 to 0
Bootstrap (based on the 1000 bootstrapped estimates (using BCa) of
the median difference): -4 to 0
Walsh averages*: -4 to –1.5
(*This method, which was used in our paper, is recommended by Gardner
& Altman.4 It was computed using Mathematica routines accompanying
the book by Prof. J. Baglivo5 and also the macro (cid.ado) written by Dr.
Patrick Royston6 for Stata. There are (n)(n+1)/2 or 9316 Walsh averages
giving the lower bound as –4 and the upper bound as –1.5. The median of
the Walsh averages (the Hodges-Lehmann estimator) is –2.5.
The ‘centile’ command with the ‘normal’ option in Stata has not been
included because it is inappropriate in view of the distribution of these
data, in spite of the large sample size.)
We used Mathematica to calculate the confidence intervals for
differences in medians (Walsh averages) in our paper. There is good
agreement between the four methods for the lower bound of –4. However,
the estimate of the upper bound of -1.5 derived from the Walsh averages
differs from the other upper bound estimates. The point estimate of the
population median using the Hodges-Lehmann estimate, -2.5, also differs
from the observed median of –1.25 but is within the reported confidence
interval and consistent with the p-value (Wilcoxon Signed Rank test).
Which of these intervals is the more appropriate in view of the
heavily tied nature of this set of observations? Gardner & Altman3
propose the method of Walsh averages. Under these particular circumstances
of heavily tied data, Sprent7 also recommends this approach and the Hodges
-Lehmann point estimate of the population median. In the case in point,
we find this argument to be compelling.
The result of the Wilcoxon Signed Rank test comparing the baseline
and follow-up data for this variable is highly significant but is not
consistent with four of the above confidence intervals which have upper
bounds including 0. Because of the large number of tied values at 0
(close to the observed median), it is impossible for a rank-based
confidence interval method to find any other upper bound as all ranks
between the 71st and the 108th are zero.
None of the above methods for confidence intervals explicitly adjust
for ties, whereas the standard method for hypothesis testing based on the
Wilcoxon Signed Rank test adjusts the estimated variance for general tied
values and also for zeros. In this instance, such adjustments have a
pronounced effect on the variance estimate and consequently on both the
test statistic and the resulting p-value. The unadjusted variance is
211,939; the adjustment for ties reduces this by 111.25 and the adjustment
for zeros reduces the variance by a further 4,754.75 to give a final
adjusted variance of 207,073. As yet, there appears to be no standard
procedure for the adjustment of either rank-based or exact Binomial
confidence intervals in the presence of ties.
To conclude, the confidence interval method recommended by both
Gardner & Altman and Sprent based on the Walsh averages and the Hodges
-Lehmann point estimator provide a more suitable set of bounds for the
population median and the result happens to be more consistent with the
reported hypothesis test. In the published table (Table 3 long / Table 1
short version), it would have been desirable to include the Hodges-Lehmann
point estimate for the median in addition to or even in lieu of the
observed median.
Similar considerations pertain to the variables “Hours exposed at
work (pub or bar) during past 7 days” and “Hours worked in past 2 days”
(Republic only) in the same table and median number of respiratory
symptoms (Republic only) in Table 4 (long version only), as both of these
variables contain large numbers of zero ties.
The above issues have no implication for the GEE-based modeling on
which the paper’s substantive conclusions are based.
1 Allwright SPA, Paul G, Greiner B, Mullally BJ, Pursell L, et al.
Legislation for smoke-free workplaces and health of bar workers in
Ireland: before and after study. BMJ, Nov 2005; 331:1117 ;
doi:10.1136/bmj.38636.499225.55
2 Bland, M. Typographical error? BMJ Rapid responses, 14 November
2005
3 Bland, M. An Introduction to Medical Statistics, 3rd Edition,
p.133, OUP, 2000
4 Gardner, M., Altman, DG. Statistics with Confidence, p.76, BMJ,
1989
5 Baglivo, J. Mathematica Laboratories for Mathematical Statistics,
SIAM, 2005
6 Royston, P. “cid.ado” a macro for calculating confidence intervals
for differences, Statalist, March. 20, 1998
7 Sprent, P. Applied Nonparametric Statistical Methods, 2nd Edition,
Chapman & Hall, 1993
Competing interests:
None declared
Competing interests: No competing interests