8
P.N. Lee et al. / Regulatory Toxicology and Pharmacology 70 (2014) 7–23
on estimates of a dummy variable set to zero pre-ban and to one
post-ban. The two approaches produce identical estimates where
there is only one post-ban period. The modified approach allows
us to fit non-linear forms for the trend, such as the quadratic.
Fourth, we test the validity of an assumption we used earlier
(Lee and Fry, 2011). In these analyses, where data for a run of sim-
ilar periods (usually years) were available pre-ban, we estimated
the ban effect based on numbers of cases, assuming that linear
trend adjustment would automatically take into account changes
in population size. This assumption is not necessarily valid, so
we have also carried out analyses based on trends in rates. This
often involved obtaining population data from other sources.
Finally, we also include results of meta-analyses comparing ban
effect estimates according to measures of the change in smoking
restrictiveness following the ban. This better reflects the situation
where bans may vary in the extent to which they limit smoking,
and may be conducted against a background of various levels of
existing restrictiveness.
44 points in states with casinos. Scores were then adjusted down
for pre-emption or up according to the percentage of the popula-
tion covered by local ordinances. Ratings under the modified sys-
tem are available up to 2013 (e.g., American Lung Association,
2013).
The Tobacco Control Scale, introduced by Joossens and Raw
(2006), included a section on smoke-free work and public places.
A score of 10 points was awarded for workplaces (excluding cafes
and restaurants), 8 points for cafes and restaurants, and 4 points
for other public places (trains, other public places and educational,
health, government and cultural places), giving a maximum of 22
points. Ratings were given for 30 European countries in 2005,
which have twice been updated (Joossens and Raw, 2007, 2011),
although referring to ‘‘bars’’ rather than ‘‘cafes’’. Ratings using the
same scheme were also given by Nguyen et al. (2012) for 11
European countries, annually from at least 1990–2010.
2.3. General approach
In many ways, the approach used is similar to that we used our
earlier (Lee and Fry, 2011). Thus:
2. Methods
2.1. Literature searches
ꢀ We estimate the effect of the ban by comparing the observed
number of AMI cases post-ban with that expected in the
absence of a ban, referring to the ratio as the ‘‘ban effect’’ or
the ban relative risk (RR).
ꢀ We consider it essential to account for the tendency for the risk
of AMI to vary seasonally by year (Ornato et al., 1996), by com-
paring numbers pre- and post-ban for whole years or the same
periods in a year (e.g., June to November), or by using results
which have adjusted for season or factors believed to cause
seasonal variation (e.g., temperature, humidity and influenza
rates). Studies taking no account of seasonal variation, e.g.,
comparing five months pre-ban and five months post-ban, are
rejected.
ꢀ Where possible, we attempt to adjust for any underlying time
trend in AMI rates. One method of doing this uses data for a
control population where trends are likely to be similar.
Another requires data for multiple similar time periods, in order
to estimate the trend. Where estimates can be obtained both by
use of a control population and by adjusting for trend, we prefer
to use the former as the shape of the trend is not always
well-defined. However, results are presented based on both
approaches.
ꢀ Consideration should be given to specific factors that might
affect the time trend, such as changes in diagnostic criteria.
ꢀ As the great majority of studies consider the post-ban period as
starting immediately or just after the ban, we derive estimates
on this basis where possible.
ꢀ Where a study provides data for multiple control populations,
the ban effect is generally estimated from the combined control
data. However, control populations with obvious weaknesses
may be excluded.
ꢀ Some studies report results for subgroups by sex, age, or smok-
ing habit. For consistency, the estimates we use in our meta-
analyses are always based on the result for the whole study
population, and not on that for subsets. However, we summa-
rize the availability of such data. Exceptionally, where studies
present results relating to different ban times in different areas,
we report these separately.
ꢀ The mathematical methods we use assume that the effect of a
ban is to multiply the risk of AMI by a given factor, with the fac-
tor invariant of the length of time post-ban. The validity of this
assumption is investigated by comparing the estimates of the
magnitude of the ban effect in studies with shorter and longer
post-ban periods.
Published studies and reviews relating smoking bans to risk of
AMI (or heart disease) additional to those considered earlier (Lee
and Fry, 2011) were sought from PubMed searches (January 1st
2009 to September 30th 2013) using the terms described by
Mackay et al. (2010), and also from papers cited in relevant
publications.
2.2. Quantifying levels of restrictiveness
Except for local US studies, and for studies presenting overall
results based on multiple bans in different locations, we sought
published scores for restrictiveness before and after the ban, using
for US studies the method of Chriqui et al. (2002) without pre-
emption (as explained below), or a modification of it (American
Lung Association, 2009), and for European studies the method of
Joossens and Raw (2006), re-expressing the scores as percentages.
Although the different ratings are not strictly comparable, this
method gives a reasonably detailed assessment of the legislation
in a variety of different environments, and of the level of change
expressed by the introduction of the ban. Where published scores
were unavailable, we conducted internet searches to supplement
the descriptions of the ban given in the study publication(s), and
estimated the scores using the Chriqui system.
The system of Chriqui et al. (2002) allocated a score of 4 points
for each of seven locations (government worksites, private work-
sites, schools, childcare facilities, restaurants including bar areas
of restaurants, retail stores and businesses, recreational and cul-
tural facilities), a bonus point for restrictions on outdoor smoking
restrictions in four of the locations (including outdoor seating at
bars and taverns under the restaurant category), and a further 5
points each for systems of penalties and enforcement, giving a
maximum score of 42 points. Points were deducted if states pre-
empted stricter local laws. Chriqui et al. (2002) gave ratings for
all states annually for 1993–1999, both with and without adjust-
ment for pre-emption, and the annual reports of the American
Lung Association published ratings without pre-emption for
2003–2006 (e.g., American Lung Association, 2008). In a later
report (American Lung Association, 2009), a modification to the
rating system gave 4 points to each of the original categories,
and allocated 4 points each to bars/taverns (in addition to the 4
points for restaurants and their bar areas) and to casinos where rel-
evant, giving a maximum of 40 points in states without casinos, or
Abutaleb, Nader S.
