Estimated public health gains from German smokers switching to risk-reduced alternatives: Results from population health impact modelling

Background Smoking is associated with cancer and cardiorespiratory disease mortality. Reducing smoking prevalence reduces deaths and life-years lost. Here, we estimate the impact of introducing heat-not-burn products and e-cigarettes in Germany from 1995 to 2015 on mortality from lung cancer, chronic obstructive pulmonary disease, ischaemic heart disease, and stroke in men and women aged 30–79 years. We used the previously described population health impact model. Modelling starts with individuals of a given sex and age range with a dened cigarette smoking distribution. They are then followed under a “Null Scenario”, where reduced-risk products are never introduced, and one of seven “Alternative Scenarios”, where they are. Transition probabilities allow tobacco product use to change annually, and the individual product histories then allow estimation of risks, relative to never users, for each year and Scenario, which are then used to estimate reductions in deaths and life-years lost for each Alternative Scenario.

nicotine replacement products [18], coupled with the high acceptance of ECigs by smokers wishing to replace cigarettes. At the same time, youth initiation continues to be low in the UK and New Zealand [19,20], and youth vaping only rarely seems to lead to smoking [21]. As for HnBs, it is noteworthy that, in Japan, 20% of smokers have switched to such products, which is a very plausible reason for the unprecedented drop in cigarette sales observed in Japan [22]. Taken together, these facts suggest that both ECigs and HnBs could play a role in reducing smoking-attributed morbidity and mortality.
In Germany, ECigs have been available since 2007, and current users now form about 2-3% of the population aged 14 or over [http://debra-study.info/wordpress//]. HnB products only became available much later, in 2016, with the number of users estimated to have risen from about 36,000 in 2017 to about 300,000 in 2019, then forming about 0.4% of the population aged 18 or over [https://www.pmi.com/investor-relations/overview]. The two products are predominantly used by current or former smokers, and only by very few never-smokers [4]. Among 12-to 17-year-olds, use in the last 30 days, which is not necessarily a good marker for regular use, remains low at 5.1% for ECigs and 0.1% for HnBs [23].
Our main objective is to estimate the population health impact of introducing HnBs or ECigs into Germany during 1995-2015 under various assumptions about their rate of uptake. We also compare our estimates with those derived by assuming that the whole population ceased smoking cigarettes immediately.
To avoid uncertainty about the future, and to take into account the effect of exogenous factors-such as medical progress and infectious disease epidemics-on future mortality rates, we use a "hindcasting" approach, in which individuals start in 1995, with a nationally representative distribution of cigarette smoking, and are then followed until 2015 under various assumptions. This approach has previously been applied for assessing the population health impact of introducing HnBs into the US [24,25] and Japanese [26] markets. In this study, we have considered two types of product, HnBs and ECigs. Both can be termed reduced-risk products (RRPs) -that is, products considered likely to present less risk of harm to cigarette smokers who switch to them.
The approach generates estimates of the number of smoking-related deaths (SRD) and number of years of life lost (YLL) in scenarios where RRPs are or are not introduced, the difference between the two scenarios being referred to as the drop in deaths (DD) and number of years of life saved (YLS). These are calculated separately for the four main diseases known to be related to cigarette smoking, lung cancer (LC), chronic obstructive pulmonary disease (COPD), ischaemic heart disease (IHD), and stroke. Note that, while the term SRD is normally used in reference to the additional deaths arising from the increased risk from cigarette smoking, it is here used in reference to the additional deaths arising from the increase in risk associated with the use of any of the three products considered: cigarettes, HnBs, or ECigs.

Methods
Outline of the approach used in population health impact modelling (PHIM) The basic method used for estimating the population health impact of introducing an RRP into a country is as described earlier [27] and involves two components, the Prevalence (P-) component and the Epidemiologic (E-) component.
The P-component starts in a speci ed year with a group of individuals of a given sex and age range with a de ned distribution of cigarette smoking. This group is then followed over discrete time intervals under both a "Null Scenario" and various "Alternative Scenarios", by using different sets of transition probabilities (TP). In the Null Scenario, RRPs are never introduced, and each individual's cigarette smoking status (never, current, or former) is updated at each yearly interval. In each Alternative Scenario, RRPs are introduced during follow-up, and the TPs allow for switching between six groups (never user, current exclusive cigarette smoker, current exclusive HnB user, current exclusive ECig user, current multiple product user, and former product user). "Never users" have never used cigarettes or either of the two RRPs considered. "Current multiple product users" currently use two or three of the products considered, while "former product users" have previously used at least one of the three products but do not currently use any of them. At the end of the Pcomponent, each individual has a complete history of use of the three products over the follow-up period under each Scenario. Note that the modelling ignores products other than cigarettes, ECigs, and HnBs.
The E-component then uses the product use histories to estimate, for each individual in the prede ned population, the relative risks (RR), compared to never users, of LC, COPD, IHD, and stroke for each follow-up year and Scenario. The estimation involves an extension of the negative exponential model (NEM), which allows for multiple changes in use, as fully described elsewhere [24]. The NEM requires estimates of the RR for continued smoking for each of the four diseases. It also requires estimates of the effective doses for current exclusive HnB use, exclusive ECig use, and multiple product use relative to that for current cigarette smoking (taken as one unit). The effective dose is a measure of harm, so that if cigarette smokers have excess risk (RR -1) for a disease, the excess risk for an RRP user with an effective dose of q is q (RR -1). The effective dose for an RRP is assumed to be the same for each disease. The NEM also requires estimates, for each of the four diseases, of the RR for continued smoking and the quitting half-life (H). H is the time from quitting when the excess relative risk (RR − 1) declines to half that for continuing smokers, the decline over time being assumed to follow a negative exponential distribution, as has been shown to be approximately true for LC [28], COPD [29], IHD [30], and stroke [31].
The estimation of the RR for an individual does not speci cally take into account the amount smoked, but the effective dose for multiple product users may be set to re ect a reduced consumption of cigarettes. A discussion of how the effective dose may be quanti ed for an RRP is given elsewhere [27].
For each Scenario, the average RR for each of the four diseases for individuals of a given sex and age group is then calculated for each follow-up year, from which the proportions of SRD can be derived. These are then converted to numbers by using national mortality estimates by sex, age group, and year. The differences in estimated numbers and proportions among the Scenarios then quantify the effect of RRP introduction.
For a given scenario, YLL is estimated by using the method of Gardner and Sanborn [32]. YLL (N) is calculated by summing up the product of the number of deaths in each age group by the number of years life remaining up to a given age of N years, with N taken here as 75 years. Thus, for an age group of 40-44 years, where the mean age is taken as 42.5 years, the number of years remaining is taken to be 75 − 42.5 = 32.5 years. For age groups above 70-74 years, the number of years remaining is taken as zero. YLS is then calculated from the difference in YLL between the Alternative and Null Scenarios.
Although the individuals in each Alternative Scenario might have death rates different from those in the Null Scenario, estimates of DD and YLS assume that the size of the populations of risk remains the same during follow-up, with no correction for differential survival.
These estimates of the effect of RRP introduction can be corrected for this possibility, if required [27].
The methodology can also compare the Null Scenario with Alternative Scenarios where RRPs are not introduced but where different sets of TPs for cigarette smoking are used.
The modelling starts with a population aged 10-79 years, with individuals dropping out of the calculations as they reach 80 years of age. This is partly because cause of death certi cation is unreliable at an older age and partly because our estimates of population health impact also include YLS, which is unaffected by deaths above the age of 74 years.

Common features of each simulation
Each simulation involved the follow-up of 100,000 individuals in 1-year intervals from 1995, with the product use status of each member of the simulated population being estimated at each year of follow-up until the year 2005 (or age 79, if that came earlier). For each situation described in the Methods section, separate simulations were conducted for each sex.
Population at baseline As previously described [24], each individual in a simulation is randomly allocated at the start of the P-component to a year of age, then to a cigarette smoking group (never, current, or former), and, then, for former smokers, to an age of quitting.
The sex-speci c age distributions used for Germany for 1995 are as published by the United Nations [33].
Sex-and age-speci c distributions of current and former smoking prevalence for Germany for individual years from 1995 to 2015 were derived by combining data from three sources: International Smoking Statistics [34], which provides results by 5-year age groups from 1980-2015 for current smoking; a report by Forey and Lee [35], which provides results by 15-year age groups from 1980-2005 for former smoking; and the German Socioeconomic Panel [36], which provides data for 2002 and 2012 for current and former smoking.
Only the estimates for 1995 are required for the baseline population.
The sex-and age-speci c distribution of quit time for former smokers used for the baseline population in 1995 was taken, in the absence of alternative data, from estimates for 2002 derived from the German Socioeconomic Panel [36]. Because this source only provided data for age groups 20-24 and above, the data for younger age groups were taken from US estimates for 2006 [24].
Additional File 1 gives further details on the derivation of the data on current and former smoking prevalence and quit time. It also includes tables summarizing the age-speci c distribution of the population and the data on smoking habits used to assign the initial status of each member of the simulated population in 1995.

Estimation of histories of cigarette smoking for the Null Scenario
The sex-and age-speci c TPs used in the P-component for developing the histories of cigarette smoking for the Null Scenario were derived as described in Additional File 2 and are shown in Table 1. In order to test the validity of the TPs, the prevalences predicted by using these TPs were compared with the estimates for Germany derived for years up to 2015 as described in Additional File 1. The rst period relates to 5 years starting from 1995, the second to 5 years starting from 2000, and the third to 10 years starting from 2005.
The probabilities of transition among the three states N (never), C (current), and F (former) are described by P followed by two subscripts, the rst representing the state changed from and the second the state changed to.
Note that RRPs are not introduced in the Null Scenario.

Estimation of histories of product use for the Alternative Scenarios
Seven different Alternative Scenarios were tested and are summarized in Table 2; Scenarios 1 to 3 are termed "Extreme Scenarios" and Scenarios 4 to 7 "Pragmatic" Scenarios. The Pragmatic Scenarios were designed to provide a range of possible uptake rates of HnBs and ECigs on the basis of known market data for Germany, with Scenario 6 (the "Conversion Scenario") being regarded as perhaps the most plausible one. Converted RRP users are de ned as the estimated number of Legal Age (over 18 years old) users that used the RRP for over 95% of their daily nicotine product consumption over the past 7 days. The link above also presents evidence on conversion rates. In 1995, all current cigarette smokers immediately switch to HnBs. The subsequent initiation, reinitiation, and quitting rates are as in the Null Scenario, but only involve transfers in or out of HnBs. 3 Complete switch to RRPs (50% HnBs and 50% ECigs) In 1995, all current cigarette smokers immediately switch to either HnBs or ECigs with equal probability. The subsequent rates are as in the Null Scenario, but only involve transfers involving the new products. Note: Multiple (product) users currently use at least one of the three products, while former (product) users have used at least one of the products, but do not currently use any).
The sum of the TPs for initiation and the sum of the TPs for re-initiation are the same as that for the Null Scenario. Each quitting TP is as for the Null Scenario. The difference between the four Pragmatic Scenarios only relates to the rates of switching among the three products. The comment for the Conversion Scenario applies here as well.
Abbreviations used: Cig = cigarette, ECig = e-cigarette, HnB = heat-not-burn, RRP = reduced risk product, TP = transition probability No RRP is introduced in Alternative Scenario 1. For the other six Alternative Scenarios, the effective doses are assumed to be 0.2 for exclusive HnB use and 0.05 for exclusive ECig use, in contrast to an effective dose of 1 for exclusive cigarette smoking. The value for HnBs was conservatively based on biomarkers and clinical ndings [37], and that for ECigs was based on a published expert opinion [38]. For multiple product use, the effective dose is assumed to be the mean of the three effective doses (i.e., 0.42).
The TPs used in the P-component for developing usage histories in the Alternative Scenario are presented in Additional File 3. Note that, for each Alternative Scenario, the sum of the initiation TTPs (for a given sex, age, and follow-up period) was constrained to be equal to the corresponding initiation TP for the Null Scenario. The same constraint was applied to the re-initiation TPs. Each cessation TP in the Alternative Scenario was also constrained to be equal to the cessation TP in the Null Scenario. These constraints were applied so that the various Alternative Scenarios considered only the effect of the RRPs introduced on the distribution of current smoking habits, without any effect on overall initiation, cessation, or re-initiation rates.
Estimating relative risks on the basis of product use histories For each disease, the estimates of the RR for continued cigarette smoking and of H were derived from meta-analyses of published data.
The sex-and age-speci c data on national population size for Germany for the years 1995 to 2015 are as published by the United Nations Department of Economic and Social Affairs Population Division [33].
The data on numbers of deaths in Germany from LC, COPD, IHD, and stroke come from the World Health Organization [41]. The data on population size and numbers of deaths for Germany for the years 1995 to 2015 are presented in Additional File 4, which gives full details on sources and disease de nitions.
The method of estimating the number of deaths and increase in death rates associated with tobacco is as described earlier [24]. Unless indicated, the results are presented without adjustment for changes in population size associated with each Alternative Scenario.

Results
Full results of the analyses are available in Additional File 5. Figure 1 compares never, current, and former smoking prevalence estimates for Germany by sex for age groups 30-34, 50-54, and 70-74 years as simulated in the Null Scenario (broken lines) with those derived as described in the Methods section (solid lines). The t is generally very good, though there is some tendency for the Null Scenario current smoking estimates to be lower than the derived estimates at age 70-74 years. As estimated by the P-Component of PHIM, 852,357 deaths were attributed to cigarette smoking over the whole follow-up period in the Null Scenario. 77.9% of these were in males, with the percentages by disease being 54.6% for LC, 26.4% for IHD, 13.4% for COPD and 5.7% for stroke.  Abbreviations used: COPD = chronic obstructive pulmonary disease, ECig = e-cigarette, HnB = heat-not-burn, IHD = ischaemic heart disease, LC = lung cancer The DDs in the Conversion Scenario are also shown by disease over the whole follow-up period in Fig. 3.
As expected, the largest DD for the four diseases combined is seen in the complete cessation Scenario. Substantial DDs are also seen in the complete switch Scenarios -more so in Scenario 3 than in Scenario 2, because the assumed effective dose is lower for ECigs (0.05) than it is for HnBs (0.2). The DDs are lower in the Pragmatic Scenarios, because the transition from Cigarettes to HnBs and ECigs is less rapid. As would be predicted from the patterns of uptake by Scenario shown in Fig. 2, the greatest DDs are seen in the Full Conversion Scenario, where smokers switch gradually to the RRPs -they are about 40% of the DDs associated with Complete Cessation, where smokers quit smoking immediately in 1995.
The patterns of DDs for the individual diseases are similar to that for the four diseases combined. Among men, the largest absolute DDs are for IHD, with LC next, followed by stroke and COPD with lower and similar DDs. Among women, the DDs for LC are higher than those for IHD, re ecting the lower overall IHD rate among women. As a proportion of all SRDs, the DDs in both sexes are substantially higher for IHD and stroke than for LC and COPD, re ecting the shorter half-lives for IHD and stroke (i.e., the more rapid reduction in cardiovascular disease risk after smoking cessation or switching to RRPs).
As is shown in Fig. 3 for the Conversion Scenario, but as is also clearly evident from Additional File 5 for the other Scenarios, a clear increase in DDs is seen with time in both sexes. This is due partly to the time needed for take-up of HnBs and ECigs and partly to the time required for the resulting decline in risk. This trend clearly suggests that the DDs would have been substantially greater had the follow-up period been extended.
In the Null Scenario, 8.61 million YLL were attributed to cigarette smoking. 76.2% of these were in males, with the percentages by disease being 51.6% for LC, 31.8% for IHD, 8.4% for COPD and 8.2% for stroke. The percentages by disease, compared to those given above for attributable deaths, re ect the higher proportion of deaths at younger ages for IHD and stroke than for LC and COPD. Table 5 and Fig. 4 summarize the results for the seven Scenarios with regard to YLS by age 75 over the whole follow-up period. The relative values for the different Scenarios are very similar to those for DD seen in Table 4. Indeed, on the basis of the results for the four diseases combined in Tables 4 and 5, the sex-and scenario-speci c ratios of YLS to DD only vary between 12.5 and 13.4. Abbreviations used: COPD = chronic obstructive pulmonary disease, ECig = e-cigarette, HnB = heat-not-burn, IHD = ischaemic heart disease, LC = lung cancer The analyses summarized above do not take into account the increase in population size associated with the reduced mortality in the Alternative Scenarios relative to that in the Null Scenario. As shown in the detailed results in Additional File 5, this had little effect on the estimated DD or YLS. For example, for the Conversion Scenario, where the overall unadjusted DDs were 57,655 (8.68% of all SRDs) in men and 17,942 (9.53% of SRDs) in women, the corresponding adjusted DDs were 57,026 (8.59%) and 17,892 (9.51%), respectively.

Discussion
Over a period of 20 years, from 1995 to 2015, in the absence of any switching to HnBs or ECigs and with the patterns of prevalence of cigarette smoking as existing in Germany, we estimate that there would have been 852,357 SRDs from LC, COPD, IHD, and stroke combined for both sexes, with about 8.61 million YLL.
In the work described here, we attempt to estimate how these numbers would have been affected had various alternative patterns of product use occurred. re ecting the assumed lower effective dose for ECigs compared with HnBs. These numbers would have been greater still had we included a Scenario in which cigarette smoking was immediately replaced by ECig use.
In practice, such Scenarios are rather unrealistic; more plausible are the estimates associated with the Pragmatic Scenarios in which a proportion of cigarette smokers move gradually to use HnBs and ECigs. Scenarios 4 to 7 vary in the extents to which uptake of these RRPs occurs and to which RRP users fully convert to exclusive RRP use rather than becoming multiple users of Cigarettes and RRPs. for YLS of the effect of immediate cessation (Scenario 1), the best albeit unrealistic scenario.
Our estimates may be regarded as conservative for three main reasons. The rst reason is that we only considered deaths from the four main SRDs because of the lack of reliable data on RR and H for all diseases associated with smoking. We have noted elsewhere [27] that our estimates of deaths saved would have to be multiplied by about 1.52 to yield an estimate for all smoking-related diseases.
The second, and very important, reason is that we only considered a 20-year follow-up period. This was because we did not wish to project into the future, where disease rates might be affected by a variety of exogenous factors, such as improvements in disease treatment. It is clear from the results in Fig. 3 that the annual numbers of lives saved increase rapidly over time, particularly from LC and COPD, where quitting takes a long time to reduce risk.
The third reason is that our analyses did not take into account the possibility that cigarette smokers who take up ECigs or HnBs might be more likely to quit cigarette smoking than those who continued to exclusively smoke cigarettes. Evidence from the US shows that use of ECigs is associated with increased cessation rates [42]. A possible limitation of our modelling is that we considered people who simultaneously used two or three out of cigarettes, ECigs, and HnBs as multiple-product users, with their effective dose taken as the mean of 1, 0.05, and 0.20. People who are dual users of cigarettes with either ECigs or HnBs might have a higher effective dose than the mean, while those who are dual users of ECigs and HnBs might have a lower effective dose. However, because the proportion of multiple product users is quite low, particularly for the Conversion and Full Conversion Scenarios, the overall effect of this limitation on the results seems likely to be quite modest.
A further possible limitation of our modeling is that we did not consider oral tobacco products like snus, the reason being that snus is not on the market in the EU outside of Sweden. Undoubtedly, snus can have a large positive impact on public health when smokers switch to it [16]. With other forms of oral tobacco having become available in Germany over the past years, uptake of such products could thus increase the overall effect on DD and YLS in the case of smokers switching to products without tobacco combustion, as has already been modelled in Sweden [44].
The rate at which smokers switch to less harmful alternatives like ECigs and HnBs is likely to depend on product risk perception, a large body of evidence having already shown this to be the case for ECigs. For instance, accurately perceiving ECigs as less harmful than cigarettes predicted subsequent ECig use among British smokers [45] and continues to correlate with ECig use among UK smokers [46].
German smokers were more likely to use ECigs for smoking cessation if they perceived them as less harmful than Cigarettes [47]. US adult dual users of ECigs and Cigarettes who perceived ECigs as less harmful than cigarettes were more likely to switch to exclusive ECig use 1 year later [48]. However, correct risk perceptions of ECigs remain low and are getting worse over time, both internationally [45,46] and in Germany, where more than half of the population perceives ECigs [49,50] and HnBs [50] as at least as harmful as cigarettes. Even among ever-users of HnBs in Germany, only just over half of them accurately perceived HnBs as less harmful than cigarettes [51]. Public health experts in the UK, the US, and Germany are, therefore, calling for better access to fact-based information [9,45,52,53].
Educational campaigns via trusted public health institutions are likely the most effective tool [54]. While such campaigns exist in the UK, they are virtually absent in Germany.
Intuitively, maximizing the bene cial population health impact of introducing ECigs and HnBs will require a combination of high uptake among smokers, with many ultimately becoming exclusive RRP users. Our modeling results support this notion, with the DD and YLS increasing between Scenarios 4 and 5, when uptake was increased, and between Scenarios 5 and 7, when exclusive product use was increased. As discussed above, RR perceptions for ECig/HnB vs. smoking are potential drivers for both product uptake and exclusive product use, with health policy actions like public education campaigns being a recommended tool. Other factors likely to have an impact include risk-proportionate regulation in general [55]-such as product health warnings [56]-and local smoking cessation guidelines and healthcare professional recommendations [57] as well as media headlines [58]. Moreover, scal policies can have an impact on relative product use. Recent US retail panel data suggest that ECig taxation increased cigarette sales [59].
Many other published papers have attempted to quantify the population health impact of introducing RRPs. These include estimates based on the methodology we have used, but applied to the USA [24, 25] or Japan [26], as well as attempts using different methodology, supported by other tobacco companies [60][61][62][63][64][65] or by public funding [66][67][68][69][70][71][72]. Despite methodological differences, most modelers have assumed that the risk from RRP use, relative to that from cigarette smoking, is low and have concluded as we have that introduction of RRPs is likely to have a bene cial impact. For example, Levy et al. [69] concluded that "The tobacco control community has been divided regarding the role of e-cigarettes in tobacco control. Our projections show that a strategy of replacing cigarette smoking with vaping would yield substantial life year gains, even under pessimistic assumptions regarding cessation, initiation and relative harm." Although this previous paper focused on e-cigarettes, the authors did note that "…heat-not-burn tobacco products have been introduced in some countries, and these may be a better substitute for cigarettes than e-cigarettes…" As noted in the introduction, the number of smoking-attributable deaths estimated by Mons and Brenner to have occurred in Germany in 2013 is 125,000 [6]. In the Null Scenario, in 2013, the number of SRDs was estimated to be 39,629. There are three main reasons for this discrepancy. First, we only considered four diseases, which form only about 67.5% of the total number of smoking-related diseases [2].
Second, we only considered the deaths of people aged 30-79 years, whereas the published estimate was related to age 35 years or above. Third, the disease-speci c RRs used by Mons [2] were derived from speci c US studies, whereas ours were derived from detailed meta-analyses (see Table 5). While the RRs from the two studies were quite similar for both IHD and stroke, those for LC (23.26 for men and 12.69 for women vs. 11.68 for both sexes) and COPD (10.58 for men and 13.08 for women vs. 4.56 for both sexes) were markedly higher in the previous study. Had we considered more diseases, a wider age range, or higher RRs, the estimated DD and YLS would, of course, have increased.
Overall, our results provide insight into how much the introduction of the two RRPs considered might affect the distribution of usage in Germany and the mortality related to cigarette smoking. Policies and regulation can accelerate switching to these RRPs, including calling for a more risk-proportionate approach and for the best available information on RRPs to be available to adult smokers. This will help increase the perception of the harm-reduction capabilities of RRPs and encourage switching, make alternatives to cigarettes more attractive for smokers, and help maintain product standards for building consumer trust in RRPs. Rather than any single measure, an integrated tobacco control strategy is likely to be more successful in encouraging smokers to switch to RRPs and thus result in an overall public health gain.

Conclusions
On the basis of plausible estimates of the rate of uptake of two RRPs (HnBs and ECigs) in Germany and their effective dose compared with cigarettes, it is estimated that there would be a drop in SRDs from LC, COPD, IHD, and stroke of approximately 40,000 to 81,000, with 0.50 to 1.05 million YLS, corresponding to 17-38% of the effect of immediate cessation (Scenario 1). While cessation is the best option for smokers, we estimate that introducing RRPs and encouraging smokers who would otherwise continue to smoke cigarettes to switch to them will result in a substantial population health bene t, even under conservative assumptions about their relative harm and rate of uptake.

Declarations
Ethics approval and consent to participate Not applicable.

Consent for publication
Not applicable.

Availability of data and materials
All data generated or analyzed during this study are included in this published article and its supplementary information les.
Competing interests RR and AKN are employees of Philip Morris International. SD was an employee of Philip Morris International at the time the work was carried out. AK, EB, PNL and JSF were contracted consultants for the project and paid by Philip Morris International.

Funding
The work described was wholly supported by Philip Morris Products S.A.

Authors' contributions
The work described here was conceived by RR, SD, AKN and PNL. Some of the data required was extracted by AK and EB. The analyses were run by JSF, RR and SD, and checked by PNL. The manuscript was drafted by PNL, with contributions by AKN, and developed following comments by RR, SD, EB and JSF. All authors read and approved the nal manuscript.