Estimated public health gains from smokers in Germany switching to risk-reduced alternatives: Results from population health impact modelling by socioeconomic group

Previously, we estimated the impact of introducing heat-not-burn products and e-cigarettes in Germany between 1995 and 2015 on mortality from the four main smoking-related diseases in men and women aged 30–79 years. Deaths would have reduced by 216,650 (from 852,357 for continued smoking) had everyone quit smoking in 1995 and by 39,818 − 179,470 had one or both of heat-not burn products and e-cigarettes been introduced and adopted to varying extents. Here, we report substantial mortality reductions separately in two socioeconomic groups (A = higher, B = lower) dened by income and education. age 10–79 years) in SEG B exceeded that in A in both sexes, this excess was only evident from age 50 years. Because the percentage of the population in group B increased steadily with age, the members in B overall (for age 10–79 years) were older than those in A (men by 3.85 years and women by 8.44 years). Among men, the prevalence of current smoking was greater in B than A, much more so at age 20–49 years than at older ages, with the overall difference between the two groups for age 10–79 years being by 6.69 percentage points (42.17% vs. 35.48%). Among women, the prevalence of current smoking was also much greater in B than A at age 20–49 years; however, the difference was reversed at older ages, so that, for age 10– 79 years, the prevalence was very similar by SEG (25.95% in A and 26.67% in B). The overall prevalence of former smoking varied little by SEG in men, being 23.21% in A and 23.96% in B; but, in women, the prevalence was higher in A than in B (19.58% vs. 9.78%). The mean age was, as expected, higher in former smokers. The mean age was similar in A and B for current smokers but greater in B for former smokers.

By comparing Scenarios where RRPs are or are not introduced, this approach generates estimates of the DD and the YLS associated with RRP introduction 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. In comparison with our companion paper, our present estimates are derived separately for two SEGs.

Outline of the approach used
The method used for estimating the population health impact of introducing an RRP in Germany, which involves a Prevalence (P-) component and an Epidemiologic (E-) component, is essentially unchanged from that used in our companion paper. However, instead of individuals of a given sex in the Pcomponent starting with a nationally representative distribution of age group and cigarette smoking, they start with a representative distribution of age group, SEG, and cigarette smoking. Additionally, they are followed up by using sets of nicotine usage transition probabilities (TP) that vary by SEG as well as by sex, age, and length of follow-up. As before, individuals are followed up under the Null Scenario, where RRPs are never introduced, and various Alternative Scenarios, where one or more RRPs may be introduced. In building up each individual's tobacco product use history over the follow-up, allowance is also made for individuals to change between the higher (A) and lower (B) SEGs.
Given the tobacco histories by SEG and the number of deaths by disease and SEG, the methodology used in the E-component to estimate the DD and the YLS associated with RRP introduction is unchanged from that used in our companion paper. The estimation is based on the negative exponential model (NEM) and, as before, requires estimates of the relative risk (RR) and quitting half-life for each disease, and of the effective doses for current exclusive HnB use, current exclusive ECig use, and multiple product use, compared with that for current cigarette smoking (taken as one unit). Note that the estimates of RR, quitting half-life for each disease, and effective dose are taken to be independent of the SEG.

Common features of each simulation
As in the companion paper, each simulation involved the follow-up of 100,000 individuals, initially aged 10-79 years, in 1-year intervals from 1995, with the product use status of each member of the simulated population being estimated annually until the year 2015 (or the members reach an age of 79 years, after which they are no longer followed up). For each scenario described below, separate simulations were conducted for each sex.

Population at baseline
At baseline, each individual is randomly allocated to a year of age, then to an SEG, 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 1995 are as in the companion paper.
The de nition of SEG was based on a combination of net annual income and mean years of education, as described by Foreman et al. [20].
The sex-and age-speci c distributions of the population by SEG for 1995 were taken from estimates for the year 2002, derived from the German Socioeconomic Panel [21]. Section 1 of Additional File 1 gives fuller details of how the higher and lower socioeconomic groups A and B were de ned.
The sex-, age-and SEG-speci c distributions of current and former smoking prevalence for individual years from 1995 to 2015 were derived from the same three sources as those in the companion paper [21][22][23], with only the estimates for 1995 being required for the baseline population. More details are given in Sect. 2 of Additional File 1.
For the baseline population in 1995, the sex-, age-and SEG-speci c distributions of quit time for former smokers were taken from the 2002 estimates derived from the German Socioeconomic Panel [21]. As in the companion paper, the data for age groups 10-14 and 15-19 years were taken from United States (US) estimates [24]. Details are given in Sect. 3 of Additional File 1. Table 1 presents the sex-and age-speci c data on population size and percentage in each SEG as well as the sex-, age-and SEG-speci c data on the prevalence of current and former smoking. Table 2 presents the distribution of quit time used. These data were used to assign the initial status of each member of the simulated population. As shown in Table 1, the percentage of the population in group A declined steadily with age from 40-44 years, and, while the overall numbers (for age 10-79 years) in SEG B exceeded that in A in both sexes, this excess was only evident from age 50 years. Because the percentage of the population in group B increased steadily with age, the members in B overall (for age 10-79 years) were older than those in A (men by 3.85 years and women by 8.44 years). Among men, the prevalence of current smoking was greater in B than A, much more so at age 20-49 years than at older ages, with the overall difference between the two groups for age 10-79 years being by 6.69 percentage points (42.17% vs. 35.48%). Among women, the prevalence of current smoking was also much greater in B than A at age 20-49 years; however, the difference was reversed at older ages, so that, for age 10-79 years, the prevalence was very similar by SEG (25.95% in A and 26.67% in B). The overall prevalence of former smoking varied little by SEG in men, being 23.21% in A and 23.96% in B; but, in women, the prevalence was higher in A than in B (19.58% vs. 9.78%). The mean age was, as expected, higher in former smokers. The mean age was similar in A and B for current smokers but greater in B for former smokers. SEG = socioeconomic group de ned based on education and income, with group A having the higher, and B the lower socioeconomic status. In the absence of available data, those aged 10-14 and 15-19 years were assumed to be equally divided between the two groups. Estimation of histories of smoking for the Null Scenario The sex-, age-and SEG-speci c TPs used in the P-component for developing the histories of smoking for the Null Scenario were derived as described in Additional File 2 and are shown in Table 3. To test the validity of the TPs, prevalences predicted by using these TPs were compared with the estimates for Germany derived as described above for years up to 2015. The rst period relates to the 5 years starting in 1995, the second period to the 5 years starting in 2000, and the third period to the 10 years starting in 2005.
The monthly probabilities of transition (per million) among the three states N = never, C = current, and F = former, are described by P (probabilities) 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. The rst period relates to the 5 years starting in 1995, the second period to the 5 years starting in 2000, and the third period to the 10 years starting in 2005.
The monthly probabilities of transition (per million) among the three states N = never, C = current, and F = former, are described by P (probabilities) 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. The rst period relates to the 5 years starting in 1995, the second period to the 5 years starting in 2000, and the third period to the 10 years starting in 2005.
The monthly probabilities of transition (per million) among the three states N = never, C = current, and F = former, are described by P (probabilities) 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. SEG = socioeconomic group de ned based on education and income, with group A having the higher, and B the lower socioeconomic status.
Estimation of histories of product use for the Alternative Scenarios In the companion paper, we had considered seven different Alternative Scenarios, numbered 1 to 7. Here, we only considered four Alternative Scenarios (1, 3, 6, and 7), with the original numbering being retained in order to facilitate comparison between this paper and our companion paper. The four Alternative Scenarios are described brie y below and in more detail in our companion paper.
1. Complete cessation: All current cigarette smokers in 1995 immediately stop smoking, with no further product use.
3. Complete switch to RRPs (50% HnB and 50% ECig): All current cigarette smokers in 1995 immediately switch, half to HnBs and half to ECigs, with subsequent initiation, re-initiation, and quitting only involving the new products.
In the two Alternative Scenarios that were termed 6 and 7 in our companion paper, the market shares of HnBs and ECigs in 2005 were assumed to be, respectively, 15.5% and 36.4% of the market share of cigarettes in 1995. Scenarios 6 and 7 vary only in the proportion of exclusive users of the two RRPs, i.e.
RRP users who have entirely given up cigarette smoking.
6. Conversion Scenario: The assumed proportions of exclusive users rise to 84% for both HnBs and ECigs.
7. Full Conversion Scenario: The assumed proportions of exclusive users rise to 100% for both HnBs and ECigs.
As previously described, various constraints were applied in Alternative Scenarios 6 and 7 to ensure comparability with the TPs for the Null Scenario, and the only difference between the two Scenarios is in the assumed rate of switching between the products.
RRPs are not introduced in Alternative Scenario 1. For the other three situations, the effective doses assumed, as measures for product harmfulness, were, as in the companion paper, 0.2 for exclusive HnB, 0.05 for exclusive ECig use, and 0.417 for multiple product use, as compared to an effective dose of 1 for exclusive cigarette use.
The full set of Alternative Scenario TPs is presented in Additional File 3.

Factors affecting TPs
As in the previous paper, the option that would allow TPs to depend on previous product history was not used. Table 4 presents the annual TPs used to allow transition among SEGs. The method of derivation is described in Additional File 4. Estimating RRs on the basis of product use histories

Transitions between SEGs
The estimates of RR and quitting half-life for each disease are as given in our companion paper. The United Nations data on population size and the WHO data on numbers of deaths from 1995 to 2015 are presented in our companion paper. On the basis of these data, the method of estimating the numbers of deaths and increase in death rates associated with smoking is essentially as described earlier [24].
The sex-and age-speci c data on national population size and numbers of deaths in Germany from LC, COPD, IHD, and stroke for the combined SEGs are as given in the companion paper, and the sources are described there. SEG-speci c estimates were obtained by multiplying these values by the proportions in the SEGs. Additional File 5 presents the mortality data by SEG.

An additional analysis
Differences in the estimated DD and YLS between SEG A and SEG B may arise both because of differences among SEGs in the prevalence of smoking habits, and in age distribution. To gain insight into which of these factors was most important, an additional analysis was run, the same as for Scenario 6, except that the prevalence of smoking habits assumed was that for SEGs A and B combined, rather than varying by SEG.

Results
The full results of the analyses are available in Additional File 6. Figure 1 compares the never, current, and former smoking prevalence estimates for Germany by sex and SEG for age groups 40-44 and 60-64 years as simulated in the Null Scenario (broken lines) with the estimates derived as described in the Methods section (solid lines). The t for these (and other) age groups is generally quite good, con rming the validity of our approach. The distributions for other age groups are shown in Additional File 6. in the percentage of current product use in 2010. As expected (from the assumptions described earlier), there were no current product users at all in Scenario 1, and the total proportion of current product users in Scenarios 3, 6, and 7 were essentially the same, with the variation among these Scenarios being only in the distribution of the four current product use groups. Table 5 presents the estimated DDs at ages 30-79 years over the whole follow-up period in all four Scenarios. The estimates were generated by using the Pcomponent of PHIM and are shown by disease, for the four diseases combined, and by SEG (A, B, and A + B). The results are expressed both as numbers and percentages of all the smoking-related deaths from the four causes studied. The DDs in the Conversion Scenario are also shown by sex, disease, and SEG over the whole follow-up period in Fig. 3.
The pattern of results in each SEG is similar to that described in our companion paper, with the largest DDs in both sexes being in Scenarios 1 and 3 and smaller DDs seen in Scenarios 6 and 7, where the switch to RRPs is gradual. In both sexes, larger DDs are seen for IHD (particularly in men) and LC than for COPD or stroke, and the percentages of DDs in smoking-related deaths are the highest for IHD and stroke, which have shorter half-lives than LC and COPD.
The DDs are, in all Scenarios, substantially larger in SEG B than A, with SEG B being older, having more individuals, and a higher prevalence of product use. Thus, the combined DDs for both sexes in Scenarios 1, 3, 6, and 7 are, respectively, The total DD in SEGs A and B combined can be regarded as an SEG-adjusted total, which can be compared with the unadjusted estimates in Table 4    COPD = chronic obstructive pulmonary disease, ECig = e-cigarette, HnB = heat-not-burn, IHD = ischaemic heart disease, LC = lung cancer, Sc = Scenario code, SEG = socioeconomic group de ned based on education and income, with group A having the higher, and B the lower socioeconomic status.
As in the companion paper, accounting for the increase in population size associated with the reduced mortality in the Alternative Scenarios relative to that in the Null Scenario made little difference to the estimated DDs. This is shown in the detailed results in Additional File 6.
In Scenario 6, the estimated DD for both sexes and all four diseases combined was 24,010 in SEG A and 50,385 in SEG B, an excess of 26,375 in B (see Table 5). In the additional analysis, which assumed that there was no variation in the prevalence of smoking by SEG, the estimates were 24,998 in A and 49,022 in B, an excess of 24,024 in B that was 91.1% of that for the original analysis. Corresponding percentages were 93.6% for men and 85.4% for women.
Based on YLS, totals over sex and disease (in thousands) in our original analysis (see Table 6) were 369 in A and 587 in B, an excess of 218. In the additional analysis, these estimates were 395 in SEG A and 587 in SEG B, a difference of 163, which is 74.8% of that seen originally. For both DD and YLS, these results demonstrate that most of the excess mortality seen in SEG B was due to the age differences among the SEG groups, though differences in smoking habits also contributed. To be clear, smoking-related diseases, as we consider here, originate from smoking, not from age differences, which is why public health measures have duly focused on lowering smoking prevalence to curb the health consequences of consuming combustible cigarettes.

Discussion
To our knowledge, this is the rst study to estimate the population health impact of introducing RRPs in Germany by SEG. Depending on the Scenario, the health gains would have reached 34-83% of that from immediate smoking cessation, regardless of the SEG. For each of Scenarios 3, 6 and 7, the gains in the lower SEG population (B) compared with those in the higher SEG population (A) would have been more than twice as high for DD and more than 1.5 times as high for YLS. Depending on the Scenario, these excess gains in B than in A corresponded to a greater DD by 27,004 to 62,262 and a greater YLS by 215,000 to 526,000.
In our companion paper, which took no account of the SEG, we estimated that, in Scenarios 1 (Complete cessation), 3 (Complete switch to RRPs -50% HnBs, 50% ECigs), 6 (Conversion), and 7 (Full Conversion), the DDs for the sexes combined and all four diseases combined would have been, respectively, 216,650; 179,470; 75,597; and 81,293. According to the results in Tables 5 and 6 in this paper, which did take into account the SEG, the corresponding estimates were quite similar (220,409; 182,424; 74,395; and 78,300, respectively) to those in our companion paper. Our results suggest that taking SEG into account had little effect on the overall estimated DDs. They also show that substantial DDs were seen in both SEG groups, A and B, with the drops being about twice as high in B as in A.
The estimates of population health impact in both SEGs are likely to be conservative for the three main reasons discussed in our companion paper -that deaths were only counted for four diseases, that only a 20-year follow-up period was considered, and that we failed to account for the possibility that cigarette smokers who took up HnBs and ECigs might subsequently be more likely to quit cigarettes than those who continued as exclusive smokers. We also note that the RRs we have used for current and former smoking (on the basis of published meta-analyses) are lower than those used by others (on the basis of speci c studies) when estimating deaths attributable to smoking in Germany [25]. Using higher RR estimates would have increased our DD estimates. However, our estimates might be optimistic if the rates of uptake of HnBs and ECigs are lower than what we have assumed, or if, in fact, taking up HnBs and ECigs makes smokers less likely to quit or increases their cigarette consumption. There is little real-life evidence, however, that the latter is the case.
While there is mounting scienti c consensus that RRPs like ECigs and HnBs represent less risk than cigarettes [26,27], there has been debate on whether smoking-related health inequalities between SEGs could be reduced by smokers switching to RRPs [28]. Some studies have found more advantaged smokers to be more likely to use ECigs in the UK [29] or HnBs in Germany [30], whereas others have recently observed that ECigs might be helping disadvantaged smokers quit [19]. ECigs might have contributed to some of the highest UK smoking cessation rates so far, with parity across SEGs [31]. This suggests that ECigs worked as quitting aids for low SEG smokers previously not reached by conventional methods and can be explained by the high acceptability of RRPs as substitutes for cigarettes, as witnessed from ECigs being the most popular quitting aids in the UK [18] and Germany [17,32]. It is notable that conventional pharmacotherapies for smoking cessation are more commonly used in Germany by smokers with higher incomes [17,32], thus possibly increasing health inequalities. The question whether this is linked to the affordability of pharmacotherapies or to factors related to education has not been conclusively answered. While a recent study found higher income of German smokers to be associated with more frequent use of pharmacotherapies, neither income nor education affected quit success [32]. On the other hand, use of the most popular quitting aid ECigs was not associated with income or education, underlining that RRPs could be a promising addition to public health strategies aimed at providing equal chances for smokers from different SEGs to exit out of cigarettes [32].
Data from several individual countries have shown that lower social grades and level of education are signi cantly linked to inaccurate harm perception of ECigs [26,33]. Currently, 61% of German smokers falsely perceive ECigs as equally harmful or more harmful than cigarettes, with only 5% correctly perceiving them as much less harmful [34]. Given that ECigs are more likely to be used for smoking cessation if they are perceived as less harmful than cigarettes [35], misperceptions might well be discouraging many cigarette smokers from trying RRPs. Improving current perception is particularly important for disadvantaged populations of smokers, such as those in lower SEGs, who could bene t more from RRPs as a harm reduction tool.
Uptake of RRPs by smokers is also affected by other factors such as taxation [36], affordability [37], moral concerns around addiction [38], and consumer choice regarding ECig avors [39]. To reduce smoking-related health inequalities, an integrated strategy has been proposed which combines targeted cessation programs, tobacco control measures, and educational media campaigns, all applied within wider attempts to address inequalities in health [9]. Clear communication of relative risks, delivered through targeted public health educational campaigns, could help realize the potential of RRPs for harm reduction among lower SEGs.
The strengths of our study include the use of nationally representative data and the hindcasting approach, which helps avoid problems in accounting for the unknown future effects of other factors, such as medical advances, infections, wars, and global warming, on future death rates. Our methodology also allows the population health impact of RRP introduction to be estimated under a variety of different assumptions.
There are, however, a number of potential limitations that need to be considered in interpreting our estimates. Issues regarding our failure to consider other sources of nicotine or environmental tobacco smoke, the possible limitations in our negative exponential model, the choice of effective doses, limiting our attention to deaths at ages 30-79 years, and the choice of appropriate uptake rates of HnBs and ECigs have been discussed in our companion paper and are not considered further here. Also failure to take into account the reduced mortality in the Alternative Scenarios compared to that in the Null Scenario was shown to have very little effect on the results.
Some issues speci cally relating to estimation of population health impact by SEG are, however, worth consideration. One is that the age-and sex-speci c prevalence data for current and former smoking in Germany by SEG was only available for 2 years (2002 and 2012); therefore, annual data had to be estimated by a combination of interpolation/extrapolation and smoothing (as described in Additional File 1). More detailed source data might have led to some revision of our estimates, but it seems unlikely that they would have made much difference, given that the adjustment for SEG had little effect on the unadjusted estimates.
Similar considerations apply to the derivation of TPs by SEG in the Null Scenario, which, as described in Additional File 2, are calculated based on the distribution of smoking habits in the same birth cohort 5 years apart. Errors in the distributions would have led to errors in the estimated TPs, and inspection of the TPs in Table 3 shows that though the general patterns by SEG and age look plausible, there are a few exceptions. For example for SEG A period 1-5 years, the estimated initiation rates in men rise between ages 25-29 and 30-34 years, while at age 15-19, but not at other ages, quitting and re-initiation rates were much higher in women than men. Given that the changes in the distribution of smoking habits over time generated by the model using the TPs matched quite well observed distributions in Germany, it is unlikely that any further attempt to improve estimation of these TPs would have materially affected our conclusions.
Another issue is that data on transfer between SEG groups A and B were not available for Germany, and the data used here (as described in Additional File 4) were derived partly from the current smoking prevalence rates in the USA coupled with assumptions about the age-speci c level of transfer from A to B. In fact, weaknesses in these data seem unlikely to be very relevant because re-running the analyses for Alternative Scenario 6 by disallowing the possibility of transfer had little effect on the overall estimates and only slightly reduced the estimated DDs shown in Table 5  One other issue relates to how we classi ed the SEGs. As noted earlier, our classi cation used a previously described standard method [20] based on a combination of net annual income and mean years of education. Using methods that also took occupation into account or classi ed individuals into more than two groups would have been possible, but it is doubtful that it would have affected the conclusion that switching to RRPs would have reduced the risk and been more bene cial in groups that smoke more.
A further issue is that, for any given year, disease, sex, or age group, the estimated number of deaths by SEG was derived from the reported number for the combined SEG by multiplying this number by the estimated proportions in A and B for that year, sex, and age group. This assumes that the mean risk of death is identical in SEG groups A and B. While data are available from the NEM on the average RR (relative to never smokers) in each SEG group, no data are available on the absolute risk separately in A and B and, therefore, a precise test of this assumption is not possible. However, some idea of the potential effect of allowing for variation in risk by SEG can be gained from an example based on the data for men and all diseases combined. Overall, our results clearly demonstrate that increasing uptake of HnBs and ECigs would reduce the adverse population health impact of cigarette smoking in both SEG groups A and B. They also show that adjustment for SEG makes little difference to the estimated impact in the total population.

Conclusions
Our population health impact model showed that introducing RRPs (ECigs and HnBs) into the tobacco market in Germany in 1995 would have resulted in substantial reductions in mortality from the four main smoking-related diseases over the following 20 years in both SEGs studied here. Depending on the Scenario chosen, these gains would have amounted to 31-82% of those achieved by immediate smoking cessation, the optimal Scenario. Considering that only 19% of German smokers even attempt cessation each year and few succeed [17], tobacco harm reduction by fully switching to RRPs represents a public health opportunity currently not exploited in Germany.
While our model predicts similar percentage reductions in mortality in both SEGs, the total predicted reductions were greater in the lower SEG, B, by about 2fold for DDs and about 1.5-fold for YLS, because of individuals in B being somewhat more numerous, of a higher average age (as the proportion of the population in B increases markedly with age), and more commonly current smokers (evident especially among men and at younger ages). These greater reductions in mortality in B correspond to an extra 27,000 to 62,000 DDs and 215,000 to 526,000 YLS, depending on whether Scenarios 3, 6 or 7 are considered.
Our results thus suggest that inclusive access to RRPs, i.e. access to acceptable, affordable products as well as comprehensible relative risk information about them, could diminish smoking-related health inequalities between SEGs. A public health approach that encourages low SEG smokers who would otherwise continue to smoke to switch to RRPs could synergize with tobacco control measures targeted at lower SEGs to further reduce such inequalities.

Declarations
Ethics approval and consent to participate Not applicable.