Modelling Potential Health Gains and Health System Savings Associated with Vaporised Nicotine Products in Canada

Objectives: To model population-wide health and cost impacts of vaporised nicotine products (VNPs) use among Canadian adults 20 years and older from 2015-2095. Methods: A multi-state lifetable model was used to project potential changes in life expectancy and health-system costs, overall and by province/territory. The simulated population was divided into 68 cohorts by sex, ethnicity, and 5year age groups. Each year, individuals could either remain in their current state, or transition to one of six smoking/vaping states. Input parameters were extracted or estimated using data from Statistics Canada and literature. Three scenarios were modelled to reflect a range of uncertainty: Status Quo (“SQ”, VNPs commercialised as they are currently in Canada); No-Vaping (“NV”, assuming VNPs never entered the Canadian market); and a Pro-Switching Policy (“PSP”, assuming increased VNP prevalence). Results: Compared to NV, SQ projected to increase life-years by 922,547, while PSP increased them further (+718,137). SQ projected a C$39.0 billion reduction in cumulative lifetime costs compared to NV; PSP would further reduce them by C$30.4 billion. Statistical variability was assessed using sensitivity analyses on input parameters, and Monte-Carlo simulations. Conclusions: Accessibility to VNPs in Canada was projected to generate net public-health gains and health-system cost savings. These projected health and economic consequences are sensitive to assumptions about accessibility and use by adult smokers and may vary by type of policy environment. Introduction Vaporised nicotine products (“VNPs”) are alternatives to smoking that may assist adult smokers in switching away from combustible cigarette smoking. While the long-term effects of VNPs cannot be known precisely, health risks associated to Qeios, CC-BY 4.0 · Article, March 22, 2021 Qeios ID: OJM4HF · https://doi.org/10.32388/OJM4HF 1/22 VNPs are believed to be substantially less than traditional tobacco smoking (Ratajczak et al. 2018; Stephens 2018). Public health authorities such as the National Academies of Sciences, Engineering, and Medicine (NASEM), Public Health England (PHE) note that vaping is significantly less harmful than smoking combustible cigarettes (McNeill et al. 2018; National Academies of Sciences, Engineering, and Medicine 2018). The difference in incremental risk of disease incidence between tobacco smoking and VNPs may contribute to improve health and cost outcomes, should use of the latter by current smokers increase. Several population health models have evaluated the impact of introducing VNPs to market (Apelberg et al. 2018; Cherng et al. 2016; Levy et al. 2018; Levy et al. 2017; Warner and Mendez 2019). Among them is the Burden of Disease Epidemiology, Equity, and Cost-Effectiveness (BODE) model, a multi-state lifetable model that was developed by researchers at the University of Otago, Wellington (New Zealand) to evaluate the effectiveness of interventions aimed at reducing specific diseases and costs associated with health problems such as obesity and smoking (Blakely et al. 2016; Burden of Disease Epidemiology n.d.). A team of researchers, Petrović-van der Deen et al. (2019), hereafter designated as “NZ”, adapted the BODE model to assess the health and cost impacts of legalizing the domestic sale of VNPs. The AppEco model presented here builds upon the core elements of the NZ cohort-based model and adapts it to the Canadian context on a national, provincial, and territorial level. The AppEco model simulates whether the introduction of VNPs in the adult population may be associated with changes in health and cost impacts over the long-term. The objective of this study was to estimate the incremental gains in life-years and associated health system costs for various vaping versus smoking scenarios to determine the ways in which different policies can yield different health and healthcare cost outcomes. As Canada considers its current vaping regulations, it may benefit from a population health model that can inform future policy decisions. Methods The AppEco model, illustrated in Figure 1, was used to simulate the changes in incremental life-years lost due to SR diseases and health system costs for adults in Canada associated with varied levels of VNP use, at a 0% discount rate. Beginning in 2015, the AppEco model tracks, cohort by cohort, the health and cost impacts of VNPs on the selected population as they age. The model was developed using Microsoft Excel® to facilitate future research. Data Sources The main lifetable was built using data from Statistics Canada for demographics and all-cause mortality. The 14 SR diseases integrated into the model are those recognized by the Canadian government namely, cerebrovascular disease, chronic obstructive pulmonary disease (COPD), heart disease, as well as the following cancers: bladder, cervical, esophageal, kidney, laryngeal, lung, oral, pancreatic, pharyngeal, stomach, and leukemia. For each SR disease, the model integrated epidemiologic parameters for disease-specific incidence, prevalence and case-fatality rates, remission rates, direct and indirect health costs using a combination of data from Statistics Canada, the Public Health Agency of Canada (PHAC), and the Canadian Institute for Health Information (CIHI). Additional input data for smoking/vaping Qeios, CC-BY 4.0 · Article, March 22, 2021 Qeios ID: OJM4HF · https://doi.org/10.32388/OJM4HF 2/22 prevalence, transition matrices, and relative risks of SR diseases were estimated using data from the Canadian Tobacco, Alcohol and Drugs Survey (CTADS), as well as existing epidemiological literature. Data for each subset of the Canadian population aged 20 years or older in 2015 was collected, as it was the first year of reliable data collection on vaping in Canada. Where possible, provincial, and territorial-specific data was also integrated (see Table 1 for a full list of data sources). Study Population The study population was selected based on the number of Canadian adults aged between 20 and 99 years old in 2015 (n = 27,816,114). All rate and cost inputs in both the main lifetable and disease-specific lifetables for this population were divided into 68 cohorts separated by age, sex, and ethnicity (Indigenous and non-Indigenous). Each cohort is separated into five-year age groups from 20-24 years old to 94-99 years old. Smoking/Vaping Prevalence For each scenario, the prevalence of smoking and vaping was modelled from 2015-2095 by attributing the six smoking and vaping states listed in Table 2 to the study population. These states divide the population into Current Smokers Not Current Vapers (“CSNV”), Former Smokers Not Current Vapers (“FSNV”), Dual Users (“DU”), Former Smokers Current Vapers (“FSCV”), Never Smokers Current Vapers (“NSCV”), and Never Smokers Not Current Vapers (“NSNV”). Smoking/Vaping State Transitions The AppEco model assumed that at every annual cycle, once a year, individuals can either transition to another smoking/vaping state or remain within the same state, depending on the scenario analysed. CSNVs can only transition to a FSNV, or DU state, as they cannot logically return to either of the two never smoker states. The same logic applies for DUs, who also cannot go back to a never smoker state. Unlike our never smoker states, CTADS definitions for vaping do not include a “Never Vaper State”, but instead a “Not Current Vaper” state, thereby making the transition from current vaper to not current vaper states possible. As such, DUs can transition to CSNV and FSCV states, and NSCVs can move to a NSNV state. Based on the seminal work of Holford et al. (2014) for cross-sectional surveys of smoking prevalence across different age cohorts, which shows net decreases in prevalence after 20 years old, we assumed no uptake in smoking and conservative rates of cessation and transition to vaping. Three transition-probability matrices were set for the following age cohorts: 20-44 years old, 45-64 years old, and 65+ years old; these were based on the findings of Holford et al. (2014) for CSNV smoking cessation rates, Zhu et al. (2017) for DU cessation rates, and Manzoli et al. (2017) for rates associated with CSNV to FSCV and DU to FSCV. We also assumed that the probability of DUs moving to a CSNV state (i.e., for DUs who quit vaping and go back to smoking alone) was equal to the probability of DU smoking cessation (Coleman et al. 2013-2015). All remaining transition probabilities were estimated using the residual sum of the initial probabilities. Transition probabilities and resulting prevalence rates were estimated using Canadian smoking and vaping prevalence Qeios, CC-BY 4.0 · Article, March 22, 2021 Qeios ID: OJM4HF · https://doi.org/10.32388/OJM4HF 3/22 data for each state, as well as current baseline trends in smoking cessation rates. As transition rates remain constant throughout the simulation, smoking/vaping prevalence per state for each year (t) was obtained by multiplying the prevalence from year t-1 by the corresponding state values of the transition matrices. See Figure 2 for a detailed account of the methods used for state transitions. Relative Risk The relative risk (“RR”) of disease incidence for vaping and smoking was estimated by integrating transition rates from never, former, and current smokers/vapers, both to and from the regular use of VNPs. The RRs of SR diseases for current versus never smokers, and for former versus never smokers were expanded to include the RRs of vaping for those using VNPs. Drawing from the methods of the NZ study, the AppEco model used the results estimated by PHE and others to calculate the RR for the remaining DU, FSCV and NSCV smoking states. According to Public Health England (PHE), vaping is 95% less harmful than smoking and dual use is 5% less harmful than smoking (McNeill et al. 2018). With respect to the NSNV state, which has no incremental risk of disease since there is an absence of both smoking and vaping, the RR for NSCV and CSCV was therefore attributed an incremental risk of 5% and 95% respectively. This does not imply that VNPs are safe to use. It is always preferable to refrain from any form of smoking and/or vaping. Considering RRS = relative risk of disease for current vs non-smokers, and RRF = relative risk of disease for former vs nonsmokers, the relative harm associated with each state was estimated using the following calculations: CSNV = RRS DU = 1 + 95%*(RRS 1); 95 % of incremental risk of CS vs NS FSCV = RRF + 5%*(RRS 1); FS vs NS risk plus 5 % of incremental risk of CS vs NS NSCV = 1 + 5%*(RRS 1); 5 % of incremental risk of CS vs NS FSNV = RRF NSNV =1; as there is no incremental risk of disease The decline of relative risks over time (according to age and time since cessation) for former smokers were modelled using the equation developed and parameters estimated by Hoogenveen et al. (2008): = 1 + (RR − 1) ∗ exp[ − Y0 ∗ exp( − ƞ ∗ age) ∗ timesincecessation] This allowed for disease incidence to increase as the population ages, and decrease from the time individuals quit smoking, vaping, or both. Model Sequence Relative risks for each SR disease “RRi”, with “i” representing one of the 14 SR diseases, were used to calculate the Qeios, CC-BY 4.0 · Article, March 22, 2021 Qeios ID: OJM4HF · https://doi.org/10.32388/OJM4HF 4/22 population impact fraction (PIF). Calculating PIFi for each disease produces changes of incidence rates over time, which were then applied to the population lifetables: PIFi = ∑pi XRRi– ∑pi XRRi/∑pi XRRi where is the prevalence of smoking and non-smoking in the no-VNP scenario, and is the prevalence of smoking in other intervention scenarios. Incidence of disease i after intervention (Ii) is Ii = I 0 i x(1– PIFi) The PIFi determines the intervention base-case scenario incidence rates for all SR disease lifetables, which cause disease-specific changes in morbidity and mortality rates, as well as health system costs. Resulting incremental life-years lost and health system costs were then calculated for each annual cycle and across sex, age, and ethnicity cohorts for all scenarios. The difference in outcomes between each pair of scenarios established the measurable impact of each scenario. Outcome Measures Measurable outcomes included cumulative disease and mortality costs, as well as life-years lost due to SR diseases, which corresponds to the amount of years lost in an individual’s lifecycle following death due to a SR disease between the ages of 20 and 85 years old. For example, if a man dies at 85 years old due to an SR disease, he gets attributed 1 year lost for not having lived his entire 85th year; if a woman dies at 30 years old, her loss in life-years will be valued at 56 years. Moreover, the incremental social cost associated with each year of life lost was estimated and incorporated into the model. Using a meta-analysis conducted by Bellavance et al. (2009), we estimated that the value of a statistical life was approximately C$40,000 per individual per year lived. This value was used to calculate the total incremental cost of life years lost due to SR diseases throughout the population’s first 85 years. Scenario Analyses Health and cost outcomes for the following three scenarios were simulated: SQ Scenario: The SQ scenario assumes that VNPs are commercialised in Canada under the legal and regulatory constraints existing on August 1, 2019 (Government of Canada: Vaping product regulations n.d.). The transition matrices for each cohort under the SQ scenario used baseline smoking and vaping prevalence rates in Canada from 2015 to project the prevalence rates for subsequent years. NV Scenario: The NV scenario assumed the continuity of pre-2015 smoking trends without the introduction of VNP products to market. This scenario was simulated using the main lifetable until death (or 100 years old) as well as allcause mortality and morbidity rates. In this scenario, because VNP does not exist, the transition matrix used to project smoking and vaping prevalence over time only includes transitions for CSNV, FSNV and NSNV states. These Qeios, CC-BY 4.0 · Article, March 22, 2021 Qeios ID: OJM4HF · https://doi.org/10.32388/OJM4HF 5/22 transitions affect smoking prevalence in the years after 2015, and as a result, influence mortality, morbidity, and related costs over time. PSP Scenario: The aim of this scenario was to model the potential effects of an increase in Canadian VNP prevalence that could result from changes in the regulatory or policy environment. In the PSP scenario, Canadian vaping prevalence rates in 2015 were increased year after year to reach prevalence rates experienced in the United Kingdom (UK) in 2018, as a hypothetical model for this policy. The choice of UK was motivated by the fact that its regulatory environment encouraged adult smokers to switch away from smoking and towards lower-risk products, such as VNPs. Because of this policy orientation, we assumed that the overall prevalence of vaping would likely be higher in the UK than in Canada during those years. For cohorts where Canadian prevalence rates were already equal to or higher than UK prevalence rates, Canadian prevalence rates were used. For years 2019-2095, smoking and vaping prevalence rates evolve as they pass through the same transition matrix as in the SQ scenario. Sensitivity Analyses To account for the uncertainties surrounding the risk of vaping, sensitivity analyses included varying the 5% (RR of vaping vs smoking) and 95% (RR of dual use vs smoking) assumptions of relative risk, to 10% and 25% and 50% (vaping vs smoking), and 70% and 100% (dual use vs. smoking) respectively. The model ran Monte Carlo simulations (n= 1,000) for baseline risk ratios of each scenario for non-Indigenous men aged 30-34 years old. This cohort was chosen because (a) it has incremental costs of similarly high magnitude, but of opposite signs for each pair of compared scenarios (SQ vs. NV and SQ vs. PSP), as well as (b) a long modelling horizon (70 years). Consequently, the impact of parameter uncertainty should be most present for this cohort. The model applied uncertainty to disease-specific RRs, as they are the single parameter that influences disease occurrence, probability of death and associated health care costs. The variable level of SR disease incidence RR was set for all cohorts in 2015 and carried over for each subsequent year of modelling. Published risk ratios included lower and upper bounds of confidence intervals, which allowed to approximate their distribution using log–normal distributions (Barendregt 2003; Olsson 2005). Results Table 3 presents the overall results of each scenario, as well as differences between SQ and NV scenarios, and SQ vs PSP scenarios. Compared to the NV scenario, the SQ scenario is associated with a 1.8% decrease in projected life-years lost due to SR diseases. This translates into 922,547 life-years gained, and an avoided social cost of life-years lost projected at C$36.9 billion over the lifetime of all cohorts. NV is also linked with an estimated C$2.1 billion in incremental health and mortality costs, mostly due to incremental health-system costs (C$2.0 billion). By contrast, under the PSP scenario, life-years lost due to SR diseases lowered by 1.4%, increasing the absolute gain in life-years by 718,137. Moreover, a total of C$30.4 billion in social cost of life-years lost, and total health and mortality costs were projected to be Qeios, CC-BY 4.0 · Article, March 22, 2021 Qeios ID: OJM4HF · https://doi.org/10.32388/OJM4HF 6/22

data for each state, as well as current baseline trends in smoking cessation rates. As transition rates remain constant throughout the simulation, smoking/vaping prevalence per state for each year (t) was obtained by multiplying the prevalence from year t-1 by the corresponding state values of the transition matrices. See Figure 2 for a detailed account of the methods used for state transitions.

Relative Risk
The relative risk ("RR") of disease incidence for vaping and smoking was estimated by integrating transition rates from never, former, and current smokers/vapers, both to and from the regular use of VNPs. The RRs of SR diseases for current versus never smokers, and for former versus never smokers were expanded to include the RRs of vaping for those using VNPs. Drawing from the methods of the NZ study, the AppEco model used the results estimated by PHE and others to calculate the RR for the remaining DU, FSCV and NSCV smoking states.
According to Public Health England (PHE), vaping is 95% less harmful than smoking and dual use is 5% less harmful than smoking (McNeill et al. 2018). With respect to the NSNV state, which has no incremental risk of disease since there is an absence of both smoking and vaping, the RR for NSCV and CSCV was therefore attributed an incremental risk of 5% and 95% respectively. This does not imply that VNPs are safe to use. It is always preferable to refrain from any form of smoking and/or vaping.
Considering RR S = relative risk of disease for current vs non-smokers, and RR F = relative risk of disease for former vs nonsmokers, the relative harm associated with each state was estimated using the following calculations: This allowed for disease incidence to increase as the population ages, and decrease from the time individuals quit smoking, vaping, or both.

Model Sequence
Relative risks for each SR disease "RR i ", with "i" representing one of the 14 SR diseases, were used to calculate the population impact fraction (PIF). Calculating PIF i for each disease produces changes of incidence rates over time, which were then applied to the population lifetables: where is the prevalence of smoking and non-smoking in the no-VNP scenario, and is the prevalence of smoking in other intervention scenarios.
Incidence of disease i after intervention (I 1 i ) is The PIF i determines the intervention base-case scenario incidence rates for all SR disease lifetables, which cause disease-specific changes in morbidity and mortality rates, as well as health system costs. Resulting incremental life-years lost and health system costs were then calculated for each annual cycle and across sex, age, and ethnicity cohorts for all scenarios. The difference in outcomes between each pair of scenarios established the measurable impact of each scenario.

Outcome Measures
Measurable outcomes included cumulative disease and mortality costs, as well as life-years lost due to SR diseases,

Scenario Analyses
Health and cost outcomes for the following three scenarios were simulated: transitions affect smoking prevalence in the years after 2015, and as a result, influence mortality, morbidity, and related costs over time.
PSP Scenario: The aim of this scenario was to model the potential effects of an increase in Canadian VNP prevalence that could result from changes in the regulatory or policy environment. In the PSP scenario, Canadian vaping prevalence rates in 2015 were increased year after year to reach prevalence rates experienced in the United Kingdom (UK) in 2018, as a hypothetical model for this policy. The choice of UK was motivated by the fact that its regulatory environment encouraged adult smokers to switch away from smoking and towards lower-risk products, such as VNPs.
Because of this policy orientation, we assumed that the overall prevalence of vaping would likely be higher in the UK than in Canada during those years. For cohorts where Canadian prevalence rates were already equal to or higher than UK prevalence rates, Canadian prevalence rates were used. For years 2019-2095, smoking and vaping prevalence rates evolve as they pass through the same transition matrix as in the SQ scenario.

Sensitivity Analyses
To account for the uncertainties surrounding the risk of vaping, sensitivity analyses included varying the 5% (RR of vaping vs smoking) and 95% (RR of dual use vs smoking) assumptions of relative risk, to 10% and 25% and 50% (vaping vs smoking), and 70% and 100% (dual use vs. smoking) respectively.
The model ran Monte Carlo simulations (n= 1,000) for baseline risk ratios of each scenario for non-Indigenous men aged 30-34 years old. This cohort was chosen because (a) it has incremental costs of similarly high magnitude, but of opposite signs for each pair of compared scenarios (SQ vs. NV and SQ vs. PSP), as well as (b) a long modelling horizon (70 years).
Consequently, the impact of parameter uncertainty should be most present for this cohort.
The model applied uncertainty to disease-specific RRs, as they are the single parameter that influences disease occurrence, probability of death and associated health care costs. The variable level of SR disease incidence RR was set for all cohorts in 2015 and carried over for each subsequent year of modelling. Published risk ratios included lower and upper bounds of confidence intervals, which allowed to approximate their distribution using log-normal distributions (Barendregt 2003; Olsson 2005). Table 3 presents the overall results of each scenario, as well as differences between SQ and NV scenarios, and SQ vs PSP scenarios. Compared to the NV scenario, the SQ scenario is associated with a 1.8% decrease in projected life-years lost due to SR diseases. This translates into 922,547 life-years gained, and an avoided social cost of life-years lost projected at C$36.9 billion over the lifetime of all cohorts. NV is also linked with an estimated C$2.1 billion in incremental health and mortality costs, mostly due to incremental health-system costs (C$2.0 billion). By contrast, under the PSP scenario, life-years lost due to SR diseases lowered by 1.4%, increasing the absolute gain in life-years by 718,137.

Results
Moreover, a total of C$30.4 billion in social cost of life-years lost, and total health and mortality costs were projected to be   Figure 3.a, younger cohorts of men were projected to reap the most benefits in terms of life-years gained and avoided costs. The increased projected benefits between 50-59 years old is related in part to the greater prevalence of vaping (single and dual use) compared to neighbour cohorts, which would shift towards smoking in the NV scenario.
Under the PSP hypothetical, higher projected savings and life-years gained were associated with Canadian cohorts who in the SQ scenario use VNPs relatively less than their UK counterparts.
Compared to the NV scenario, non-Indigenous men were associated with a projected reduction ( Table 4.
Statistical variability was assessed using sensitivity analyses on input parameters and Monte-Carlo simulations. Both analyses were performed on a single cohort, namely non-Indigenous men aged between 30-34 years old. Figure 4 presents the results of the sensitivity analyses conducted on life-years lost due to SR diseases. At baseline, the results for Monte-Carlo simulations (

Discussion
The AppEco multi-state lifetable model was used to estimate potential changes in life-years lost and health-system costs associated to different levels of smoking and VNP use in the Canadian population alive in 2015. The SQ scenario was projected to contribute to a reduction in life-years lost and a substantial increase in healthcare system savings when compared to the NV scenario. These results were further reinforced in the PSP scenario, where both outcomes were projected to improve compared to the SQ scenario. The differences in life-years lost and health care savings between scenarios stem from demonstrated differences in health outcomes associated with smoking/vaping states. These results illustrate that scenarios with higher VNP use yield significant and positive outcomes compared to scenarios with lower levels of observed VNP prevalence, and hence higher smoking prevalence, in the study population. project the effects of 7 hypothetical scenarios, found that VNPs generated substantially more population-wide changes in smoking cessation than initiation. They observed that VNPs would have to increase smoking initiation by more than 100% to have a noticeable impact on smoking prevalence. These patterns are consistent with the results of our model.

Policy Implications
The flexibility of the AppEco model, namely its ability to analyse health outcomes and costs between different VNP use level scenarios, can be informative for public health decision-making. Though the results of this study reflect the effects of smoking/vaping prevalence in the Canadian context, the model can also be used as a policy tool for different subsets of any population across a variety of health modelling topics, most notably in contexts where the distribution of the study population among subgroups varies from one time-period to the next.
This study has two main results. First, the use of VNPs in Canada may be associated with positive and significant health and economic outcomes compared to a counterfactual scenario in which smoking cigarettes would have been the alternative option. Secondly, these outcomes may be further improved with wider use of VNPs by adult smokers. Our research suggests that more attention should be given to the potential benefits of authorized VNP products on the health of Canada's adult smoking population and its health care system. Henningfield et al. (2018) commented that in failing to recognize the potential positive impacts of VNPs on population health, policymakers could be negatively impacting overall net health outcomes. The absence or delay of a policy recognizing the potential benefits of VNPs could lead to more deaths due to SR diseases and higher health system costs than a policy that would see wider prevalence of VNPs (Paradis et al. 2012). Our results also suggest that policymakers ought to consider the potential benefits of lower-risk products, such as VNPs, on adult population health when developing regulations on product use and access. They can also work to increase public awareness regarding the risks and potential benefits of these lower-risk products for current smokers. In Canada and elsewhere, awareness initiatives could help fill the information gap for adult consumers and improve health and cost outcomes accordingly.
The Canada-wide scenario analysis for Indigenous peoples suggests a positive potential for this population. Tobacco smoking prevalence in Indigenous populations across Canada is markedly higher than that of non-Indigenous people,