Using System Dynamics modeling to assess the impact of launching e-cigarettes in the US market Oscar M. Camacho 1 , Andrew Hill 2 , Eleni Mavropoulou 1 , Stacy Fiebelkorn 1 , Christopher Poctor 1 and James Murphy 1 1 British American Tobacco, Research and Development, Regents Park Road, Southampton, SO15 8TL, United Kingdom 2 Ventana Systems UK, Salisbury, , United Kingdom Poster 209 at SRNT Annual Meeting, 21 st -24 th February 2018, Baltimore, USA Model Structure The model is initiated with data from the year 2000 and it is not a cohort model, it aims to represent the whole of the population, including births, deaths and migration rates. All possible transitions for a two-nicotine product model are considered with differentiation between current/former and dual NGP users with smoking history and without. The reason of separating stocks with different smoking history is not only because those categories are likely to have different relative risks but it is also necessary to investigate initiation from never and former smokers. Input Data Study/Report/Survey Data Source Adult Smoking Prevalence National Health Interview Survey (NHIS) for years 2000 to 2003 and 2005 to 2012. https ://www.cdc.gov/nchs/nhis/data- questionnaires-documentation.htm (Accessed 13Feb2018) Youth Smoking Prevalence National Youth Tobacco Survey (NYTS) for years 2000, 2002, 2004 , 2006, 2009, 2011 and 2012. https ://www.cdc. gov/tobacco/data_statist ics/surveys/nyts/index .htm (Accessed 13Feb2018) Smoking Initiation Rates Holford TR, Levy DT, McKay LA, et al. Patterns of Birth Cohort–Specific Smoking Histories, 1965–2009. American journal of preventive medicine. Mortality Rates National Vital Statistics Reports, Volume 50, Number 15, Deaths: Final Data for 2000 https://www.cdc.gov/nchs/data/nvsr/nv sr50/nvsr50_15.pdf (Accessed 13Feb2018) Relative Risks National Center for Chronic Disease Prevention and Health Promotion (US) Office on Smoking and Health. The Health Consequences of Smoking—50 Years of Progress: A Report of the Surgeon General. https://www.surgeongeneral.gov/library /reports/50-years-of-progress/full- report.pdf (Accessed 13Feb2018) Office for National Statistics In 2012, the FDA suggested using mathematical models as tools for assessing the impact in terms of population health outcome of releasing new nicotine or tobacco products. Since then, several models have been developed using different approaches (1,2,3). These models, although based on distinct underlying methodologies, all of them try to provide simplified representations of the behaviours and mechanisms associated with nicotine use such as, initiation, switching and quitting nicotine use. Projections from models rely on relevant historic data and/or assumptions, which are generally expressed in a comparative manner with hypothesised scenarios. As response to this guidance, BAT in collaboration with Ventana Systems UK has developed a System Dynamics compartmental model for two nicotine product categories (2). This initial model was built and calibrated using data from the United Kingdom. To better represent the US population 4 race/ethnicity categories have been included in the model. In addition, there has been a further break down of age categories to increase model resolution and some mechanisms have been simplified. This new model configuration is used to investigate scenarios as result of launching e- cigarettes in the US. In this poster we focus on the models assessment aspects. We compared our projections to official population projections and projections from other published model in an attempt to ‘validate’ the outcomes from our model. Introduction Table 1. Model conceptual and structural assumptions. Table 2. Data sources for smoking related data inputs in US population. Lack of data and differences on the definitions among data sources and data collection methodologies provided inconsistent inputs which made necessary the introduction of assumptions and calculation of some parameters through model calibration. Comparative scenarios with respect to other published data and projections from other models suggest that our SD model yields sensible outputs which could provide valuable information to assess nicotine products in terms of population health outcomes. Conclusions Main Assumptions Relative risks of under 35 years old is the same across nicotine use statuses Dual users have the same RRs as current smokers People relapsing to smoking will have the same RRs that any other smokers of that age category (there is not a benefit from quitting smoking for short periods of time) Nicotine usage initiation rates start to be applied from the age of 10 and before that age are considered 0. Aim The aim of this work was to further develop a compartmental population impact model based on System Dynamics methodology and assess its applicability to real life data. Approaches for ‘validating’ the outcomes of these type of models are also assessed. Methods System Dynamics Different smoking statuses are represented by stocks (compartments) and arrows represent the flows (Figure 1). It allows representation of complex non-linear mechanisms, including feedback effects, by simply calculating inbounds and outbounds based on integration of flows in relation to time. Smokers NGP Users Stocks Flows NGP initiation rate Figure 1. Stock and flows are the basic elements of System Dynamics Figure 6. Smoking and E-cigarette prevalence from SD model Scenario A (Top) and Vugrin model (Middle (2)). Below are cumulative deaths with respect of Status Quo scenario and scenarios A and B from SD model. US population, e-cigarettes and comparative number of deaths and life-years saved as health outcome of interest We investigated the potential benefit of launching e-cigarettes by comparing scenarios with different switching rates from smoking to sole e-cigarette use. Data inputs The model was initialised at year 2000 with US demographic data including smoking prevalence by gender, age and race/ethnicity categories as well as birth and death rates by these same categories. The available data provided a calibration period of 13 years up to 2012 and with a time step of a year. Data sources are listed in Table 2. Mortality relative risks between smokers and never smokers were extracted from a report of the surgeon general (Figure 2). These estimates were provided by age and gender but race specific estimates were not available. RRs for former smokers were calculated based on the negative exponential curve previously published 1 . Similarly, smoking initiation rates were not readily available by race category (Figure 3. Correspondence: [email protected] References 1. Hill A, Camacho OM. A system dynamics modelling approach to assess the impact of launching a new nicotine product on population health outcomes. Regulatory Toxicology and Pharmacology 2017 Jun;86:265-278. 2. Vugrin ED, Rostron BL, Verzi SJ, Brodsky NS, Brown TJ, Choiniere CJ, et al. Modeling the Potential Effects of New Tobacco Products and Policies: A Dynamic Population Model for Multiple Product Use and Harm. PLoS ONE 2015 10(3): e0121008. 3. Bachand AM and Sulsky SI. A dynamic model for estimating all-cause mortality dues to lifetime exposure history. Regulatory Toxicology and Pharmacology 2013; 67 (2): 246-51. Never Smoker Current Smoker Former Smoker NGP User (NeverSmoker) Former NGP User (Never Smoker) Former NGP User (Smoking History) NGP User (Smoking History) NGP Dual User Former NGP Dual User Relapse to NGP Relapse to NGP Relapse to Smoking Relapse to Smoking Relapse to Smoking Initiate NGP Initiate Smoking Switch to Smoking Initiate Dual Use NGP to Dual Use Relapse to Dual Use Relapse to Dual Use Initiate Smoking Switch to NGP Quit Smoking Quit NGP Quit NGP Smoker to Dual Use NGP to Dual Use Revert to NGP Revert to Smoking Relapse to Dual Use Dual Use to NGP Relapse to Smoking Quit Birth Rate Figure 2. Representation of the full model. This includes all possible outcomes that can occur at a time step. The new model structure includes 4 race categories as White, Black or African American, Hispanic not black and Other. We have increased granularity for time-steps by extending the age range and increasing the number of age categories to: under 5, 5 to 9, 10 to 14, 15 to 17, 18 to 24 and then 5 year cohorts up to 85+ years. Data gaps for other model inputs were filled by calibration. These include race adjustments to initiation rates, quitting rates by age and gender which was also scaled for race categories. Figure 4 illustrate quitting rates for females after calibration and initiation rates by race category. Underlying Model Assumptions The model is built on two different types of assumptions: 1. conceptual and structural assumptions form part of the core model and do not change with implementation. These assumptions relate to the methodological limitations in the modelling approach to represent real world complexity so simplifications for some mechanisms are introduced. In this category of assumptions we also include those beliefs that are widely accepted by the scientific community, for example, disease relative risks (RRs) for smokers and never smokers are not different before the age of 35 years old. Type 1. assumptions are displayed in Table 1. 2. The second type of assumptions are directly related to data availability (or lack of it) for a specific implementation (Table 3). 0.0% 1.0% 2.0% 3.0% 4.0% 5.0% 6.0% 7.0% 10-14 15-17 18-24 25-29 Age Cohort Annual Smoking Initiation Rates Male Female 0.00 0.50 1.00 1.50 2.00 2.50 3.00 3.50 Under 5 years 5 to 9 years 10 to 14 years 15 to 17 years 18 to 24 years 25 to 29 years 30 to 34 years 35 to 39 years 40 to 44 years 45 to 49 years 50 to 54 years 55 to 59 years 60 to 64 years 65 to 69 years 70 to 74 years 75 to 79 years 80 to 84 years 85+ years Current Smoker Male Current Smoker Female Smoker Relative Risk Figure 3. Smoker RRs (Left) and smoker initiation rates (Right) by age and gender. Scenarios With the data presented to this point we generated a Status Quo scenario, i.e., without considering e-cigarette use. Projections from this Status Quo scenario were assessed against US Census data to confirm that was able to draw sensible projections (Figure 5). 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 WHITE BLACK OR AFRICAN AMERICAN OTHER RACE HISPANIC AGED 10 TO 14 YEARS AGED 15 TO 17 YEARS AGED 18 TO 24 YEARS AGED 25 TO 29 YEARS 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 WHITE BLACK OR AFRICAN AMERICAN OTHER RACE HISPANIC Figure 4. Smoking initiation (Right) and quitting rates (Left) for females by age and gender. Female Smoker Quitting Rates Female Smoking Initiation Rates Population (000s) 500,000 375,000 250,000 125,000 0 2000 2005 2010 2015 2020 2025 2030 2035 2040 2045 2050 2055 2060 2065 2070 2075 2080 2085 2090 2095 2100 Time (Year) Thousand people US Census Projection total population : Status Quo Scenario 2100 5.0 4.0 3.0 2.0 1.0 0.0 1.0 2.0 3.0 4.0 5.0 Under 5 years 5 to 9 years 10 to 14 years 15 to 17 years 18 to 24 years 25 to 29 years 30 to 34 years 35 to 39 years 40 to 44 years 45 to 49 years 50 to 54 years 55 to 59 years 60 to 64 years 65 to 69 years 70 to 74 years 75 to 79 years 80 to 84 years 85+ Percentage of Population Population Age Distribution at 2050 Female Census Projection Female Model Male Census Projection Male Model Figure 5. US population projection Status Quo scenario up to 2100 vs. Census projection (Left) and Population distribution by age and gender from Status Quo scenario vs. US Census projections (Right). Alternative Scenarios To facilitate cross-model comparisons we used the values published by Vugrin et al. (2) for our main alternative scenario, referred as Scenario A (Table 3) and then we changed the RR for e-cigarettes to 1.05 with respect to never smokers (Scenario B). Transition Assumption NS to EC Smoking Initiation * 0.5 NS to Dual use 0 CS to EC 1.5% annually CS to Dual use 1.5% annually EC(NS) to CS 5% annually EC(NS) to FEC(NS) Smoking Quit Rate * Scalar EC(NS) to Dual use 5% annually EC(SH) to CS Smoking Initiation Rate EC(SH) to Dual use Smoking Initiation Rate EC(SH) to FEC(SH) Smoking Quit Rate * Scalar Dual user to CS Smoking Quit Rate * Scalar Dual user to EC(SH) Smoker Quit Rate Dual user to Former Dual User (SH) Smoking Quit Rate * Scalar Additional Parameters Proportion of switchers and dual users coming from smokers that would have quit in that year 0.25 Proportion of new product initiates who would have otherwise initiate cigarettes in that year 0.5 NGP Relative Risk Scalar 0.25 NGP(SH) Quit Probability Scalar 1 NGP(NS) Quit Probability Scalar 1 Dual User Quit Probability Scalar 1 Smokers & E-cigarette Prevalence 30 22.5 15 7.5 0 2000 2005 2010 2015 2020 2025 2030 2035 2040 2045 2050 2055 2060 2065 2070 2075 2080 2085 2090 2095 2100 Time (Year) Dmnl Smokers Status Quo E-cigarette Smokers E-cigarette Scenario Projected Value Vugrin Model 2050 SD Model 2050 SD Model 2100 Status Quo Scenario Smoking Prevalence 12.5% 9.6% 7.8% Scenario A (25% Harm) Smoking Prevalence 11.7% 9.3% 7.7% Scenario A (25% Harm) E-cig Prevalence 8.0% 6.6% 6.2% Scenario A (25% Harm) Lives Saved 175,000 187,000 63,000 Life Years Saved - -- 6.1M 3.0M Difference in Cumulative Deaths 0 -1.5 -3 -4.5 -6 2000 2010 2020 2030 2040 2050 2060 2070 2080 2090 2100 Time (Year) Hundred thousand people cumulative difference in deaths : Status Quo Scenario cumulative difference in deaths : Scenario A cumulative difference in deaths : Scenario B - Risk Ratio 0-05 Table 3. Assumptions for scenario A. Table 4. Comparison of projections between different scenarios. SD model projections are systematically lower than Vugrin model (2), however when assessed comparatively vs. Status Quo ,i.e., lives saved, both seem to reach comparable conclusions (Table 4). Reinforcing this observation, nicotine use behaviours suggest to follow similar patterns (Figure 6 Top and Middle). With Scenario B we investigate the SD model sensitivity to the relative risk parameter (Figure 6. Bottom).