Upload
others
View
3
Download
0
Embed Size (px)
Citation preview
Using System Dynamics modeling to assess the impact of launching e-cigarettes in the US market
Oscar M. Camacho1, Andrew Hill2 , Eleni Mavropoulou1 , Stacy Fiebelkorn1 , Christopher Poctor1 and James Murphy1
1 British American Tobacco, Research and Development, Regents Park Road, Southampton, SO15 8TL, United Kingdom2Ventana Systems UK, Salisbury, , United Kingdom
Poster 209 at SRNT Annual Meeting, 21st-24th 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 AssumptionsRelative 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 published1.
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 (SmokingHistory)
NGP Dual User Former NGP DualUser
Relapse to NGP
Relapse to NGP
Relapse toSmoking
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 NGPQuit 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 toSmoking
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
Un
der 5
years
5 to
9
years
10
to 1
4
years
15
to 1
7
years
18
to 2
4
years
25
to 2
9
years
30
to 3
4
years
35
to 3
9
years
40
to 4
4
years
45
to 4
9
years
50
to 5
4
years
55
to 5
9
years
60
to 6
4
years
65
to 6
9
years
70
to 7
4
years
75
to 7
9
years
80
to 8
4
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 RatesFemale 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)
Thousa
nd p
eople
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)
Dm
nl
Smokers Status Quo
E-cigarette
Smokers E-cigarette Scenario
Projected Value Vugrin Model 2050
SD Model 2050 SD Model 2100
Status Quo ScenarioSmoking 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)
Hun
dre
d t
ho
usa
nd
peo
ple
cumulative difference in deaths : Status Quo Scenariocumulative difference in deaths : Scenario Acumulative 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).