Ventana Systems UK, Salisbury, , United Kingdom

Page 1

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).