©
E-Cigarettes: Promise, Peril, and
Probabilistic Population Prediction
Bill Poland, PhD
INFORMS Annual Meeting, 10 November 2014
E-cigarettes, which deliver nicotine without carcinogenic tar, hold the
promise to save the lives of many smokers who switch to them, but risks
include failure to quit cigarettes (dual use), increased initiation to nicotine
products among youth, relapse of former smokers to e-cigarettes, and e-
cigarettes becoming a “gateway to smoking.” To capture these uncertainties
and weigh benefits vs. risks, prediction of e-cigarette health impacts must
use a broad range of probability-weighted scenarios.
©
After the “cigarette century”: what’s the end game?
2http://www.surgeongeneral.gov/library/reports/50-years-of-progress/index.html#fullreport
2014
©
Smoking still kills about 480,000 Americans per year, reducing life
spans 11-12 years.
3
http://www.surgeongeneral.gov/library/reports/50-years-of-progress/index.html#fullreport
©Sugerman DT. e-Cigarettes. JAMA 2014;311(2):212. http://jama.jamanetwork.com/article.aspx?articleid=1812964
* Herzog B, Wells Fargo Securities presentation, 2014. http://www.ecigarette-politics.com/files/WF-DallasMarch2014.ppt4
Sales of e-cigarettes have been roughly doubling annually since
US introduction in 2007 and could overtake cigarettes by 2023*,
raising hopes and fears.
©
Cigarettes deliver nicotine (addictive but relatively safe) and
tar (toxic partially combusted tobacco); e-cigs avoid the tar.
Nicotine
Produced by the tobacco plant as a natural pesticide
• Used as a pesticide in early 1900s
Highly addictive for many, mildly addictive for others
Suggestion but no solid evidence linking to cancer
• Ames assay for genotoxicity negative
• But promotes angiogenesis and tumors in some experimental models
Exposure during adolescence and in utero appears to cause long-term brain changes
Nicotine Replacement Therapy (NRT) such as patches and gum, and drugs like Chantix, minimize addiction risk but have a low success rate
Tar
Contains hundreds of mutagens,
carcinogens, and other toxins
• “Cigarette smoking has been causally
linked to diseases of nearly all organs
of the body”
• Surgeon General’s Report, 2014
• Top diseases: lung cancer, COPD,
CHD
• Also diabetes, rheumatoid arthritis, and
colorectal cancer, as well as
inflammation and impaired immune
function.
Damages lungs (coats the cilia
causing them to stop working and
eventually die) and mouth.
5
©
One eminent researcher concluded “e-cigarettes, with prudent ...
regulations, do have the potential to make the combusting of
tobacco obsolete ... just as digital cameras made film obsolete.”
6 Abrams DB. JAMA 2014;311(2): 135-136. http://jama.jamanetwork.com/article.aspx?articleid=1812971
©
Another researcher emphasizes a worst-case scenario.
Time for probabilistic modeling!
Best Case
Massive migration of smokers into e-cig vaping
• Like digital cameras replacing film
• CV and respiratory disease decline, followed by cancers
Dual use is temporary
Many e-cig users then quit nicotine entirely
• Users work their way down to low and no-nicotine e-cigs
• Total nicotine use drops
Youths who would have smoked take up e-cigs instead
Long-term vapor inhalation is found safe.
Worst Case
Massive migration of smokers and non-smokers into e-cig vaping
• Like cell phone adoption
• Non-users drawn in by purported safety
Dual use persists
• Mortality benefit of fewer cigarettes smoked/day is less than hoped
E-cigs attract youth and experimenters, addict them to nicotine, and become a “gateway to smoking”
E-cigarette advertising renormalizes tobacco product use so that smoking prevalence increases.
Long-term exposure to fine particles in vapor turns out to be harmful.
7
©
Population models predict smoking prevalence and
mortality under various scenarios.
8
Mendez D 2011. Modeling the Population Dynamics of Smoking Prevalence and Health Effects, Washington DC, 9 May2011. http://www.iom.edu/~/media/Files/Activity%20Files/PublicHealth/ReducedRiskTobacco/MendezPresentation.pdf
©
Mortality risk relative to nonsmokers has been modeled vs. years
since quitting. Quitting by 40 may return risk to non-smoker levels.
9
Mendez D, Warner KE 2001. “The relative risk of death for former smokers: the influence of age and years-quit.” Unpublished
research monograph. www.umich.edu/~dmendez/tobacco/RRiskmonograph.doc
©
Few attempts have been made to relate mortality to use levels
(intensity)—important with e-cigs. Here a Tobacco Exposure Index
balances lifetime smoke toxin accumulation with natural purging.
10
Miller LS et al. 2010. Evaluation of the economic impact of California’s Tobacco Control Program: a dynamic model
approach. Tobacco Control 19(Suppl 1):i68-i76. http://tobaccocontrol.bmj.com/content/19/Suppl_1/i68.full.pdf
never-smoker2 packs/day
©
A new generation of population models tries to predict
effects of two tobacco products and dual use.
11
Verzi S 2014.
http://www.fda.gov/downloads/Advis
oryCommittees/CommitteesMeetingMa
terials/TobaccoProductsScientificAdvis
oryCommittee/UCM394231.pdf
Cigarette/e-cig
“gateway
to smoking”
relapse
by former
smoker
via e-cigs
©
Even with simplifications, over a dozen highly uncertain
transition rates remain.
12
Cigarette/e-cig
x
x x
x
x
x
Assume
• one product change at a time
• no relapse, except former
smokers via e-cigs
x
“gateway
to smoking”
x x
x
x
relapse
by former
smoker
via e-cigs
xx
©
Data sources include:
For cigarette smoking initiation, prevalence, use levels, and cessation rates: US national surveys, with age and gender breakdown
• National Survey on Drug Use and Health (NSDUH)
• National Health Interview Survey (NHIS)
• Tobacco Use Supplement to the Current Population Survey (TUS-CPS)
• A decade or more of annual data allows model calibration
For smoking-related mortality: large survey-based relative risk studies
• By age, gender, use levels, time since quitting
For e-cig use patterns:
• Randomized controlled studies of smokers offered e-cigs
• Surveys of e-cig use (problematic due to biases)
• Studies of Swedish snus as a possible e-cigarette analog
• In the 1970s snus started displacing cigarettes in Sweden, resulting in the lowest smoking rate in Europe for the past 15+ years.
• BUT e-cig technology and use patterns are evolving quickly!
13
©
Assuming a linear system with constant coefficients results in a weighted
sum of matrix exponentials describing the states over time, e.g., smoker,
e-cig user, dual user, former user, and never-user categories.
14
Continuous Discrete
Equation
(x is a vector, A &
P are matrices)
dx/dt = A x + b,
x(0) = x0
x(t+1) = P x(t) + B (t=0, 1, ...),
x(0) = x0
Units A: units of x/t (≥0)
b: units of x
P: no units (range 0-1)
B: units of x
xEquilibrium: x(∞)
(dx/dt = 0,
x(t+1) = x(t))
-A-1 b (I–P)-1 B
Solution x(t) eAt x0 + (I-eAt) xEq Pt x0 + (I-Pt) xEq
Equivalent if: P = eA,
B = -(I-P) A-1 b
Exponential approach to equilibriumExponential approach to equilibrium
Proportional transition rates Constant growth rateConstant growth rate“Survival” proportion
Most population models are discrete, but continuous and discrete linear models are equivalent:
©
One insight from such linear models is that even if parameters are
fixed, equilibrium (-A-1 b) may take centuries to reach.
15
Current Smokers
Former Smokers
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
2000 2050 2100 2150 2200
Ne
ve
r-S
mo
ke
r P
rop
ort
ion
Sm
oke
r P
rop
ort
ion
Year
Note Former Smoker prevalence increases, then decreases.
©
For flexibility, we use quasi Monte Carlo simulation of individual
tobacco use histories across a large population.
Most population health effect models define Markov states and
calculate annual proportions of users in each state (deterministically).
Advantage of individual simulation: the number of states considered is
no longer a limitation.
• E.g., mortality can be made a function of year of age, gender,
smoking intensity, e-cig use status, former status history, etc
Disadvantage: tens of thousands of individuals need to be simulated for
stable results in each sub-category.
Use of quasi Monte Carlo numbers (selected to cover the space rather
than fully random) reduce this disadvantage, by reducing the number of
simulated individuals needed by several-fold.
16
©
Probabilistic analysis translates input uncertainty into
output uncertainty.
Wide input distributions result in wide distributions on possible net
health impacts of e-cigs, from negative to positive.
• Wide distributions are more “honest” than point estimates
Deterministic and probabilistic sensitivity analysis
show which variables are most important for
reducing uncertainty on health impacts.
• Tornado charts sweep one variable at a time
through a range, sorting by bar length to form
a tornado shape.
Breakeven analysis shows how much beneficial effect (like switches
from smoking to e-cigs) is needed to offset a harmful effect (like
switches the other way).
• Called “tipping point analysis” in Swedish Match’s 2014 snus
application to the FDA’s Center for Tobacco Products.
17
0.80
1.00
0.38
0.92
0.9%
0.4%
0.35
1.20
1.85
0.60
1.10
2.7%
1.2%
0.45
Base
-200,000 0 200,000 400,000 600,000
©
Key tasks in evaluating net health effects of e-cigs include:
Improve estimates and ranges of e-cig transition rates
• Initiation, full and partial switching from cigarettes, relapse back to cigarettes, “gateway” from e-cigs to cigarettes, quit rates
• Snus analogy is useful as a scenario
• Elicit subjective ranges from an expert panel?
• As a function of time (e.g., 2020 and 2050)
• Conditioned on various e-cig market growth scenarios
Adjust mortality rates to account for history of cigarette use levels
• Critical because smokers trying e-cigs reduce, but are slow to eliminate, smoking.
Predict impact of e-cigs on morbidity as well as mortality
• Morbidity effects show up sooner than mortality effects.
18
©
Thank you! Questions?
19
Abrams 2014. e-Cigarettes: Can we use them to make combusting of tobacco obsolete - end the “cigarette century” and its
preventable deaths? Presentation to American Academy of Health Behavior 14th Annual Meeting.
http://www.aahb.org/Resources/Pictures/Meetings/2014-Charleston/PPT%20Presentations/Sunday%20Welcome/Abrams.AAHB.3.13.v1.o.pdf