Download Development and Validation of a Microsimulation Economic

VALUE IN HEALTH 14 (2011) S51–S59
available at www.sciencedirect.com
journal homepage: www.elsevier.com/locate/jval
Development and Validation of a Microsimulation Economic Model to
Evaluate the Disease Burden Associated with Smoking and the CostEffectiveness of Tobacco Control Interventions in Latin America
Andres Pichon-Riviere, MD, MSc, PhD1,2,*, Federico Augustovski, MD, MSc1,2, Ariel Bardach, MD, MSc1,
Lisandro Colantonio, MD, MSc1, for the Latinclen Tobacco Research Group3
1
Institute for Clinical Effectiveness and Health Policy, Buenos Aires, Argentina; 2School of Public Health, University of Buenos Aires, Buenos Aires, Argentina;
LatinCLEN Tobacco Research Group: Argentina (Institute for Clinical Effectiveness and Health Policy): Andres Pichon-Riviere, MD, MSc, PhD, Adolfo
Rubinstein, MD, MSc, PhD, Federico Augustovski, MD, MSc, Ariel Bardach, MD, MSc, Lisandro Colantonio, MD, Andrea Alcaraz, MD, Sebastian Garcia-Marti,
MD, Luz Gibbons, MSc; Argentina (Clinical Epidemiology Unit–University of Tucuman): Evelina Chapman, MD, MSc, Ramon Gonzalez, MD, Sara Aulet, MD;
Bolivia (Clinical Epidemiology Unit–San Andres University): Maria del Pilar Navia Bueno, MD, MSc, Lijia Elizabeth Avilés Loayza, MD, MSc; Brazil
(Universidade Federal do Rio de Janeiro): Antonio Jose Ledo Alves da Cunha, MD, MPH, PhD, Alberto José de Araujo, MD, MSc; Chile (Research and Training
Center in Clinical Epidemiology–Frontera University): Sergio Muñoz, MS, PHD, Carlos Vallejos, MD, MS; Colombia (Clinical Epidemiology & Biostatistics Unit–
Pontificia Universidad Javeriana): Juan Manuel Lozano Leon, MD, MSc, Esperanza Peña Torres, RN, MSc; Mexico (Clinical Epidemiology Unit–Instituto
Mexicano del Seguro Social–Faculty of Medicine): Fernando Carlos Rivera, MSc, Araceli Camacho Chairez, MSc, Araceli Aguirre Granados, Patricia Clark
Peralta, MD, MSc, PhD; Peru (Clinical Epidemiology Unit–Universidad Peruana Cayetano Heredia): Leandro Huayanay Falconi, MD, MSc, Cezar Loza Munárriz, MD
3
A B S T R A C T
Objective: To describe the development and validation of a health economic model (HEM) to address the tobacco disease burden and the
cost-effectiveness of smoking cessation interventions (SCI) in seven
Latin American countries. Methods: The preparatory stage included
the organization of the research network, analysis of availability of
epidemiologic data, and a survey to health decision makers to explore
country-specific information needs. The development stage involved
the harmonization of a methodology to retrieve local relevant parameters and develop the model structure. Calibration and validation was
performed using a selected country dataset (Argentina 2005). Predicted
event rates were compared to the published rates used as model inputs. External validation was undertaken against epidemiologic studies that were not used to provide input data. Results: Sixty-eight decision makers were surveyed. A microsimulation HEM was built
considering the availability and quality of epidemiologic data and relevant outcomes conceived to suit the identified information needs of
Introduction
Smoking is the single most preventable cause of disease and death
worldwide, and this burden is increasingly shifting from upper to
lower and middle-income countries. In the year 2000 there were 4.83
million premature tobacco-related deaths [1], and this number is expected to grow to 10 million per year by 2030 [2,3]. Currently half of
the current tobacco-attributable deaths occur in high-income countries [1,3]; however, by the year 2030 7 out of 10 of these deaths are
expected to occur in developing countries. This represents one out of
six of all the deaths around the world [3].
decision makers. It considers all tobacco-related diseases (i.e., heart,
cerebrovascular and chronic obstructive pulmonary disease, pneumonia/influenza, lung cancer, and nine other neoplasms) and can incorporate individual- and population-level interventions. The calibrated
model showed all simulated event rates falling within ⫾ 10% of the
sources (-9%–⫹5%). External validation showed a high correlation between published data and model results. Conclusions: This evidencebased, internally and externally valid HEM for the assessment of the
effects of smoking and SCIs incorporates a broad spectrum of tobacco
related diseases, SCI, and benefit measures. It could be a useful policymaking tool to estimate tobacco burden and cost-effectiveness of SCI.
Keywords: cost-effectiveness, cost-utility, disease burden, economic
model, Latin America, Monte Carlo microsimulation, smoking cessation interventions, tobacco, validation.
Copyright © 2011, International Society for Pharmacoeconomics and
Outcomes Research (ISPOR). Published by Elsevier Inc.
Although the Framework Convention Tobacco Control from
World Health Organization has been signed by almost every country in the Latin American region [4], tobacco control policies are
still scarce in these countries.
The lack of quality information related to the health and
economic consequences of tobacco use in our region is an important barrier for the implementation of evidence-based tobacco control policies. This has led to a biased assessment by
policy makers, resulting in a distorted prioritization of health
policies where tobacco control interventions are considered
less urgent than action on other diseases [5].
Conflicts of interest: The authors have indicated that they have no conflicts of interest with regard to the content of this article.
* Address correspondence to: Andres Pichon-Riviere, Institute for Clinical Effectiveness and Health Policy, University of Buenos Aires,
Arevalo, 1457 (1414), Buenos Aires, Argentina.
E-mail address: [email protected].
1098-3015/$36.00 – see front matter Copyright © 2011, International Society for Pharmacoeconomics and Outcomes Research (ISPOR).
Published by Elsevier Inc.
doi:10.1016/j.jval.2011.05.010
S52
VALUE IN HEALTH 14 (2011) S51–S59
Model-based health economic evaluations (HEEs) are widely
accepted as decision-making tools [6] that can provide valuable
information for the optimization of health resource allocation
[7]. Although in many developed countries this “fourth hurdle”
based on health economic evidence is required to shape health
policies [8], there is still little experience in Latin America [9].
This project constitutes a collaboration among seven Latin
American countries that aims to provide relevant evidence to
inform tobacco control policies.
The LatinCLEN Tobacco Research Group, an international
and interdisciplinary network, is composed of eight research
units from the Latin American chapter of the International Clinical Epidemiology Network in Argentina, Bolivia, Brazil, Chile,
Colombia, Mexico, and Peru. The specific aims for this project
were to select and develop the most suitable methodologic
framework, as well as to elaborate a common health economic
model to estimate the smoking-related disease burden and the
cost-effectiveness of smoking cessation interventions (SCIs). In
this article we present the details of the model’s development,
structure, and validation using data inputs from one of the participating countries.
Material and Methods
The final model structure and its inputs were agreed after two
stages; the preparatory stage and the development stage.
Preparatory stage
Organization of a research network to monitor the model building process and to ensure the generalizability to all participant
countries. The LatinCLEN Tobacco Research Group was formed in
2004. The group designed two surveys that were completed in
each country to 1) evaluate the availability, cost, and current coverage policies of SCIs; and 2) assess the availability and the quality
of relevant information to be incorporated in each country-specific analysis (e.g., local epidemiology and cost of smoking-related
diseases).
Performance of a rapid systematic review of existing health economic evaluations regarding tobacco cessation strategies. Fortyfour individual studies and seven reviews published between 1984
and 2003 (search data up to November 2004) were critically assessed.
Design and administration of a survey to health decision makers
to explore country-specific information needs when deciding on
the implementation and coverage of SCIs. As future users of the
HEE, relevant decision makers from the different health sectors
in the seven participant countries defined the key aspects to be
considered (e.g., relevant time horizon and relevant perspective) and the outcomes to be reported (e.g., the number of cases
prevented, life years gained, or quality-adjusted life years
[QALYs]) in the HEE.
Development stage
This stage involved the following tasks: 1) definition of the methodology for the information source selection and parameter incorporation; 2) development of the model structure; and 3) calibration
and validation of the model.
The research group used Email and a Web-based platform to
exchange documents, outlines, and ideas. The development of
the model was completed in three phases: 1) based on information obtained in the preparatory stage, a first draft of the health
economic model was sent to the participating countries for
feedback, including the basic structure, disease states to be incorporated, and main assumptions; 2) three consultation
rounds for refining the model description and structure; and 3)
a face-to-face research meeting carried out in Buenos Aires,
Argentina, during November 2006 where participants agreed on
the final version of the HEE.
Excel (Professional Edition 2003, then updated to version 2007,
Microsoft Corp., Redmond, WA) with Visual Basic Macros (version
6.3, Microsoft Corp., Redmond, WA) was selected as the model
platform to easily share the information. A software package was
installed to improve the original Excel’s random number generator function [10,11].
Results
Preparatory stage
The surveys and data retrieved in each country showed that
implementation of tobacco control interventions was a relevant
issue in the region. Sixty-eight decision makers (9 –10 from each
of the participating countries) completed the survey. The majority of decision makers belonged to the public (56%) or the
social security (25%) health care sector and 80% considered that
the lack of coverage for SCI adversely affected the prevalence of
smoking in their institutions and countries. Ninety-three percent considered that this level of coverage should be increased
and 83.3% believed that SCI should be included in the national
lists or basket of mandatory coverage in their countries. When
asked about what would be the most relevant information on
the interventions needed when having to decide their incorporation into the health system, most of the decision makers identified the cost per QALY, cost per life-year saved, and budget
impact information. The decision makers also mentioned a
wide range of interventions that they considered should be
evaluated for coverage, from population-wide interventions to
pharmacological treatments, so the HEE had to be able to include all these aspects. On the other hand, the survey of epidemiologic data showed that the availability and quality of information in the region was very heterogeneous and poor,
especially with regard to the incidence of events, thus making it
necessary to design and harmonize a methodology to estimate
locally relevant information in each country. All the results obtained during this first stage strongly influenced many of the
decisions made later and shaped the type and structure of the
model to be developed.
Development stage
Information source selection and parameter incorporation. We
defined a decision rule that would establish a priority order among
the possible data sources to populate the model: 1) use good quality local (country-specific) sources when available [12-14]; 2) use
international sources when local data were unavailable or poor
and when the parameter was considered transferable from other
settings; or 3) derive or estimate the parameter from the best available local data when international sources were considered nontransferable.
Special attention was paid to the estimation of baseline disease
event incidence in nonsmokers because these data are keys to the
generalizability of the model. Given the low availability of information encountered in the region, we defined a common methodology, anchored on national health statistics, to derive these parameters from mortality data. This methodologic assumption
linking mortality to incidence data is a widely used assumption in
epidemiologic and health economic models, used by the World
Health Organization in tools such as DisModII or the WHO-
VALUE IN HEALTH 14 (2011) S51–S59
CHOICE, and by GLOBOCAN [15-20]. Different approaches were
taken for acute events or chronic conditions. For acute events,
such as cardiac or cerebrovascular events, the first step was to
obtain the age-, sex- and country-specific absolute risk of the
event based on the specific mortality rate and the lethality of the
event:
Rpop.event ⫽
Rdeath
(1)
L
where L is the lethality of the event and Rdeath is the age- and
sex-specific mortality of the condition. Once this absolute risk is
known, the baseline risk in nonsmokers was calculated based on
the age-, sex- and country-specific smoking prevalence as well as
disease-specific smoking relative risk:
Rnosmk ⫽
Rpop.event
(RRsmk ⫻ fsmk) ⫹ (RRformersmk ⫻ fformersmk) ⫹ fnosmk
(2)
where Rnosmk is the baseline event annual incidence in non-smokers, Rpop.event is the age- and sex-specific population risk (obtained
with formula 1), RRsmk and RRformersmk are the relative risks of the
event in smokers and former-smokers versus nonsmokers, and
ƒsmk, ƒformersmk and ƒnosmk are the age- and sex-specific proportion
of smokers, former-smokers, and nonsmokers. For cancer (as
chronic conditions), the age and sex estimation of the probability
of diagnosis was calculated using a more complex approach that
considered both the annual mortality rate from national statistics
as well as the estimated yearly survival rate since diagnosis. The
age- and sex-specific risk of diagnosis for each cancer was calculated with the following formula:
Rdxi ⫽
冋兺
10
n⫽0
册
Rmi ⫻ Pn ⫻
1
1 ⫺ S10
(3)
where Rdxi is the risk of diagnosis at age i; Rmi⫹n is the population
risk of death from the specific cancer at age i⫹n; Pn is the conditional probability of dying in year n after the diagnosis, conditional
on dying within 10 years; and S10 is the proportion of survivors
after 10 years. We assumed that those subjects surviving 10 years
after a lung cancer diagnosis, or five for other cancers, return to the
general population risk of death. Then, the formula (2) was applied
to derive the baseline risk in nonsmokers.
A special case was chronic obstructive pulmonary disease
(COPD). Because national statistics are known to significantly underestimate COPD-related mortality, we estimated its incidence
and prognosis based on international studies [21,22].
Model structure and operation. A first order Monte Carlo, or probabilistic microsimulation of individual subjects, was built. This
model incorporates the natural history, costs, and quality of life of
all the tobacco-related adult-specific diseases: coronary and noncoronary heart disease, cerebrovascular disease, COPD, pneumonia, influenza, lung cancer, and nine other neoplasms. This model
allows the follow-up of the lives of thousands of individuals in
hypothetical cohorts, calculating all outcomes for each patient in
an annual basis, using the simulation of each individual’s history
to ultimately obtain aggregated population results in terms of
health and costs. Subjects can be assigned demographic and disease specific characteristics. The model updates the values of the
various input parameters for each patient in a yearly basis and
calculates event rates for outcomes on the basis of the variables
and the underlying risk equations.
The model runs on Visual Basic and captures the key parameters from four main spreadsheets: 1) sex- and age-specific epidemiologic data; 2) unit cost; 3) quality of life; and 4) interventions
effects. It consists of two main modules. The first one is used for
the analysis of the disease burden associated with smoking, in
which age and sex detailed information of selected epidemiologic
S53
and economic data is kept (e.g., the number of events suffered by
the cohort throughout specific ages, the distribution of COPD
stages or the prevalence of coronary heart disease). The second
module is the one that performs the cost-effectiveness analysis,
and it focuses on the comparison of the experience of two cohorts
for whom different sets of interventions are defined. The cohorts
are then followed during their lifetime, and the aggregated results
are compared in terms of costs and benefits.
Disease incidence, progression, and mortality. For each time period, the model estimates the individual risks of occurrence of
each event, disease progression and death, based on the subject’s
demographic attributes, smoking status, and clinical conditions.
Table 1A shows a list of possible events a subject can suffer in each
time period and the calculation method used to derive it. Table 1B
shows the different health states considered in the model. The risk
of death is calculated for each time unit as the age- and sex-specific general mortality, excluding the 16 disease-specific risks of
death considered in the model, plus the risk of death of the events
and conditions that the individual experiences during that time
unit (Table 1C).
For instance, in the case of myocardial infarction the model
estimates its risk multiplying the age-and sex-specific risk in nonsmokers (baseline incidence) in each time period and for each
subject, by the relative risk related to his smoking status. Then, a
random number is generated; if this number is less tan or equal to
the individual probability, the model assumes that the event is
present. The risk of death, in this case, will be the general risk of
death plus the myocardial infarction age- and sex-specific case
fatality. When there is more than one simultaneous cause of death
for a given subject in the same time period, a probabilistic approach is used to assign the final cause of death, weighted by the
baseline risk of each competing cause.
For COPD, besides the risk of acquiring this condition, the risk of
progression to more severe stages in those already affected is estimated according to the individual’s smoking status. For oncologic
conditions, the specific risk of death is estimated according to the
number of years since diagnosis. This type of individual-based model
allows for having multiple events in a given year, as they are not
mutually exclusive. This was the main reason to choose this model
instead of a state transition cohort model (i.e., Markov cohort).
Smoking status and interventions. We considered three smoking status states: current smokers, former smokers, and never
smokers. Smokers have a given probability of making a quit
attempt in each time unit as well as a probability of succeeding
in that attempt without any active intervention (background
quitting rates). These probabilities are age- and sex-related, and
in case of considering a population SCI, its effect could be modeled by directly influencing this background quitting rate (Table
1A). Similarly, former smokers have a given relapse probability
related to the time elapsed since the successful quit attempt
(background relapsing rates). The model was built to consider a
wide range of intervention modalities: 1) interventions/policies
with the objective of improving smoking cessation rates in
smokers who have a quit attempt (e.g., nicotine replacement or
behavioral interventions); 2) interventions/policies that increase the probability of smokers attempting to quit (e.g., media
campaign) and; 3) mixed interventions/policies that influence
both the probability of attempting to quit and the success rates
(e.g., training primary care physicians in brief counseling interventions, including pharmacotherapy in benefit plans).
Resource use, cost, and quality of life. The model is programmed
to calculate the use of resources and the QALYs in each time unit
as a summary of the events the subject experienced in that particular time unit with the active health conditions coming from
S54
VALUE IN HEALTH 14 (2011) S51–S59
Table 1 – (A) Acute events, (B) chronic disease states, and (C) causes of death included in the model.
A. Acute events
Disease acute events
MI
Non MI CHD event
Stroke
COPD diagnosis
COPD progression
Pneumonia/Influenza
Cancer diagnosis: lung, bladder, renal, lip/oral/pharynx,
larynx, stomach, esophagus, pancreas, cervical
cancer, leukemia
Smoking behavior events
Performing a quit attempt
Succeeding in a quit attempt
Relapsing after successful quit attempt
Probability calculation
Disease events:
Baseline Risk in non-smokers (age/sex specific) x
RRsmoking.status
COPD progression:
Baseline Risk in non-smokers (sex and years-inprevious-stage specific) x RRsmoking.status
Performing or succeeding quit attempt:
Baseline population probability (age/sex specific) x
RR.INTERVENTION
Relapsing:
Risk based on years in former-smoker state (sex
specific)
B. Chronic
disease states
CHD patient
Post - Stroke
COPD Stage
Lung cancer
Bladder cancer
Renal cancer
Lip/oral/pharynx
cancer
Larynx cancer
Stomach cancer
Esophagus cancer
Pancreas cancer
Cervical cancer
Leukemia
Smoking status:
Smoker
Former smoker
Never smoked
C. Causes of death
MI
Non MI CHD event
Stroke
Pneumonia/Influenza
Non-ischemic CV death
COPD
Lung cancer
Bladder cancer
Renal cancer
Lip/oral/Pharynx cancer
Larynx cancer
Stomach cancer
Esophageal cancer
Pancreas cancer
Cervical cancer
Leukemia
Mortality for all other causes
Probability calculation
Acute events deaths:
Probability of the event x its lethality (age/sex specific)
Non-ischemic CVD death:
Baseline Risk in non-smokers (age/sex specific) x
RRsmoking.status
COPD:
Stage specific mortality (sex specific)
Cancer:
Tumor annual specific mortality during the first five years
after diagnosis (except lung cancer: 10 years)
Mortality for all other causes
General population mortality minus the mortality of the
diseases included in the model (sex and age specific)
CHD, coronary heart disease; COPD, chronic obstructive pulmonary disease; CV,cardiovascular; MI, Myocardial infarction; RR.INTERVENTION,
relative risk of the intervention (either to improve the probability of performing a quit attempt or to improve the success rate of the quit attempt);
RRsmoking.status, disease specific relative risk according to smoking status.
prior time units. Costs and QALYs are simultaneously calculated
in undiscounted and discounted fashions.
Main assumptions. The occurrence of all the modeled events is
independent and not mutually exclusive. In the case of coexistence
of two or more events or conditions, the costs and the absolute risks
of death are additive and the QALYs are multiplicative. Each event or
condition has two sets of cost and QALYs estimates: one for the first
year and the other for the subsequent years up to a customized specific time horizon. In the current version, this time horizon was set to
a lifetime for the cardiovascular events and COPD, to 10 years in lung
cancer, and to 5 years in all other neoplasms. Cancer survival was
modeled as dependant of sex and the years since diagnosis, independent of age. COPD incidence was modeled as dependant of sex, age,
and smoking status; its progression as dependant of sex, years in
current stage, and smoking status. COPD deaths were modeled as
dependant of sex and stage.
Model outputs. Results can be presented according to age, sex,
previous cardiovascular history, and other specific epidemiologic data. Different benefit measures that can be used to report
the cost-effectiveness of the different smoking cessation strategies are cost per quitter; cost per year of life gained, cost per
event averted, and cost per QALY. To reflect decision uncertainty, a graphical depiction of the results of several simulations (second-order uncertainty) each comparing a specific population of interest (first-order uncertainty) can be made. This
can be shown in the cost-effectiveness plane as the incremental
cost-effectiveness rate dispersion, as 95% confidence intervals,
and also as cost-effectiveness acceptability curves to show the
probability of each strategy being cost-effective according to
locally relevant threshold values. To perform a probabilistic
sensitivity analysis, the probability distribution of selected parameters are incorporated into the model. For each iteration,
parameters values are recalculated and applied to the equations. In Figure 1 we present an example of the cost-effectiveness results of two hypothetical SCIs in 500 simulations of 5000
50-year old men.
Different one-way or scenario sensitivity analyses can also be
incorporated for the estimation of the effects of selected parameters. The detailed information provided by the model can also be
used to perform budget impact analysis and to guide for locally
relevant research priority setting [23,24].
Calibration and validation
We applied the International Society for Pharmacoeconomics
and Outcomes Research criteria for model development and
reporting [25]. The model structure and the parameters’ calculation approach were validated and calibrated using a selected
country dataset (Argentinean National Health Statistics year
2005) [26].
Internal validation. Internal testing and debugging were performed to ensure that the mathematical calculations were accurate and consistent with the specifications of the model. The
model was checked and tested during the modeling process to
identify any errors relating to data incorporation and modeling
syntax. Null and extreme input values were used and the test of
S55
VALUE IN HEALTH 14 (2011) S51–S59
(b)
2000
1500
1000
500
0
-0.2
-0.1
0
0.1
0.2
-500
0.3
0.4
probabilty new SCI is cost-effective
Cost difference (Argentine pesos 2007)
(a)
80%
70%
60%
50%
40%
30%
20%
10%
0%
-1000
Effect difference (QALYs)
0K
10K
20K
30K
40K
50K
60K
70K
maximun acceptable ceiling ratio in $'000 Argentine pesos
Fig. 1 – (A) Cost effectiveness plane of cost- and effectrelated differences between two smoking cessation
interventions (SCI). SCI-a: cost $472, effectiveness: 15%
smoking cessation at 1 year; SCI-b: cost $1.254,
effectiveness: 23% smoking cessation at 1 year
(Argentine pesos 2007). Preliminary results for 500
simulated cohorts of 5000 50-year old male smokers
after one quit attempt. Discount rate for costs and
effects: 5%. (B) Cost-effectiveness acceptability curve
showing the probability that SCI-b is cost-effective as
compared to SCI-a over a range of values for the
maximum acceptable ceiling ratio.
replication using equivalent input values was applied. Inconsistencies were detected and programming errors corrected.
Calibration. Calibration was performed to ensure that the model
can reproduce the results of the sources used to run the model.
General mortality and all age- and sex-specific death rates predicted by the model were compared with local health statistics,
with a total of 16 parameters (excluding COPD mortality, universally agreed to be underestimated in national statistics) [21,22].
Sex- and age-specific model outputs were compared to the source
rates and deviations from the expected values were analyzed.
Mean simulated event rates within ⫾ 10% of the mean reference
event rates were considered acceptable, and in cases of higher
deviations, the risk equation for that particular event was modified to provide a better fit to the published data. The search
stopped as soon as all the outputs were within 10% of the target
results. As explained earlier, the disease event incidence was estimated from the age- and sex- specific mortality and the lethality
of the event for acute conditions and COPD, and the yearly survival
rate since diagnosis for cancers. These last two parameters (lethality and survival rate) were estimated from local and international
studies and were allowed to vary ⫾ 15% to determine the best
fitting parameter set. Table 2 describes the data sources, baseline
risk parameters calculation process, and the allowed variation
during the calibration process. Besides ensuring that the simulated results were within the prespecified range, the total number
of events and the event incidence were graphed for each parameter according to age and sex. The resulting observed and expected
curves were visually explored to confirm a good fit. Closeness of fit
was additionally assessed by plotting predicted versus observed
values outcomes, fitting a linear curve through the points with the
intercept set at zero, and obtaining a squared linear correlation
coefficient (R2). The final simulation set was composed of 20 cohorts (10 of men and 10 of women) of 25,000 continuing smokers,
25,000 smokers who quit smoking during follow-up and 25,000 lifelong nonsmokers followed through their lifetime. The sample size of
the simulations was estimated on the basis of the standard error of
the parameter with greater variability (lifelong risk of death from oral
cavity cancer) to be able to obtain 95% confidence intervals within
10% in each cohort (smokers, former smokers, and nonsmokers). Incidence rates estimated from the simulated cohorts were transformed into absolute numbers of events in each age and sex strata
following Argentina 2005 population distribution [26]. After calibration, the differences between published data and the model results
ranged from -9% to ⫹5%. The curves’ shapes of the age- and sexspecific simulated number of events adequately overlapped with
those of the expected values, showing, in all cases, an excellent internal validity. Figure 2 shows results in four conditions: myocardial infarction, kidney cancer, nonischemic cardiovascular disease, and oral cavity cancer. As expected, correlation between
predicted and observed results was better among high incidence events (such as myocardial infarction, stroke, or lung
cancer) and weaker for less frequent (thus with greater variability) events such as leukemia or oral cavity cancer. When predicted values were plotted against observed data to assess
goodness of fit, the majority of values were on or close to the y ⫽
x line, indicative of perfect fit. Evaluation of the correlation between predicted and observed data produced R2 values that
ranged from 0.758 to 0.999 (perfect fit ⫽ 1) indicating a very
strong correlation. The regression lines obtained for the 16 parameters ranged from a gradient of 0.874 to 1.272, close to the
Table 2 – Baseline annual risk in nonsmokers: data sources, calculation process and allowed variation during the
calibration process.
Parameter
Calculation process
Data inputs
Baseline event incidence
in nonsmokers – acute
conditions (stroke,
cardiac events,
Pneumonia/Influenza)
Baseline cancer incidence
in nonsmokers
Formula 1 and Formula 2
Baseline risk of COPD
incidence
Formula 2
– Age- and sex-specific mortality
– Lethality
– Smoking prevalence
– Disease specific RR for smokers and former
smokers
– Age- and sex-specific mortality
– Survival rate
– Smoking prevalence
– Disease specific RR for smokers and former
smokers
– Population COPD incidence (age and sex
specific)
– Smoking prevalence
– Disease specific RR for smokers and former
smokers
Age- and sex-specific mortality
Formula 3 and Formula 2
Risk of death from all
other causes
COPD, chronic obstructive pulmonary disease.
Source
Allowed
variation
[26], [36]
[37], [38]
[33]
[39]
None
⫾ 15%
None
None
[26]
[16], [17]
[33]
[39]
None
⫾ 15%
None
None
[21], [22]
⫾ 15%
[33]
[39]
None
None
[26]
⫾ 15%
S56
VALUE IN HEALTH 14 (2011) S51–S59
Fig. 2 – Calibration: Annual number of deaths predicted by the model in Argentina for each age strata compared to the 2005
Argentinean national health data [26] in four selected conditions: (A) myocardial infarction (women); (B) kidney cancer
(women); (C) non-ischemic cardiovascular disease (men) and; (D) oral cavity cancer (men).
perfect fit line (gradient ⫽ 1). These results are shown in Figure
3 for stroke, lung cancer, pancreatic cancer, and esophageal
cancer.
External validation. Model results were validated against selected
published epidemiologic and clinical studies not used to provide input data. Stroke and myocardial infarction age- and sex- specific
event rates predicted by model were compared with those from international and local available data: the World Health Organization
MONICA Project data on stroke and myocardial infarction incidence
(World Health Organization monitoring of trends and determinants
in cardiovascular disease) [27,28] and the only population-based
myocardial infarction incidence study performed in Argentina
[29,30] (see Fig. 4A and Fig. 4C). Age- and sex-specific COPD predicted
prevalence was compared with the results from the Latin American
Project for the Investigation of Obstructive Lung Disease, a population-based study carried out in five Latin American cities [31] (see Fig.
4B). Lung cancer incidence and lung cancer mortality predicted by
the model were compared to those of the global cancer estimations
reported by the International Agency for Research on Cancer [16,17]
(see Figs. 4D and 4E). As a final endpoint that is influenced by the
others, the toll on life expectancy of smokers and former smokers
predicted by the model was analyzed together with the effect of quitting smoking at age 55 years. Results were compared to the population-based study of male British doctors [32] (see Fig. 4F). In all cases
a high correlation between published data and model results was
observed.
Discussion
Smoking is the single most preventable cause of disease and death all
around the world [1]. In Latin America tobacco control policies are
still poor and access to SCI is very limited [33-35]. In a context of more
limited resources local evidence from cost-effectiveness studies is
essential to implement more efficient health policies. Although international evidence regarding the burden of tobacco-related diseases is extensive, it is widely known that the results of health economic evaluations cannot be directly transferred from one setting to
another. Recently, countries such as the United Kingdom have
changed tobacco intervention policies using cost-effectiveness data,
suggesting that the presence of regional, accurate information in
Latin America could lead to an increased availability of effective interventions and a better definition of local research priorities.
Our study describes the development and validation of a HEE
model to evaluate the disease burden associated with smoking
and the cost-effectiveness of SCIs in Latin America. To ensure
the local relevance of this model, decision makers and researchers from each participant country provided input from the beginning of the project. The main characteristics of the HEE were
S57
VALUE IN HEALTH 14 (2011) S51–S59
(b)
250
500
450
400
350
300
250
200
150
100
50
0
Model predicted values
Model predicted values
(a)
Stroke deaths
y =0.9858x R² =0.9959
200
150
Lung cancer deaths
y =0.9677x R² =0.9923
100
50
0
0
100
200
300
400
500
0
50
Argentinean vital statistics data
(c)
150
200
250
(d)
60
80
70
60
50
40
30
20
10
Pancreatic cancer deaths
y =0.9734x R² =0.7709
Model predicted values
90
Model predicted values
100
Argentinean vital statistics data
50
40
30
Esophagus cancer deaths
y =1.0543x R² =0.8964
20
10
0
0
0
10
20
30
40
50
60
70
80
Argentinean vital statistics data
0
10
20
30
40
50
Argentinean vital statistics data
Fig. 3 – Correlation plot of model predicted versus reported age specific deaths in four selected conditions: (A) stroke
(women); (B) lung cancer (men); (C) pancreatic cancer (women); (D) esophageal cancer (men). The gradients of regression
lines (y) and the correlation coefficients (R2) are reported in each graph. Reference population: Argentinean National Vital
Statistics year 2005 [26].
defined taking into consideration the availability and quality of
the required epidemiologic data in the region. The relevant outcomes of the HEE model were conceived to suit the different
policy makers’ information needs identified in the preparatory
stage of the study. The HEE model showed internal validity with
all simulated event rates falling within ⫾ 10% of the source
publications and it also showed an excellent external validity
when model results were compared to selected published studies. This external validation considered different conditions analyzed from different perspectives: death incidence, disease
prevalence, and survival experience of the simulated cohorts.
This comprehensive validation process is reassuring regarding
the adequate performance of the model. It showed to be a reliable tool that can be used to estimate the tobacco-related disease burden and the cost-effectiveness of different SCI in the
participant countries.
It is important to note that the validation that we are presenting here corresponds to a single country data set (Argentina), and
that a similar validation process is required for each country in
which the model is applied.
This is the first multicountry collaborative project that, to our
knowledge, developed and validated a microsimulation model capable to evaluate the cost-effectiveness of a wide range of SCIs in
Latin America, from public health interventions to specific individual therapies. As opposed to most published tobacco economic
models that are based on state transition structures, the main
advantages of the microsimulation-based approach include the
possibility of tracking each subject’s history (as opposed to the
“memorylessness” of Markov models) and how it influences future
events; the possibility of simultaneously experiencing different
health states during follow-up (which would render a Markov
model nearly unmanageable); and being intrinsically probabilistic,
allowing to depict first order (person-level) uncertainty. Second order
(parameter-level) uncertainty can be estimated in the probabilistic
sensitivity analysis through the incorporation of defined distributions in selected parameters. In addition, it can be easily adapted to
new information availability. Whereas other tobacco-related economic models are usually limited to cardiac disease and lung cancer,
our model included most tobacco-related diseases and additionally
encompassed cerebrovascular disease, COPD, pneumonia, and influenza as well as nine other neoplasms. It also allowed incorporation of
background quitting and relapsing rates.
Some considerations and possible limitations to our model include the following: 1) it is highly dependent upon local health
statistics, which may, in some settings, misestimate the actual
disease-specific toll; 2) it does not analyze the effect of secondhand smoking; and 3) it does not incorporate the effect of tobacco
on mother and child health.
This project was supported by grants from international
agencies, allowing us to develop a generic tool not oriented to a
particular SCI. In addition, because this model was designed
taking into consideration the low availability and quality of information in the seven participant countries, it could be easily
adapted to be used in other “information-poor” settings such as
most low and middle income countries.
Apart from being able to depict the incremental cost and effect of
interventions for the economic evaluation, the model was conceived
as a tool to provide burden of tobacco-related diseases and budget
impact data. This is of utmost relevance to raise awareness of the
health and economic consequences of smoking in the region. The
need to more precisely estimate the burden of the smoking-associated diseases measured in terms of economics and clinical consequences (smoking related illnesses and quality-adjusted survival)
still exists. Also, it is highly important to incorporate these tools in
Latin America to inform decision makers about the cost-effectiveness of tobacco control policies.
S58
VALUE IN HEALTH 14 (2011) S51–S59
(b)
Annual MI attack rate per 10.000
250
70%
60%
200
Model
results
150
Danish
MONICA
100
COPD prevalence
(a)
Buenos
Aires
study
50
50%
Model
results
40%
30%
PLATINO
study
20%
10%
0%
0
35-39
45-49
55-59
65-69
35-39
75-79
45-49
55-59
700
(d)
Model
results
600
500
Finland
MONICA
400
Russia
MONICA
300
200
Lithuania
MONICA
100
Lung Cancer incidence per 100,000
Annual stroke attack rate p/100.000
Age group
(c)
0
45-54
Age group
≥ 85
300
250
Model
results
200
Argentina
150
World
100
South
America
50
55-64
15-44
300
(f)
Model
results
250
200
Argentina
150
World
100
South
America
50
45-54
55-64
Age group
≥ 65
100
Percentage survival from age 35
Lung Cancer mortality per 100,000
75-79
0
35-44
(e)
65-69
Age group
89 (89)
80
(81) 82
60
74 (76)
Non-smokers
(58) 57
46 (50)
Smokers
40
Stopped
smoking at
age 55
(26) 26
20
0
0
15-44
45-54
55-64
≥ 65
40
50
60
70
Age (years)
80
90
100
Age group
Fig. 4 – External validation against selected published epidemiologic studies. Results correspond to the male
population. (A) Annual myocardial infarction event rates predicted by the model compared to population-based
incidence studies: Danish WHO MONICA study register [27] and the myocardial infarction incidence study conducted
in Argentina (Coronel Suarez-Province of Buenos Aires) [29,30]; (B) Model predicted COPD prevalence compared to a
population-based prevalence study performed in Latin America (PLATINO Latin American Project for the Investigation
of Obstructive Lung Disease) [31]; (C) Model predicted annual stroke event rates compared to WHO MONICA study
register in selected countries with high stroke incidence rates (Finland WHO MONICA study register North Karelia
province, Russia WHO MONICA study register Novosibirsk city, Lithuania WHO MONICA study register Kaunas city)
[28]; (D) Model predicted lung cancer incidence compared to the International Agency for Research on Cancer
estimations for Argentina, South America, and the world [16,17]; (E) model predicted lung cancer mortality compared
to the International Agency for Research on Cancer estimations for Argentina, South America, and the world [16,17];
(F) Survival from age 35 for continuing smokers, lifelong non-smokers and effects on survival of stopping smoking at
age 55 years. Model results for simulated cohorts of 250,000 subjects. In parenthesis, at age 60, 70, and 80 years model
results are compared to the cohorts of male British doctors [32].
Sources of financial support: The study was made possible
by unrestricted research grants awarded by the International
Clinical Epidemiology Network, the Initiative for Cardiovascular
Health Research in the Developing Countries; and the International Development Research Centre, Canadian Tobacco Control Research Initiative, American Cancer Society, Cancer Research UK, Institut National du Cancer France, and Department
for International Development UK.
The funding agreement ensured the authors’ independence in
designing the study, interpreting the data, writing, and publishing
the report.
REFERENCES
[1] Ezzati M, Lopez AD. Estimates of global mortality attributable to
smoking in 2000. Lancet 2003;362:847–52.
[2] Jha P, Chaloupka FJ, Moore J, et al. Tobacco addiction. In: Jamison DT,
Measham AR, Breman JB et al. Disease Control Priorities in Developing
Countries. (2nd ed.). New York: Oxford University Press, 2006.
[3] Jha P, Chaloupka FJ. Curbing the epidemic: governments and the
economics of tobacco control. Washington, DC: The World Bank, 1999.
[4] Convenio Marco de la OMS para el control del tabaco. Organización
Mundial de la Salud [Frame- work Convention on Tobacco Control
VALUE IN HEALTH 14 (2011) S51–S59
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
(FCTC). World Health Organization]. Available from: http://www.
who.int/tobacco/framework/WHO_fctc_spanish.pdf. [Accessed March
9, 2011].
Abdullah AS, Husten CG. Promotion of smoking cessation in
developing countries: a framework for urgent public health
interventions. Thorax 2004;59:623–30.
Sculpher MJ, Claxton K, Drummond MF, McCabe C. Whither trialbased economic evaluation for health care decision making. Health
Econ 2006;15:677– 87.
Hjelmgren J, Berggren F, Andersson F. Health economic guidelines—
similarities, differences and some implications. Value Health 2001;4:
225–50.
Taylor RS, Drummond MF, Salkeld G, Sullivan SD. Inclusion of cost
effectiveness in licensing requirements of new drugs: the fourth
hurdle. BMJ 2004;329;972–5.
Banta D. Health technology assessment in Latin America and the
Caribbean. Int J Technol Assess Health Care 2009,25(Suppl. 1):253– 4.
Knusel L. On the accuracy of statistical distributions in Microsoft Excel
2003. Comput Stat Data Anal 2005;48:445–9.
Keeling K, Pavur R. Numerical accuracy issues in using excel for
simulation studies. Available from: http://portal.acm.org/citation.
cfm?id⫽1162012. [Accessed March 9, 2011].
Silva LC, Ordúñez P, Paz Rodríguez M, Robles S. A tool for assessing
the usefulness of prevalence studies done for surveillance
purposes: the example of hypertension. Rev Panam Salud Publica
2001;10:152– 60.
Moher D, Schulz KF, Altman DG. The CONSORT statement: revised
recommendations for improving the quality of reports of parallelgroup randomised trials. Lancet 2001;357:1191– 4.
Terwee CB, Bot SD, de Boer MR, et al. Quality criteria were proposed
for measurement properties of health status questionnaires. J Clin
Epidemiol 2007;60:34 – 42.
Barendregt JJ, Van Oortmarssen GJ, Vos T, Murray CJ. A generic model
for the assessment of disease epidemiology: the computational basis
of DisMod II. Popul Health Metr 2003;1:4.
Ferley J, Bray F, Pisani P, Parkin DM. GLOBOCAN 2002. Cancer
incidence, mortality and prevalence worldwide. IARC CancerBase No.
5, version 2.0. Available from: http://www.emro.who.int/ncd/cancer_
globocan.htm. [Accessed 10 December 2010].
Parkin DM, Bray F, Ferlay J, Pisani P. Estimating the world cancer
burden: GLOBOCAN 2000. Int J Cancer 2001;94:153– 6.
Lauer JA, Rohrich K, Wirth H, Charette C, et al. PopMod: a longitudinal
population model with two interacting disease states. Cost Eff Resour
Alloc 2003;1:6.
Gail MH, Kessler L, Midthune D, Scoppa S. Two approaches for
estimating disease prevalence from population-based registries of
incidence and total mortality. Biometrics 1999;55:1137– 44.
Pisani P, Bray F, Parkin DM. Estimates of the world-wide prevalence of
cancer for 25 sites in the adult population. Int J Cancer 2002;97:72– 81.
Mannino DM. COPD: epidemiology, prevalence, morbidity and
mortality, and disease heterogeneity. Chest 2002;121(5 Suppl.):
121S– 6S.
Stang P, Lydick E, Silberman C, et al. The prevalence of COPD: using
smoking rates to estimate disease frequency in the general
population. Chest 2000;117(5 Suppl. 2):354S–9S.
Levy DT, Bauer JE, Lee HR. Simulation modeling and tobacco control:
creating more robust public health policies. Am J Public Health 2006;
96:494 – 8.
Orme ME, Hogue SL, Kennedy LM, et al. Development of the health
and economic consequences of smoking interactive model. Tob
Control 2001;10:55– 61.
Weinstein MC, O’Brien B, Hornberger J, et al. Principles of good
practice of decision analytic modeling in health care evaluation:
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
S59
report of the ISPOR Task Force on Good Research Practices-Modeling
Studies. Value Health 2003;6:9 –17.
Dirección de estadísticas e información de salud. Argentina:
Ministerio de Salud de la Nación, 2006. Estadísticas vitales –
Información 2005, Serie 5, No. 29. [Directorate of statistics and health
information. National Ministry of Health, 2006. National Health
Statistics year 2005]
Kirchhoff M, Davidsen M, Bronnum-Hansen H, et al. Incidence of
myocardial infarction in the Danish MONICA population 1982–1991.
Int J Epidimiol 1999;28:211– 8.
Thorvaldsen P, Asplund K, Kuulasmaa K, et al. Stroke incidence, case
fatality, and mortality in the WHO MONICA Project. Stroke 1995;26:
361–7.
Caccavo A, Alvarez A, Bello F, et al. Incidencia poblacional del infarto
con elevación del segmento ST o bloqueo de rama izquierda a lo largo
de 11 años en una comunidad de la provincia de Buenos Aires. [Eleven
Years Incidence of Infarction with ST Elevation or Left Bundle Branch
Block on the Population of a Community in the Province of Buenos
Aires] Rev Argent Cardiol 2007;75:185– 8.
Ferrante D, Tajer C. ¿Cuantos infartos hay en la Argentina? [How
many heart attacks there in Argentina?] Rev Argent Cardiol 2007;75:
161–2.
Menezes AM, Perez-Padilla R, Jardim JR, et al. Chronic obstructive
pulmonary disease in five Latin American cities (the PLATINO study):
a prevalence study. Lancet 2005;366:1875– 81.
Doll R, Peto R, Wheatley K, Gray R, Sutherland I. Mortality in relation
to smoking: 40 years’ observations on male British doctors. BMJ 1994;
309:901–11.
Programa VIGIA Argentina. Encuesta de tabaquismo de las 5 grandes
ciudades de la Argentina. Buenos Aires, Argentina: Ministerio de Salud
y Ambiente de la Nación, 2004. [VIGIA program Argentina. Survey of
smoking in the big 5 cities of Argentina. Ministry of Health, 2004]
Noble M, Pérez-Stable E, Casal E. El comportamiento médico en
relación al tabaquismo. Boletín de la Acad Nac de Medicina 1996
[Physician behavior in relation to smoking. Bulletin of the National
Medicine Academy 1996];74:413–25.
Schoj V. Relevamiento regional de cobertura de tratamientos de la
cesación tabáquica. Buenos Aires, Argentina: Ministerio de Salud y
Ambiente de la Nación, 2004. [Regional survey on coverage of smoking
cessation treatments. Buenos Aires, Argentina: National Ministry of
Health, 2004]
Sistema de Información Sanitaria del SNS. Estadísticas Sanitarias
España 1991-2000: 5.8.6. Neumonía e Influenza. Número de
defunciones por edad y sexo. [Health Information System of the SNS.
Spain Health Statistics 1991–2000: 5.8.6. Pneumonia and influenza.
Number of deaths by age and sex] Available from: http://www.msc.es/
estadEstudios/estadisticas/inforRecopilaciones/generales.htm.
[Accessed April 28, 2008].
Egresos de establecimientos oficiales según variables seleccionadas.
República Argentina - Año 2000. Buenos Aires, Argentina: Secretaría
de Políticas, Regulación y Relaciones Sanitarias. Ministerio de Salud,
2003. [Discharges from public institutions according to selected
variables. Argentina - Year 2000. Buenos Aires, Argentina: Secretariat
of Policies, Regulations and Sanitation. Ministry of Health 2003]
Gutierrez F, Masia M, Mirete C, et al. The influence of age and gender
on the population-based incidence of community-acquired
pneumonia caused by different microbial pathogens. J Infect 2006;53:
166 –74.
Centers for Disease Control and Prevention. Smoking-attributable
mortality, morbidity, and economic costs (SAMMEC). Adult SAMMEC,
Relative Risk - CPS–II (82-88). Available from: http://apps.nccd.cdc.gov/
sammec/. [Accessed April 28, 2008].