Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
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].