Smoking Cessation

820735761.pdf

Home >Literature homework help >Smoking Cessation

Economic Evaluation of Smoking-Cessation Therapies A Critical and Systematic Review of Simulation Models

Kristian Bolin

Department of Economics, Lund University, Lund, Sweden

Abstract Background: Smoking is probably the most important among preventable health risks. Health economic evaluation of smoking-cessation interventions,

applying a lifetime perspective, is made possible by available epidemiological

knowledge. The well established method of performing cost-effectiveness analyses

of smoking-cessation interventions involves mathematical modelling (both deter-

ministic and stochastic) of future events important for cost effectiveness.

Objectives: This study surveys cost-effectiveness analyses of smoking cessa-

tion, with a particular focus on the mathematical modelling and simulation

analyses performed.

Data Sources: A systematic literature search was performed using the data-

bases MEDLINE, Econlit and Academic Search Complete.

Study Selection: Health economic evaluations, published as full-length jour-

nal articles, were searched for.

Results: 423 studies were identified and 78 were finally included, of which 30

were assessed as being highly relevant, based on the application of simulation

modelling.

Conclusions: In general, studies are well performed as regards modelling.

Common weaknesses include reporting of modelling details; validation of

used simulation models; and the handling of structural uncertainty and dif-

ferent types of heterogeneity.

Key points for decision makers

� Smoking-cessation interventions are likely to produce substantial gains in QALYs at a relatively low cost

� Smoking-cessation interventions are likely to reduce future healthcare costs

� The gains from smoking cessation are likely to be significantly larger than most studies sug- gest, since effects on productivity (sickness absenteeism and premature mortality) have not been modelled in most studies

Health risks associated with tobacco smoking are substantial, which has been known for a long time.[1-8] Adverse health effects may even occur for non-smokers – passive exposure increases the risk of lung cancer, heart disease and respiratory illness.[9] In fact, smoking is likely to be the single most important preventable health risk. Estimates suggest that smoking caused about 5 million pre- mature deaths per year worldwide at the beginning of the 21st century, and within the next 10 years, the estimated annual number of premature deaths due to smoking will be approximately 9million.[10]

Thus, in terms of economic costs, considerable healthcare utilization can be attributed to smok- ing, and all premature deaths incur a loss of pro- ductivity (although, due to lack of data, the value of non-market production among retirees is sel- dom included in evaluations; exceptions are the three Swedish studies reported in table S1 in the Supplemental Digital Content [SDC], http://links. adisonline.com/PCZ/A138).

These large losses caused by smoking provide powerful incentives for tobacco policy makers to launch initiatives trying to lower smoking preva- lence rates. In several countries, public policy makers have used tools, such as informational campaigns about the adverse health effects of smok- ing, and legislation, such as banning smoking in specific places, for years. It has been argued in the tobacco literature, though, that the new smoking- cessation medical technologies, involving both healthcare efforts and pharmaceutical utilization, are underutilized and, furthermore, that these types of intervention are an important part of a comprehensive tobacco policy aiming at lower smoking prevalence rates.[11,12]

Value for money is one important piece of in- formation, both for individuals when deciding to spend part of a scarce budget on smoking-cessation therapy and for legislators when deciding about the allocation of public funds. Studies of the cost effectiveness of smoking-cessation interventions have been performed and published for decades. The validity of the earlier studies for current in- dividual decisions and public policy making is likely to be limited, however, since time has brought changes that influence cost effectiveness. More- over, the current state-of-the-art methodology

applied when assessing cost effectiveness includes mathematical modelling using simulation tech- niques, drawing from the considerable improve- ment in epidemiological knowledge achieved during the last decade.[13] In particular, several models have been developed that facilitate modelling of the dynamic features of smoking and smoking cessation on health. The effects of smoking on health are truly dynamic because current smok- ing influences future health risks and, similarly, a smoking cessation today will cause smoking- related health risks to tail off gradually. Thus, in order to gauge total effects of smoking cessation, a lifetime perspective is necessary, and a variety of different costs and effects have to be taken into account. In practice, the methods applied to this end vary between different simulation models. Thus, when assessing the cost effectiveness of smoking-cessation interventions, it does not suf- fice to evaluate the quality of the data used for the calculations; how and what data have been used must also be scrutinized. In other words, an as- sessment of the modelling used is at least as im- portant as assessing the quality of data.

1. Methodology Framework

Available smoking-cessation therapies have in general been found to be cost effective at mod- erate levels of willingness to pay for an additional life-year or QALY.[14,15] For some settings, these interventions have even been found both more ef- fective and cost saving with relevant alternatives. No systematic review of cost effectiveness in smok- ing cessation that focuses on modelling has been published. Thus, this study will review the scientific, peer-reviewed literature regarding the economic value of smoking-cessation therapies, focusing on modelling issues. The majority of studies identi- fied and included in this review apply Markov-type simulation models, and most are cohort-based, dynamic (not necessarily as regards transition prob- abilities), time-discrete models. However, more variation is demonstrated when it comes to the specification and parameterization of a given model, in order to analyse the cost effectiveness of an in- tervention for a specific setting. Differing health- sector cost structures between different countries

552 Bolin

http://links.adisonline.com/PCZ/A138

necessitates that (new) health interventions are evaluated for each setting. Consequently, a num- ber of different specifications of one and the same model exist. In addition, health economic evalua- tions that are locally valid are increasingly a man- datory piece of information when decisions are made about reimbursement of pharmaceuticals.

Before presenting the strategy for the literature search and the structure of the rest of the paper, I will provide a brief account of mathematical mod- elling and simulation models (a thorough tax- onomy of simulation models used in health eco- nomic evaluation is provided by Brennan et al.[16]).

1.1 Mathematical Modelling and Simulation – Why?

Mathematical simulation is used for studying the properties of a real system by means of a theoretical model of that system. The properties of the theoretical model define the set of simula- tion models that are possible to use in order to simulate the theoretical model. Let us illustrate this by means of a model familiar to all econo- mists: the economic model of consumer beha- viour. The simplest version of this model is static (does not include time) and assumes perfect knowl- edge of all relevant parameters. A slightly more complicated model of consumer behaviour re- sults from introducing an additional time period, in which case a model that is dynamic and time discrete was obtained. Further generalizations of the theoretical model of consumer behaviour could include, first, several and, then, indefinitely many time periods. The latter case means that the model is formulated in continuous time. In all of these versions of the consumer model, it is poss- ible to take various degrees of imperfect infor- mation into account.

More specifically, a theoretical model (implicitly) underlying any attempt at simulating lifetime effects of a smoking-cessation intervention in- corporates two fundamental processes: first, the effect of the particular therapy on smoking ab- stinence, and second, the time-dependent effects of smoking abstinence on health risks and risk of relapse to smoking. Although there are several ways in which a model could be constructed in

order to simulate these processes, any construction needs to be parameterized as regards (i) treatment effects and (ii) morbidity and mortality risks ac- cording to smoking status and how these evolve over time. These pieces of information will make it possible to project, for instance, life-years in two different scenarios. A full health economic eval- uation demands additional information about costs, and health-related quality of life (HR-QOL), if the effects of smoking cessation are preferred in terms of utility.

Following a more or less established structure, mathematical models used for simulating a process can be classified according to three fundamental dimensions: (i) static versus dynamic simulation model – static and dynamic alluding to whether the variables of interest change over time (dynamic) or not (static); (ii) deterministic versus stochastic – deterministic and stochastic alluding to whether the values of the variables of interest can be per- fectly predicted at every point in time (determi- nistic) or not (stochastic); and (iii) continuous versus discrete – continuous and discrete alluding to whether the variables of interest change only at a countable number of points in time (discrete) or not (continuous). Most simulation models employed in health economic evaluation are dy- namic and time discrete, and often display both stochastic and deterministic features. In addition to these dimensions, mathematical models used for evaluating health technologies are often clas- sified according to the properties of the principal object of study – a cohort (aggregate level) or an individual – and whether or not interaction be- tween individuals is allowed.[16]

2. Outline

First, the systematic literature search is de- scribed. Second, a summary account of encountered mathematical simulation models is provided. Third, each included highly relevant study (see section 3.4) was summarized focusing on simulation-modelling issues, in particular model parameterization. Fourth, the assessment of included studies, ac- cording to published guidelines for modelling in health technology appraisal, is presented.[17] The studies are presented chronologically (by date

Economic Evaluation of Smoking-Cessation Therapies 553

published according to the literature search), be- ginning with the most recent work. The reason for this is that a clear development of applied methods has occurred over the last two or three decades. The review is concluded with a discus- sion of findings, focusing on weaknesses revealed by the assessment, and possible and warranted improvements.

3. Literature Review

3.1 Search Strategy and Inclusion Criteria

A systematic literature search was performed along the following principles: (i) the databases MEDLINE, Econlit and Academic Search Com- plete were searched for full-length articles in English, published in peer-reviewed journals, using the search phrases (a) ‘cost effectiveness’ and ‘smok- ing cessation’ and/or ‘tobacco’, and (b) ‘cost uti- lity’ and ‘smoking cessation’ and/or ‘tobacco’, to appear anywhere in the document; (ii) Health Technology Assessment and the Cochrane Colla- boration of Systematic Reviews were searched for systematic reviews of relevant studies; and (iii) published reviews of cost effectiveness of smoking-cessation interventions were scanned for additional studies published in journal arti- cles. Thus, 423 journal articles and five systematic reviews[14,15,18-20] were identified. The search of the literature lists of the systematic reviews did not locate any additional study that was assess- ed as relevant. In total, 78 articles were initially identified.

The content of each initially indentified study was further examined by reading the article ab- stract; when no abstract was available, the main text was examined. Studies were then finally in- cluded according to the following criteria: the study should treat smoking cessation exclusively and, more specifically, (i) perform a health economic evaluation of a smoking-cessation intervention or an intervention influencing smoking indirectly (for in- stance, reimbursement of smoking-cessation ther- apy); and (ii) use simulation modelling.

A further sorting of studies was performed ac- cording to relevance (high–low). High relevance was given to studies applying intertemporal model-

ling using an appropriate time horizon. Since smoking cessation affects health risks, the appro- priate time horizon is lifetime, which was taken to mean 20 years or more. Low relevance was given to studies that (i) did not fulfil this criterion and (ii) relied on previous studies in order to es- timate life-years saved due to smoking cessation. In addition, since the focus was on modelling, it seemed appropriate to give low relevance to studies performed before a specific date – studies per- formed before 1995 were thus given low relevance. Thirty studies were considered highly relevant (see section 3.4) and were comprehensively summarized in table S1 in the SDC, and assessed for following good practice guidelines for modelling in health technology assessments (see table S2 in the SDC). Studies that were assessed as having low relevance are summarized briefly in table S3 (provided in the SDC). It should be noted that the low-relevance articles in general were high-quality studies and, hence, may provide policy-relevant information.

Figure 1 outlines how the final set of high- and low-relevance studies was obtained.

3.2 Search Results and Extraction of Information

The following information was extracted from each included study: (i) type of intervention, set- ting and perspective; (ii) description of the mod- elling used, model availability and assumptions made in stochastic sensitivity analyses; (iii) sour- ces of core input data: epidemiological data, treat- ment effects, relative risks and QALY weights; (iv) cost-data information, type of costs and how these are measured; and (iv) assumptions made in stochastic analyses.

3.3 Summary Descriptions of Applied Mathematical Simulation Models

Several different simulation models have been applied in the cost-effectiveness analyses included in this review. Most of these models are closely related and, fundamentally, function by applying exposure-specific mortality risks and etiological fractions to epidemiological data regarding mor- tality and the incidence of specific diseases to a state-transition Markov dynamic structure. The

554 Bolin

Benefits of Smoking Cessation on Outcomes (BENESCO) model is a state-transition Markov model and the most frequently applied simula- tion model in published smoking-cessation cost- effectiveness studies. The first study was published in 2008, and to date ten studies that performed calculations by means of the BENESCO model have been published in international peer-reviewed journals (all of these were included in this review as high-relevance studies). A detailed account of the BENESCO model, and less detailed descriptions of other encountered models, are provided below.

The structure and functioning of the BENE- SCO model rests on the same principles as those behind the Health Economic Consequences of Smoking (HECOS) simulation model, which was prepared for and reviewed by the World Health Organization European Partnership Project to Reduce Tobacco Dependence (for a description of the HECOS model, see Orme et al.[21]). However, there is one difference in the basic functioning: while both models apply (the same) published relative risks of dying from smoking,[13] the BENESCO model simulates total morbidity and mortality associated with the included morbidities, and es- timates the relative effects on morbidity and mor- tality of smoking cessation, utilizing a particular therapy, as the difference between morbidity and mortality when that therapy is used and morbidity and mortality when a competing therapy is used. In

contrast, the HECOS model calculates smoking- attributable morbidity and mortality directly.

The BENESCO model simulates the lifetime development of morbidity and mortality for the population at hand, in the age span 18-100 years. The simulations utilize absolute risks of developing and/or dying from each of the considered diseases that were calculated for smokers and former smokers. The model distinguishes between three age groups (cohorts): (i) 18–34 years, (ii) 35–64 years and (iii) 65 years and older; as well as between three health conditions: (i) no morbidity, (ii) mor- bidity and (iii) dead. Morbidity and mortality risks were calculated and updated each cycle for the different age and health states and according to smoking status: (i) current smoker (those at- tempting to quit face the same morbidity and mortality risks as smokers); (ii) recent quitter – stopped smoking between 1 and 5 years ago; and (iii) long-term quitter – had been abstinent for at least 6 years. Recent quitters face decreasing morbidity and mortality risks, approaching that of long-term quitters, who face the same risks as never-smokers.

The model distinguishes between five diseases, using seven different states (chronic obstructive pulmonary disease [COPD] and coronary heart disease [CHD] are each modelled using two se- parate states distinguishing between first-time events and recurrence): (i) asthma; (ii) COPD; (iii) CHD;

Initial search using the phrases ‘smoking cessation’ and ‘cost effectiveness’; and ‘smoking cessation’ and ‘cost utility’; and ‘cost effectiveness’ and ‘tobacco’ (n = 423)

Studies that did not meet the inclusion criteria (n = 345)

Studies that did meet the inclusion criteria (n = 78)

Duplicates (n = 0)

Studies that were assessed as highly relevant (n = 30)

Studies that were assessed as having low relevance (n = 48)

Comprehensive description and assessment (tables S1 and S2)1

1 Tables can be found in the Supplemental Digital Content, http://links.adisonline.com/PCZ/A138.

Listing of reasons for assessment (table S3)

Fig. 1. Description of the literature search and assessment process.

Economic Evaluation of Smoking-Cessation Therapies 555

(iv) stroke; and (v) lung cancer. A number of health conditions seem to have firm associations with smoking according to present epidemiolo- gical knowledge. Even though there are other smoking-related diseases, these conditions cover most of the health problems associated with smok- ing.[22] Furthermore, the model allows for the modeller to distinguish between acute and recur- ring COPD and CHD events, to take differences in cost between the two types of events into ac- count, and to predict mortality conditional on previous COPD and CHD events.

The BENESCO model also incorporates the dynamic relationship between time since quitting and risk of relapse: the model distinguishes be- tween (i) those attempting to quit (the first year); (ii) recent quitters – stopped smoking between 1 and 5 years ago; (iii) medium-term quitters – stopped smoking between 6 and 10 years ago; and (iv) long-term quitters – stopped smoking more than 10 years ago. For those attempting to quit, relapse risk is measured by treatment effective- ness. The BENESCO model employs randomized controlled trial (RCT)-based efficacy values re- garding varenicline versus bupropion, and var- enicline versus nicotine replacement therapy. In addition, the model has been updated to include the efficacy of extended varenicline treatment. Recent quitters are assumed to face a 6% risk of relapse, while medium-term quitters face a risk of 2%. Long-term quitters face a relapse risk of 1%.[23,24]

The model is programmed as a Microsoft�

Excel spreadsheet and comprises a built-in visual- basic facility for performing stochastic sensitivity analysis. Choices of distributions for the parameters considered as stochastic are possible to change (within limits allowed by Excel). Probability dis- tributions are specified using the specific input data provided. The BENESCO model was com- missioned by Pfizer and is not freely available. The HECOS model is available through the World Health Organization.

While the BENESCO and HECOS models were developed with the specific purpose of projecting health effects of smoking cessation, two earlier models that have been much applied in this area – the PREVENT[22] model and the Chronic Dis-

ease Model (CDM)[25] – were originally devel- oped for a somewhat broader purpose: to project population health effects of changes in risk ex- posures. Both models use an epidemiological approach when projecting future health effects of smoking cessation (or change in some other risk factor). The PREVENT model is a dynamic population-based simulation model that allows multiple mortality risks to be simulated jointly. It also allows for the user to specify epidemiological specifics of the studied population. Moreover, the model takes into account the dynamics of risk factors, for instance the decline in risk starting at the time of cessation of exposure, and allows for interventions to be user specified with respect to risk factors, and computes morbidity-specific mortality projections with and without the inter- vention. The PREVENT model has been evaluated against corresponding predictions produced by a micro-simulation model.[26] The results show that the discrepancies between predictions are small for most realistic scenarios. Several modifications have been incorporated since it was first introduced, and it is available free of charge (see details on the EpiGear website[27]). The PREVENT model allows for taking morbidity-specific healthcare costs into account and adjusts life-years using disability weights.

The CDM (Markov model) is also a model constructed for projecting public health effects of interventions that influence risk exposure levels. As in the PREVENT model, the CDM incorpo- rates facilities for simulating the effects of risk exposure changes on morbidity-specific health- care costs, life-years and the quality of those years. The CDM is not freely available. The model was recently extended in order to facilitate simula- tion, which involves multiple states in an ordin- ary Markov model.[28]

Two other models that were encountered among the included studies are the Quit Benefits Model (QBM)[29] and the Tobacco Policy Model.[30,31]

The QBM was developed in order to calculate health and economic benefits of smoking cessa- tion. The model projects avoided cases of acute myocardial infarction, COPD, lung cancer and stroke, and was initially developed and para- meterized for an Australian setting, adopting the

556 Bolin

requirements for submission to the Australian federal government for subsidization of medicines. The model was developed so that it can be adapted to evaluation of smoking-cessation interventions in other settings than Australia. The QBM is not freely available. The Tobacco Policy Model was constructed with the purpose of calculating costs and health gains associated with any particular intervention or policy aiming at reduced smoking. The model projects births, deaths and smoking status for a US population. The Tobacco Policy Model is not freely available.

In addition to the models mentioned in this section, a number of less widely applied simula- tion models were utilized (see study descriptions in table S1 in the SDC). In what follows, I will assume that the models explicitly mentioned in this section have been appropriately validated.

3.4 Description and Comparison of Included Highly Relevant Studies

In this section, the included highly relevant studies are described and compared. Table S1 in the SDC provides general information for each study (including cost-effectiveness measures), while this section compares more detailed modelling issues, using a division of modelling character- istics similar to the one found in the modelling guidelines referred to in section 2 (see table S2 in the SDC).

The studies are compared with respect to (i) structure – (a) type of model and time horizon, (b) morbidities modelled, assumptions regarding morbidity-risk structures and cost components included, (c) modelling of HR-QOL and (d) model- ling of quit attempts and relapse; and (ii) data – type and source of (a) smoking epidemiology data, (b) relative morbidity and mortality risks, (c) treat- ment effectiveness, (d) QOL weights, (e) unit costs and (f) the handling of uncertainty.

For convenience, and since the BENESCO model is currently the most widely applied simula- tion model in published cost-effectiveness studies of smoking-cessation interventions, the compar- ison of modelling aspects between the different studies is done with the BENESCO model as a point of reference.

3.4.1 Structure

The majority (25) of the assessed studies em- ployed Markov simulation models that per- formed calculations for representative cohorts. The BENESCO model follows three separate cohorts (see section 3.3). The youngest cohort (18–34 years) successively enters into the middle cohort, in which smoking-associated morbidity starts to appear. Most of the Markov model-based studies that used some other simulation model were based on the same principle, although the exact definitions of age cohort may vary.

Of the 25 Markov model-based studies, ten employed Markov simulation models that were not presented, or mentioned, in section 3.3.[32-41]

Of these, seven studies were based on models that were not readily available for validation by the reader.[33,34,36-39,41] Similarly, three of the five studies that were based on non-Markov models utilized models not mentioned in section 3.3 (two were based on the PREVENT model[42,43]). The majority of models adopted a time horizon be- tween 50 years and lifetime; see tables S1 and S2 in the SDC (highly relevant classification demanded at least a 20-year time horizon).

The BENESCO model-based studies that were based upon the original set of morbidities in the model all included four of the five diseases that the model was constructed to handle. Asthma was excluded in four of the studies, due to lack of data. Some studies were based on models that included more morbidities than the BENESCO model. Both the PREVENT model and the CDM incorporate morbidities in addition to the ones included in the BENESCO model (mainly cancer diagnoses).[42-45] The study based on the QBM in- cluded the same diseases as the BENESCO model, except for asthma (the model used in the Hurley et al.[36] study was developed by partly the same authors).[46] The South Korean study was based on a version of the BENESCO model, which had been modified in order to take cancer of the stom- ach and the liver into account.[47] The Japanese study, which was based on a Markov model con- structed by the authors, included several condi- tions in addition to the diseases incorporated in the BENESCO model.[32] In total, 19 separate dis- eases, of which ten were cancer diagnoses, were

Economic Evaluation of Smoking-Cessation Therapies 557

included. Morbidity-related healthcare costs were included in most studies; six studies included in- tervention costs only.[35,39-43] Six studies included indirect effects on productivity.[42,48-52] The three Swedish BENESCO-based studies included in- direct effects on production and consumption values induced by the changes in mortality fol- lowing smoking cessation.[48-50] The calculations were performed using an extended version of the BENESCO model. Two studies included cost in- duced by sickness absenteeism.[51,52] One study considered the cost of patient time.[42]

Furthermore, a few studies simulated the num- ber of successful quitters over time and utilized previously published data in order to directly cal- culate estimates of gains in life-years, or gains in terms of reduced incidence of a particular condi- tion,[36] and intervention costs and effects on healthcare utilization produced by smoking ces- sation.[34-36,38,39,53] The other studies included the same diseases as the BENESCO model-based studies (except for asthma).

Although most of the studies were based on models that calculated QALYs, several studies re- ported cost-effectiveness measures without using QALYs (one study reported cost per disability- adjusted life-year saved[37]).[33-35,38-40,42,43,46,52,53]

The BENESCO model allows for one quit at- tempt, at the outset of the simulation, and takes into account that morbidity risks and the risk of relapse decrease with time. All the BENESCO model-based studies included these characteristics (although different types of data may have been used; see section 3.4.2). All other studies also in- cluded one quit attempt, except the Danish study which allowed for more than one attempt.[38]

Regarding morbidity and relapse risk structures, some studies did not take into account that the excess smoking-related morbidity risks diminish over time,[32] and/or that the risk of relapse also decreases with time.[32]

3.4.2 Data

The BENESCO model requires detailed epi- demiological information pertaining to the ger- mane setting, regarding the prevalence of smokers and former smokers, and the prevalence and in- cidence of and mortality from included morbid-

ities. Furthermore, the model is pre-programmed with relative morbidity risks of included dis- eases,[13] although these may be chosen at differ- ent values by the user. The different BENESCO model studies collected the necessary epidemio- logical information from public registries and/or published studies, and in all cases but one,[54] the data pertained to the studied country (local data). As regards relative risks, all but two studies[47,55]

utilized model default values. The studies based on other models employed local epidemiological data, collected from public registries or previous publications; see table S1 in the SDC for details. While most of these studies employed smoking- related relative morbidity (and mortality) risks that were produced by data from the US Cancer Society Prevention Study II, several studies col- lected and utilized local risk data. This was the case for the Dutch, Japanese, Danish and Scot- tish studies.[32,38,43-45]

Furthermore, the BENESCO model-based studies utilized (i) pre-programmed treatment ef- fectiveness values from head-to-head RCTs, re- garding varenicline versus bupropion, varenicline versus nicotine replacement therapy and 12-week course of varenicline versus 12 + 12-week course of varenicline; and (ii) pre-programmed utility weights, which were obtained from a systematic search of published studies.[56] There were excep- tions, though: the Dutch, Finnish, Belgian, and South Korean studies all used local morbidity- related utility weights.[47,57-59] Various sources of treatment-effectiveness and utility-weight data were used in the studies that were based on other models; see table S1 in the SDC for details. As regards efficacy, a majority of the studies collected infor- mation from international peer-reviewed pub- lications.[34,35,37,40-45,51-53,55,60] Similarly, utility weights were collected from published studies, ex- cept in one case;[60] seven studies employed local utility weights.[44,45,51,55,57-59]

Intervention costs and morbidity-related health- care costs were collected from public or admin- istrative registries, or from previous published studies, and were local in all but one case.[46]

Finally, different aspects of uncertainty were con- sidered in all but one study.[43] The included studies span the years 1996–2010. A clear development as

558 Bolin

regards practice can be observed during this period. The more recent studies are generally more com- prehensive and systematic as regards sensitivity analysis, often employing models with built-in facilities for stochastic sensitivity analysis.

3.5 Assessment of Included Studies That Were Assessed as Highly Relevant

The detailed assessment was done according to Philips et al. guidelines for good modelling prac- tice[17] – each study was assessed in relation to each item listed in table II of that publication. The results are reported in table S2 in the SDC. The Philips et al. guidelines are divided into three sections: Structure, Data and Consistency. Each section comprises a number of detailed issues that should be checked against the content in the as- sessed study. As regards the included studies in this review, the Structure section did not evoke many did-not-meet-requirement assessments (marked by ‘no’; ‘yes’ indicated the opposite). No doubt, this is a consequence of the fact that most models, at least among the recent ones, were developed on a more or less established structure. Most studies, however, had shortcomings in the Data and/or Consistency sections.

Studies that failed to meet all criteria in the Structure section applied simulation models that were insufficiently reported, both as regards the specifics of the process modelled and whether or not input data were consistent with the scope of the model.[33,39,46] In two cases, disease pathways were modelled assuming that patients completely recover after a cycle (1 year)[32] and that 5-year lung cancer survivors completely recover,[51] which are questionable assumptions. In some cases, model specifics such as cycle length, pathways and type of model could not be assessed.[40,42,45,46,53,60,61]

In the Data section, studies suffered from un- availability of input data[47] and treatment effects that were not appropriately supported by em- pirical evidence.[45] The uncertainty part of the Data section evoked at least one did-not-meet- requirement remark for all studies – no study performed analysis of the importance of struc- tural uncertainty. For the other three entries in the checklist, the findings are mixed (see table S2

in the SDC). Several studies considered method- ological uncertainty by performing calculations for various discount rates and time horizons. Het- erogeneity was also considered in several cases by performing calculations separately for men, women and age groups. In one case, differences between geographical regions were considered.[42] The im- portance of parameter uncertainty was considered by most studies, both by performing determinis- tic one-way or two-way sensitivity analyses and by Monte Carlo simulations. In general, these analyses were well performed, although the choice and specification of probability distributions in most cases were done without proper motivation.

Finally, in the Consistency section, several studies used a simulation model that was neither available for performing validity checks nor re- ported as having been previously validated (see table S2 in the SDC).

4. Discussion

Efforts taken in order to mitigate the adverse effects of smoking need to be governed by firm and reliable evidence as regards the costs and benefits that results from smoking cessation. This is where health economic evaluations, using math- ematical modelling and simulation, can be useful. For this to be the case, however, appropriate data must be employed appropriately. This review fo- cused on the application of modelling in published cost-effectiveness studies of smoking-cessation in- terventions. The assessment was conducted using published guidelines for good modelling practice in health economic evaluations.[17] In general, health economic modelling has been used in a way that is consistent with the different criteria of the guidelines. Studies failed to meet the require- ments concerning (i) sufficient reporting of model structures; (ii) sufficient reporting of input data; (iii) appropriate analysis of the consequences of uncertainty; and (iv) validity of the model used. The more recent studies are more aligned with published guidelines concerning how to perform and report health economic evaluations and how to adhere to good practice when using simulation models – no doubt, this owes to the fact that a practice for how modelling should be performed

Economic Evaluation of Smoking-Cessation Therapies 559

has been established. Also, the improved avail- ability of data and software tools for modelling are likely to contribute.

The studied interventions range from drug- based interventions combined with counselling to broad policy measures targeting whole popula- tions. Although applied methods for assessing treatment effects vary within these boundaries, reduced morbidity, mortality and improved QOL are the essential health effects that constitute the gains making smoking cessation worthwhile and are thus the ultimate goal for any interven- tion regardless of its practical shape. It is no co- incidence then that the applied models presented in this review have about the same structure. Even though there are some computational differences and differences as regards, for instance, considered morbidities, all models include the most important cancer diagnoses, cardiovascular diseases and res- piratory diseases. Performed studies suggest that these models produce comparable results – at least judged from the qualitative conclusions that they provide – and, hence, differences between studies seem to be more a question of differences as regard input data and, of course, the exact set of calculations carried out. Most applied models were of Markov type and followed basically the same structure – change in risk exposure influences smoking-related morbidity and mortality, which, in turn, are projected over at least 20 years.

In addition to the assessment of simulation modelling, this review also adds to the stock of evidence regarding cost effectiveness of smoking- cessation interventions. Published evidence con- cerning cost effectiveness of smoking-cessation interventions suggests that, using a reasonable willingness-to-pay threshold, smoking-cessation interventions are among the most cost-effective interventions for health. This conclusion reaches beyond pharmaceutical-based interventions and is further corroborated by the findings in this re- view. Some particular interventions, however, have only been subject to a small number of evalua- tions and cannot, although promising results have been obtained, be classified as cost effective without more evidence. For instance, reimburse- ment of smoking-cessation costs accruing to the individual smoker has been proposed by several

authors, arguing that therapies for smoking cessa- tion are underutilized. The evidence, however, for reimbursement to be cost effective is scarce, at best.

While many of the revealed modelling weak- nesses of included studies are readily amendable by more elaborate reporting of modelling details, there are more intricate issues: 1. The inclusion of indirect effects is a much dis- cussed issue in the health economic literature.[61-64]

Smoking cessation influences morbidity and life expectancy and, hence, will also affect future productivity and consumption. It is by no means self-evident that these effects will work in one way or the other. The inclusion of these effects has the potential of altering conclusions about cost ef- fectiveness. On the one hand, reduced morbidity and mortality will increase productivity among individuals who are active in the labour market, but, on the other hand, also increase healthcare costs and consumption among both working individuals and retirees. None of the simulation models en- countered in this review incorporated indirect effects in their original versions, although two of the models were modified in order to do so.[50,65]

2. None of the included studies addressed the issue of structural or model-related uncertainty. This type of uncertainty could – in principle – have been dealt with in any of the Markov models encountered, for instance, by considering an alternative set of states and alternative transi- tion probabilities,[66] or by constructing a parallel non-Markov model, and comparing results. The proposal to convey the magnitude of structural uncertainty is to perform the analysis repeatedly for different model specifications and then to use a weighted average as the final measure of cost effectiveness. 3. Heterogeneity as regards both treatment ef- fects and future health effects has been taken into account in some studies by performing calcula- tions separately for men, women and different age groups. Most studies, however, did not make any attempts at taking possible additional sources of heterogeneity into account. Several of the para- meters used in lifetime simulations of health effects resulting from smoking cessation are likely to vary between individuals according to more characteristics than sex and age. For instance,

560 Bolin

education is a variable known to influence health- related behaviour and, hence, we should expect education to influence not only treatment effects but also the size of the benefits from cessation that arise over time. Naturally, the lack of data may obstruct a more comprehensive analysis of the importance of heterogeneity. 4. Several studies suffered from a lack of evidence as regards internal and external validity. Since most models were not freely available, the validity of derived results cannot be assessed. It goes with- out saying that this reduces the policy value that can be attached to a particular study. 5. Most encountered models included only a part of those diseases for which there is epide- miological evidence of a relationship to smoking. For instance, the BENESCO model included only one cancer diagnosis (lung cancer). The effects of smoking cessation on the incidence of COPD, CHD, cardiovascular diseases and lung cancer were most frequently modelled. There are, how- ever, several more diagnoses that could have been included. The conservative approach adopted by most models means that the benefits of smoking cessation are underestimated, ceteris paribus. 6. While a number of studies benefited from head- to-head RCTs concerning the efficacy of the treat- ments studied (the BENESCO studies), most studies had to rely on indirect comparisons. Naturally, studies that utilize RCT data produce more (in- ternally) valid results than studies that employ other types of data. 7. None of the studies considered adverse effects induced by the smoking-cessation intervention. At least for the interventions based on pharma- ceuticals, this tends to overestimate the cost ef- fectiveness of the intervention. Even though the adverse health effects of smoking are severe, it may be the case that a smoking cessation not only reduces the risks associated with several diseases but also influences individual behaviour in such a way that the risk of overweight and obesity in- creases. This has neither been included in any of the encountered simulation models, nor discussed in any of the conclusions.

Finally, even though the assessed studies showed several significant similarities, essential differences regarding model structure and data were also re-

vealed. As regards structural modelling issues, potentially important differences were found vis- à-vis time horizon, included morbidities and in- cluded cost components. First, the relationship between adopted time horizon and measured cost effectiveness of a smoking-cessation intervention is ambiguous, since a prolonged time horizon will include both additional life-years and additional costs. In particular, the indirect effects (on pro- duction and consumption) of increased survival due to reduced smoking may be large enough to induce radical changes in measured cost effec- tiveness, as demonstrated by the Swedish stud- ies.[48-50] Moreover, the results in those studies suggest a weak connection between time horizon beyond 20 years and cost effectiveness. Second, total health benefits of smoking cessation is likely to involve risk reductions for a much larger set of morbidities than included in any of the assessed studies. In addition, the effects of passive smoking, which may be substantial, have not been con- sidered in any of the studies. Thus, the benefits of smoking cessation have most likely been under- estimated in all the studies. Furthermore, the effect of including (or excluding) a particular dis- ease from the analysis has to be considered from case to case. The principle, however, is as follows: benefits will increase more, ceteris paribus, when including a disease, (i) the more smoking affects the risk of getting the disease, (ii) the higher the incidence of the disease, and (iii) the more severe the health consequences of the disease. Third, the specification of costs to be included in the ana- lysis – obviously – has the potential of being de- cisive for the results. However, the effect of leaving out healthcare costs depends both on the prevailing healthcare utilization practice for each included disease and on healthcare sector productivity. Thus, no general conclusions can be drawn and, hence, the importance of a cost component has to be assessed from setting to setting.

The reliability of the data encountered in the assessed studies varies. It seems reasonable to think about information reliability as varying between information published in the peer-reviewed in- ternational scientific literature (most reliable) and information collected by the authors themselves, without any corresponding publication (least

Economic Evaluation of Smoking-Cessation Therapies 561

reliable). Generally, then, data concerning treatment effects and risk of relapse into smoking are the most reliable data encountered in this assessment. Furthermore, QOL weights were for the most part collected from peer-reviewed journals and, hence, could be regarded as relatively reliable. The si- tuation is different when it comes to epidemiolo- gical and unit cost data, though. In several studies, this information was collected from unpublished sources. However, epidemiological data were fre- quently collected from public registries, or compar- able sources, facilitating third-party validation.

5. Conclusions

Although reviewed studies have made largely satisfying use of health economic modelling, there is scope for improvement. In particular, improve- ments can be made by certifying that all input data necessary for assessing the policy relevance of a study are readily available, by routinely pro- viding validation calculations produced by the simulation model used (either own calculations or previously published calculations) and by taking heterogeneity into account so far as allowed by available input data. Moreover, the absolute ma- jority of perceived models were of Markov type and performed calculations on a cohort level. Using an individual-level model (often referred to as discrete-event model) would facilitate taking individual heterogeneity into account and also the modelling of processes that naturally involve states with multiple diseases. In a Markov setting, such processes have to be modelled by defining new states for each combination of diseases, which rapidly gets very complex.

Acknowledgements

No sources of funding were used to prepare this article. The author has no conflicts of interest that are directly rel- evant to its content and the opinions expressed are the au- thor’s own.

References 1. Wynder EL, Graham EA. Tobacco smoking as a possible

etiologic factor in bronchiogenic carcinoma: a study of 684 proved cases. JAMA 1950; 143: 329-96

2. Doll R, Hill AB. The mortality of doctors in relation to their smoking habits: a preliminary report. BMJ 1954; 1: 1451-5

3. Doll R, Hill AB. Lung cancer and other causes of death in relation to smoking: a second report on the mortality of British doctors. BMJ 1956; 2: 1071-81

4. Hammond ED, Horn D. Smoking and death rates – report on forty-four months of follow-up on 187,783 men: I. Total mortality. JAMA 1958; 166: 1159-72

5. Hammond ED, Horn D. Smoking and death rates – report on forty-four months of follow-up on 187,783 men: II. Death rates by cause. JAMA 1958; 166: 1294-308

6. Royal College of Physicians. Smoking and health: summary and report of the Royal College of Physicians of London on smoking in relation to cancer of the lung and other diseases. New York: Pitman Publishing, 1962

7. U.S. Department of Health, Education, and Welfare. Smoking and health: report to the Advisory Committee to the Surgeon General of the Public Health Service. Washington, DC: U.S. Government Printing Office, 1964. PHS publication no.: 1103

8. World Health Organization. Tobacco or health: a global status report. Geneva: WHO, 1997

9. Fagerström K. The epidemiology of smoking: health con- sequences and benefits of cessation. Drugs 2002; 62 Suppl. 2: 1-9

10. Shafey O, Dolwick S, Guindon GE, editors. Tobacco con- trol country profiles. 2nd rev. ed. Atlanta (GA): American Cancer Society, 2003

11. Abrams DB, Graham AL, Levy DT, et al. Boosting pop- ulation quits through evidence-based cessation treatment and policy. Am J Prev Med 2010 Mar; 38 (3 Suppl.): S351-63

12. Hughes JR. How confident should we be that smoking ces- sation treatments work? Addiction 2009; 104 (10): 1637-40

13. Thun MJ, Apicella LF, Henley SJ. Smoking vs other risk factors as the cause of smoking-attributable deaths: con- founding in the courtroom. JAMA 2000; 284: 706-12

14. Woolacott NF, Jones L, Forbes CA, et al. The clinical ef- fectiveness and cost-effectiveness of bupropion and nico- tine replacement therapy for smoking cessation: a system- atic review and economic evaluation. Health Technol Assess 2002; 6 (16): 1-245

15. Keating GM, Lyseng-Williamson KA. Varenicline: a phar- macoeconomic review of its use as an aid to smoking ces- sation. Pharmacoeconomics 2010; 28 (3): 231-54

16. Brennan A, Chick SE, Davies R. A taxonomy of model structures for economic evaluation of health technologies. Health Econ 2006; 15 (12): 1295-310

17. Philips Z, Bojke L, Sculpher M, et al. Good practice guide- lines for decision-analytic modelling in health technology assessment: a review and consolidation of quality assess- ment. Pharmacoeconomics 2006; 24 (4): 355-71

18. Elixhauser A. The costs of smoking and the cost effective- ness of smoking-cessation programs. J Public Health Pol- icy 1990; 11 (2): 218-37

19. Song F, Raftery J, Aveyard P, et al. Cost-effectiveness of pharmacological interventions for smoking cessation: a literature review and a decision analytic analysis. Med Decis Making 2002; 22 (5 Suppl.): S26-37

20. Ronckers ET, Groot W, Ament AJ. Systematic review of economic evaluations of smoking cessation: standardiz- ing the cost-effectiveness. Med Decis Making 2005; 25 (4): 437-48

562 Bolin

21. Orme ME, Susan LH, Kennedy LM, et al. Development of the health and economic consequences of smoking inter- active model. Tob Control 2001; 10: 55-6

22. Gunning-Schepers LJ. The health benefits of prevention: a simulation approach. Health Policy 1989; 12: 1-256

23. Wetter D, Cofta-Gunn L, Fouladi R, et al. Late relapse/ sustained abstinence among former smokers: a longitudi- nal study. Prev Med 2004; 39: 1556-63

24. Krall E, Garvey A, Garcia R. Smoking relapse after 2 years of abstinence: findings from the VA Normative Aging Study. Nicotine Tob Res 2002; 4: 95-100

25. Hoogenveen RT, de Hollander AEM, van Genugten MLL. The chronic disease modeling approach. Bilthoven: Na- tional Institute of Public Health and the Environment (RIVM), 1998

26. Brønnum-Hansen H. How good is the Prevent model for estimating the health benefits of prevention? J Epidemiol Community Health 1999; 53 (5): 300-5

27. Prevent [computer program]. Version 3.01. Wilston (QLD): EpiGear International, 2009 [online]. Available from URL: http://www.epigear.com/index_files/prevent.html [Accessed 2011 Feb 5]

28. Hoogenveen RT, van Baal PHM, Boshuizen HC. Chronic disease projections in heterogeneous ageing populations: approximating multi-state models of joint distributions by modeling marginal distributions. Math Med Biol 2010; 27: 1-19

29. Hurley SF, Matthews JP. The Quit Benefits Model: a Mar- kov model for assessing the health benefits and health care cost savings of quitting smoking. Cost Eff Resour Alloc 2007; 5: 2-20

30. Tengs TO, Ahmad S, Savage JM, et al. The AMA proposal to mandate nicotine reduction in cigarettes: a simulation of the population health impacts. Prev Med 2005; 40: 170-80

31. Tengs T, Ahmad S, Moore R, et al. Federal policy man- dating safer cigarettes: a hypothetical simulation of the anticipated population health gains or losses. J Policy Anal Manage 2004; 3: 857-72

32. Igarashi A, Takuma H, Fukuda T, et al. Cost-utility analysis of varenicline, an oral smoking-cessation drug, in Japan. Pharmacoeconomics 2009; 27 (3): 247-61

33. Thavorn K, Chaiyakunapruk N. A cost-effectiveness ana- lysis of a community pharmacist-based smoking cessation programme in Thailand. Tob Control 2008; 17 (3): 177-82

34. Dino G, Horn K, Abdulkadri A, et al. Cost-effectiveness analysis of the Not On Tobacco program for adolescent smoking cessation. Prev Sci 2008; 9 (1): 38-46

35. Welton NJ, Johnstone EC, David SP, et al. A cost- effectiveness analysis of genetic testing of the DRD2 Taq1A polymorphism to aid treatment choice for smoking cessation. Nicotine Tob Res 2008; 10 (1): 231-40

36. Hurley SF, Matthews JP, Guymer RH. Cost-effectiveness of smoking cessation to prevent age-related macular degen- eration. Cost Eff Resour Alloc 2008; 6: 18

37. Bertram MY, Lim SS, Wallace AL, et al. Costs and benefits of smoking cessation aids: making a case for public re- imbursement of nicotine replacement therapy in Australia. Tob Control 2007; 16 (4): 255-60

38. Olsen KR, Bilde L, Juhl HH, et al. Cost-effectiveness of the Danish smoking cessation interventions: subgroup analysis

based on the Danish Smoking Cessation Database. Eur J Health Econ 2006; 7 (4): 255-64

39. Cornuz J, Gilbert A, Pinget C, et al. Cost-effectiveness of pharmacotherapies for nicotine dependence in primary care settings: a multinational comparison. Tob Control 2006; 15 (3): 152-9

40. Cornuz J, Pinget C, Gilbert A, et al. Cost-effectiveness analysis of the first-line therapies for nicotine dependence. Eur J Clin Pharmacol 2003; 59 (3): 201-6

41. Fiscella K, Franks P. Cost-effectiveness of the transdermal nicotine patch as an adjunct to physicians’ smoking cessa- tion counseling. JAMA 1996; 275 (16): 1247-51

42. Parrott S, Godfrey C, Raw M, et al. Guidance for commis- sioners on the cost effectiveness of smoking cessation in- terventions: health educational authority. Thorax 1998; 53 Suppl. 5: 1-38

43. Ratcliffe J, Cairns J, Platt S. Cost effectiveness of a mass media-led anti-smoking campaign in Scotland. Tob Con- trol 1997; 6 (2): 104-10

44. Vemer P, Rutten-van Mölken M, Kaper J, et al. If you try to stop smoking, should we pay for it? The cost-utility of re- imbursing smoking cessation support in the Netherlands. Addiction 2010; 105 (6): 1088-97

45. Feenstra TL, Hamberg-van Reenen HH, Hoogenveen RT, et al. Cost-effectiveness of face-to-face smoking cessation interventions: a dynamic modeling study. Value Health 2005; 8 (3): 178-90

46. Hurley SF, Matthews JP. Cost-effectiveness of the Australian National Tobacco Campaign. Tob Control 2008; 17 (6): 379-84

47. Bae JY, Kim CH, Lee EK. Evaluation of cost-utility of var- enicline compared with existing smoking cessation therapies in South Korea. Value Health 2009; 12 Suppl. 3: S70-3

48. Bolin K, Mörk AC, Wilson K. Smoking-cessation therapy using varenicline: the cost-utility of an additional 12-week course of varenicline for the maintenance of smoking ab- stinence. J Eval Clin Pract 2009; 15 (3): 478-85

49. Bolin K, Mörk AC, Willers S, et al. Varenicline as compared to bupropion in smoking cessation therapy: cost-utility results for Sweden. Respir Med 2008; 102 (5): 699-710

50. Bolin K, Lindgren B, Willers S. The cost utility of bupropion in smoking cessation health programs: simulation model results for Sweden. Chest 2006; 129 (3): 651-60

51. Johansson PM, Tillgren PE, Guldbrandsson KA, et al. A model for cost-effectiveness analyses of smoking cessation interven- tions applied to a Quit-and-Win contest for mothers of small children. Scand J Public Health 2005; 33 (5): 343-52

52. Halpern MT, Khan ZM, Young TL, et al. Economic model of sustained-release bupropion hydrochloride in health plan and work site smoking-cessation programs. Am J Health Syst Pharm 2000; 57 (15): 1421-9

53. Heitjan DF, Asch DA, Ray R, et al. Cost-effectiveness of pharmacogenetic testing to tailor smoking-cessation treat- ment. Pharmacogenomics J 2008; 8 (6): 391-9

54. Bolin K, Wilson K, Benhaddi H, et al. Cost-effectiveness of varenicline compared with nicotine patches for smoking cessation: results from four European countries. Eur J Public Health 2009; 19 (6): 650-4

55. Vemer P, Rutten-van Mölken M. Crossing borders: fac- tors affecting differences in cost-effectiveness of smoking

Economic Evaluation of Smoking-Cessation Therapies 563

http://www.epigear.com/index_files/prevent.html

cessation interventions between European countries. Value Health 2010; 13 (2): 230-41

56. Knight C, Howard P, Baker CL, et al. The cost-effectiveness of an extended course (12+12 weeks) of varenicline compared with other available smoking cessation strategies in the Uni- ted States: an extension and update to the BENESCO Model. Value Health 2010; 13 (2): 209-14

57. Linden K, Jormanainen V, Linna M, et al. Cost effectiveness of varenicline versus bupropion and unaided cessation for smoking cessation in a cohort of Finnish adult smokers. Curr Med Res Opin 2010; 26 (3): 549-60

58. Annemans L, Nackaerts K, Bartsch P, et al. Cost effectiveness of varenicline in Belgium, compared with bupropion, nicotine replacement therapy, brief counselling and unaided smoking cessation. Clin Drug Investig 2009; 29 (10): 655-65

59. Hoogendoorn M, Welsing P, Rutten-van Mölken MP. Cost- effectiveness of varenicline compared with bupropion, NRT, and nortriptyline for smoking cessation in the Netherlands. Curr Med Res Opin 2008; 24 (1): 51-61

60. Tengs TO, Osgood ND, Chen LL. The cost-effectiveness of intensive national school-based anti-tobacco education: results from the Tobacco Policy Model. Prev Med 2001; 33: 558-70

61. Meltzer D. Accounting for future costs in medical cost- effectiveness analysis. J Health Econ 1997; 16: 33-64

62. Lundin D, Ramsberg J. On survival consumption costs: a reply to Nyman. Health Econ 2008; 17 (2) 293-7

63. Nyman JA. Should the consumption of survivors be in- cluded as a cost in cost-utility analysis? Health Econ 2004; 13: 417-27

64. Nyman JA. More on survival consumption costs in cost- utility analysis. Health Econ 2006; 15: 319-22

65. Howard P, Knight C, Boler A, et al. Cost-utility analysis of varenicline versus existing smoking cessation strategies using the BENESCO simulation model: application to a population of US adult smokers. Pharmacoeconomics 2008; 26 (6): 497-511

66. Briggs A, Sculpher M, Claxton K. Decision modelling for health economic evaluation. Oxford: Oxford University Press, 2006

Correspondence: Professor Kristian Bolin, Department of Economics, Lund University, P.O. Box 705, SE-220 07 Lund, Sweden. E-mail: [email protected]

564 Bolin

mailto:[email protected]

copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written

permission. However, users may print, download, or email articles for individual use.