Cost-effectiveness of a high-intensity versus a low-intensity smoking cessation intervention in a dental setting: long-term follow-up

Objectives The aim of this study was to conduct a cost-effectiveness analysis (CEA) of a high-intensity and a low-intensity smoking cessation treatment programme (HIT and LIT) using long-term follow-up effectiveness data and to validate the cost-effectiveness results based on short-term follow-up. Design and outcome measures Intervention effectiveness was estimated in a randomised controlled trial as numbers of abstinent participants after 1 and 5–8 years of follow-up. The economic evaluation was performed from a societal perspective using a Markov model by estimating future disease-related costs (in Euro (€) 2018) and health effects (in quality-adjusted life-years (QALYs)). Programmes were explicitly compared in an incremental analysis, and the results were presented as an incremental cost-effectiveness ratio. Setting The study was conducted in dental clinics in Sweden. Participants 294 smokers aged 19–71 years were included in the study. Interventions Behaviour therapy, coaching and pharmacological advice (HIT) was compared with one counselling session introducing a conventional self-help programme (LIT). Results The more costly HIT led to higher number of 6-month continuous abstinent participants after 1 year and higher number of sustained abstinent participants after 5–8 years, which translates into larger societal costs avoided and health gains than LIT. The incremental cost/QALY of HIT compared with LIT amounted to €918 and €3786 using short-term and long-term effectiveness, respectively, which is considered very cost-effective in Sweden. Conclusion CEA favours the more costly HIT if decision makers are willing to spend at least €4000/QALY for tobacco cessation treatment.

Several smoking cessation interventions, targeted at current smokers, are available; 2 furthermore, evaluations so far have confirmed the effectiveness of the majority of them. 3 Additionally, some recent studies emphasise that higher level of intervention intensity, such 4 as additional counselling sessions 4 and intensive support through a mobile application 5 , 5 resulted in the highest smoking cessation rates. However, due to increasing number of 6 available interventions, decision-makers have to decide which intervention to implement, 7 taking into account that intervention intensity reflects intervention costs. Relative costs and 8 benefits of those interventions are important criteria, thus, increasing the attention on 9 economic evaluations in recent years 6 7 . Economic evaluations combine the costs and 10 outcomes of different interventions and aim to determine which intervention provides the best 11 value for money 8 . Several studies on the cost-effectiveness of smoking cessation 12 interventions comparing different intensity of support have been performed during the last 13 few years. For example, Quit-and-Win programme 9 , comparison of standard, enhanced and 14 intensive smoking cessation interventions using cell phones 10 , and two smoking cessation 15 approaches of different level of intensity for cancer patients 11 . The results suggested that the 16 higher intensive interventions are preferable from health economics point of view, but all 17 those evaluations were based on 6-or 12-months follow-up, long-term follow-ups are scarce 18 in randomised controlled trials. 19 The effects of smoking on health occur during many years because current smoking 20 influences future health risks; similarly, a smoking cessation today will cause smoking related 21 health risks to tail off gradually. Thus, in order to estimate cost-effectiveness of smoking   [12][13][14] . Berg at al. 13 3 identified 64 economic evaluations in smoking cessation, and the state-transition Markov 4 model was most frequently used. The majority of the models simulates the lifetime 5 development of morbidity and mortality for smoker vs former smoker using relative risks for 6 four diseases, chronic obstructive pulmonary disease (COPD), coronary heart disease (CHD), 7 stroke, and lung cancer. The authors concluded that existing economic evaluations in smoking 8 cessation vary in quality, resulting mainly from the way in which they selected their 9 populations, measured costs and effects, and assessed the variability and generalisability of 10 their own findings 13 . One of the reasons is that all those studies are based on short-term 11 follow-up (from six months to one year), and they never had a chance to validate the 12 sustainability of short-term effectiveness in real life; thus they cannot confirm the reported 13 cost-effectiveness results and policy recommendations. Moreover, the long-term assumption 14 might change the results of the smoking cessation cost-effectiveness 15 . 15 Our previous economic evaluation of high-and low-intensity programmes (HIT and LIT) for 16 smoking cessation in a dental setting was based on the reported number of quitters measured 17 as point prevalence abstinent (not one puff of smoke during the past seven days prior to [1][2][3][4][5][6][7][8][9][10][11][12][13][14][15][16][17][18] year follow-up). The conclusion was that high-intensity treatment support is the preferred 19 option if the decision-maker's willingness-to-pay exceeds €5,100 (50,000 SEK) per QALY.

20
The base-case scenario of the analysis assumed a sustained abstinence for the quitters 16 . The 21 long-term follow-up of the programmes was performed five to eight years later 17 . In this 22 study, we used a unique opportunity to compare cost-effectiveness analyses of a high-and a 23 low-intensity smoking cessation intervention in a dental setting, using data from short-term 24 (1-year) and relatively long-term (5-8 years) follow-up. smoking cessation date. The effectiveness analysis showed that the difference in outcome 2 between the HIT and LIT programmes remained relatively constant and significant in favour 3 of HIT, and that abstinence at 1-year follow-up was a good predictor for long-term abstinence 4 17 . All analyses were done using the "intention to treat" approach where non-responders were 5 considered as smokers. The original study, as well as the long-term follow-up, was approved 6 by the ethical committee at Uppsala University (Dnr:Ups 02-457, Dnr: 2010/172). 7 The mean age of the participants was 49 years, and 78% were women. Short-term follow-up 8 (one year) questionnaire was answered by 84% of the randomised participants (88% for HIT 9 vs 81% for LIT). Fourteen per cent (41 of the 300 participants) reported 6-month continuous 10 abstinence (not one puff of smoke during the past 6 month); 27 (18%) individuals in HIT vs 11 14 (9%) in LIT. At long-term follow-up (5-8 years), 241 persons answered the questionnaire 12 (80% for both HIT and LIT). Of those, 24 were sustained abstinent (17 vs 7 for HIT vs LIT) 13 since the planned smoking cessation date. Characteristics of the study participants as well as 14 abstinence at the 1-year and at the long-term follow-up are presented in Table 1. 15 16  with discounting (3% per year) and sensitivity analyses 8 19 . The ICER was calculated by 2 dividing the difference in total costs for the programmes (incremental cost) by the difference 3 in the health outcomes in QALYs (incremental effect) to provide a ratio of extra cost per extra 4 unit of health effect. The intervention costs were collected prospectively by interviewing the three dental 9 hygienists who carried out the patient work as well as the project leader and the project 10 coordinator. The costs were divided into joint costs for the two programmes and programme- 11 specific costs, and undiscounted because of the short 3-year project time. The joint costs were 12 assumed, divided equally between the programmes while the programme-specific costs 13 included staff time for patient work, material, and participant costs. Estimation of the 14 intervention costs has been described in detail previously 16 . All costs were measured in 15 Swedish kronor (SEK) in For CEA short-term, we used 6-month continuous abstinence at 1-year follow-up reported by 23 41 participants (14 from HIT and 27 from LIT). For CEA long-term, we used sustained 24 abstinence at 5-8 years reported by 24 participants (17 from HIT and 7 from LIT), see Table   25 1. Both measures were significant different between the treatment programs.     incremental QALYs slightly decreased which resulted in higher incremental cost of €9,263 2 per QALY, see Table 3.

4
At all values of willingness-to-pay for a QALY, including zero, the HIT was more cost- 5 effective than the LIT, see the probabilistic sensitivity analysis in Figure 1. Main results 17 18 In this study, we performed a cost-effectiveness analysis using the long-term follow-up data 19 from a RCT of a high-and a low-intensity treatment programme (HIT and LIT) for smoking 20 cessation in a dental setting. We also validated the cost-effectiveness results of the previous 21 study based on short-term follow-up. HIT was more effective in getting participants to quit 22 smoking and to keep sustained abstinent, resulted in higher societal costs avoided and more 23 QALYs gained among both men and women, compared with LIT and thus was more cost-24 effective. The incremental cost-effectiveness ratios (ICERs) were €936 and €1,020 using 25 short-and long-term effectiveness, respectively, which are below the Swedish willingness-to-26 pay threshold of €50,000 per QALY 25 , thus, indicating that the resource intensive HIT was 27 highly cost-effective. The results also confirm the conclusions of the previous cost-28 effectiveness analyses based on short-term follow-up data, and we would recommend the use 29 of the HIT programme as a cost-effective option for smoking cessation. QALY decreased for both programmes. As mentioned in our previous study 16 , the Markov 2 model indicates considerably lower smoking-related disease risks for women reported by 3 large epidemiological studies (see model technical report for details) 21 , and thus lower cost 4 savings and health gains from tobacco cessation for women than for men. Finally, the 5 intervention costs for the RCT study calculation was based on the trial protocol and might be 6 overestimated in comparison with routine praxis; however, in the ICER, those extra costs 7 were divided equally between the programmes, and thus disregarded. Comparison with other studies 10 11 We could not find any cost-effectiveness analyses based on more than 1-year follow-up, and 12 therefore we compared our results with other studies estimating cost-effectiveness of 13 interventions with different level of intensity using 6-or 12-month follow-up. Thus, a cost-14 effectiveness analysis of high intensity multiple contests and low intensity enhanced contest 15 of a Quit-and-Win programme reported that high intensity Quit-and-Win leads to an average 16 gain of 0.03 QALYs and was cost-saving, in comparison with lower intensity 9 . Another study 17 presented a cost-effectiveness analysis of three smoking cessation interventions with different 18 intensity levels: Standard Care (SC) (brief advice to quit, nicotine replacement therapy and 19 self-help written materials), Enhanced Care (EC) (SC plus cell phone-delivered messaging) 20 and Intensive Care (IC) (EC plus cell phone-delivered counselling) 10 . The overall conclusion 21 was that the higher intensive intervention (IC) was the most cost-effective strategy both for 22 men and women, which is in line with our results. Additionally, a cost-effectiveness analysis 23 of two smoking cessation approaches for cancer patients was presented in a study from 24 Canada 11 . The basic programme consisted of screening for tobacco use, advice and referral, 25 whereas the best practice programme included a basic programme and pharmacological 26 therapy, counselling and follow-up. The incremental cost-effectiveness ratio of the best practice programme compared to the basic programme was $3,367 per QALY gained for men, 2 and $2,050 per QALY gained for women. These results are very similar to our findings. In 3 our previous study 16 , based on the same RCT and 1-year follow-up, a higher ICER of 4 €9,900/QALY and €5,500 /QALY was calculated for point prevalence and continuous 5 abstinence respectively, but the overall conclusion confirmed the cost-effectiveness of HIT at 6 a willingness-to-pay of €10,000. In conclusion, the more costly HIT smoking cessation programme has the potential to be an 11 economically attractive option when compared to the LIT programme over a broad range of 12 assumptions. The HIT programme was sustainable cost-effective in comparison with the LIT 13 programme in the long-term. Cost-effectiveness analysis favours the more costly HIT if 14 decision-makers are willing to spend at least €1,000/QALY for tobacco cessation treatment. 15 These findings can support and guide implementation of smoking cessation programmes.   None declared. 5 6 Ethics approval 7 8 The Ethical Committee Uppsala University gave clearance for the smoking cessation study 9 Dnr: Ups 02-457. 10 11 Data sharing statement 12 13 Data is available from corresponding author (IF) on reasonable request. 14 References 15 16 1. [Folkhälsomyndigheten] PHAoS. Public Health Agency of Sweden [Folkhälsomyndigheten] 17 Stockholm2018 [Available from: https://www.folkhalsomyndigheten.se/the-public-health- 18 agency-of-sweden/living-conditions-and-lifestyle/alcohol-narcotics-doping-tobacco-     risks ............................................................................................................................................................ 7 Death risks ............................................................................................................................................................ 10

Introduction
This is a technical report on an updated version of a model, originally developed in year 2004 , to enable systematic cost-effectiveness analyses of tobacco cessation interventions in Sweden. It aims to follow international and Swedish recommendations of cost-effectiveness analyses in health and medicine. The model holds a societal perspective, aiming to incorporate available disease-specific costs for all sectors of Swedish society. The updated model contains more recent data on societal costs, disease and death risks, and quality-of life-estimates, to enable estimates that reflects current Swedish conditions. The model simulates the lifetime societal effects of quitting smoking on three diseases: lung cancer, chronic obstructive pulmonary disease (COPD), and cardiovascular disease (CVD) including coronary heart disease (CHD) and stroke. The model incorporates the smoking-related disease risks, the remaining disease risks after tobacco quitting, the death risks in the diseases and unrelated diseases, as well as the societal effects of the diseases. The societal effects include medical treatment costs, costs for institutional care, drug costs, costs for informal care and other costs for patients and relatives, and morbidity productivity costs, as well as loss of life-years and quality-adjusted life-years (QALYs).
This technical report contains a description of the model structure, of all the data sources used and of the assumptions made. For validation purposes, it also reports model estimates for some selected age-groups and more detailed outcomes and sensitivity analyses for one age-group, men and women aged 50 years at the start of the simulations. To investigate model uncertainty, univariate and multivariate sensitivity analyses are reported, as well as a probabilistic analysis. The model validity is discussed in the final section of the report.

Method The diseases
The model incorporates the three most smoking-related diseases: lung cancer, chronic obstructive pulmonary disease (COPD), and cardiovascular disease (CVD) including coronary heart disease (CHD) and stroke, see table 1. The model is restricted to the effects on the individual smoker/quitter, thus not incorporating any side-effects on other people.

The model
The stochastic model simulates the societal effects of smoking cessation on three smokingrelated diseases. It is constructed as a Markov-cycle tree model appropriate for microsimulations.
The Markov model is a health state-transition model  using probabilities for transitions between health states. These probabilities are the age-and gender-specific disease risks, conditional on smoking status and years since quitting, and age-, gender-and disease-specific death risks. The states are mutually exclusive and collectively exhaustive, and transitions between disease states are not allowed. The only exits from disease states are death, in the disease in question or in unrelated diseases, except for 5-year survivors in lung cancer which are assumed to recover to complete health. All other disease states are assumed to last life-long. See figure  1 for the state-transition diagram.
The Markov stages are one year-long, with no half-cycle correction. The starting point is the state healthy. The model covers the ages 15 to 95 years. The Markov-cycle tree has been created in Treeage Pro (Treeage Inc., 2015).

Material
The model is based on principles for cost-effectiveness analysis in health and medicine  and Swedish methodological recommendations (TLV, 2004). The model holds the societal perspective, aiming to incorporate diseasespecific costs for all sectors of Swedish society.
The model uses Swedish register data and secondary data from previously published scientific articles. The secondary data was found through searches in the database MEDLINE and the reference lists of retrieved articles, choosing the data that is considered most relevant to present-day Swedish circumstances and the target group. No metaanalysis nor other synthesis of data was performed.
All costs are expressed in year 2014 SEK (USD 1=SEK 6.86; Euro 1=SEK 9.10), converted if necessary by the Swedish CPI (consumer price index). The annual discount rate is 3% for both costs and health effects.

The risks
Disease risks All disease risks are annual age-and gender-specific excess incidence risk until the age of 95 years, see tables 2 to 5.
The COPD death risk is estimated from the study Obstructive Lung Disease in Northern Sweden (OLIN), which reported the 20-year mortality in three age groups. Comparison with the general age-specific mortality risks revealed no excess risk of death among those younger than 58 years and older than 70 years, but a considerable increased risk among those aged 58-70 years at follow-up. The excess risk was estimated at on average around 5 times the age-and gender-specific general population death risk, see table 2.
The lung cancer death risk is based on survival data from the Swedish National Cancer Registry, see table 8. The death risks for year 3 and 4 after diagnosis are estimated by linear interpolation between years 2 to 5. Lung cancer survivors at 5 years are assumed recovered, and returned to the health state healthy.
The death risks from CHD and stroke are taken from Swedish registers, see tables 9 to 11, or published scientific reports, see table 5. The death risks for AMI, stroke and IHD are divided into risks the first year after the first event and the second and following years after first event.  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y 11  The model also incorporates the possibility of dying in unrelated diseases. The death risk in the health state Healthy is the average 5-year age group-and gender-specific risk adjusted for all model disease deaths, the last column in table 12. In disease health states, the risk of dying in unrelated disease is the average 5-year age group-and gender-specific  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y   12 risk adjusted for the deaths in each respective disease. For ages below 39 years the risk in the age group 35-39 years is used, and for ages 80-84 years the risk >79 years. For ages above 84 years, the general population age-and gender specific death risk is used for the unrelated death risk. As the lung cancer death risks are so high, the unrelated death risks for lung cancer among individuals aged above 84 years had to be adjusted, by deducting 0.05. For those aged below 85 years, the age-and gender-specific general population risk of death is only used for calculating some disease-specific death risks, see tables 2 and 4. The risk is taken from the Swedish national mortality statistics for the year 2014 (Statistics Sweden, 2015).

Changes in risk after quitting smoking
The excess disease risks for smokers are not eliminated immediately after quitting smoking. This "lead time" is 36 years for lung cancer, 16 years for CHD, and 11 years for stroke, while for COPD some excess risk remain life-long, see heading effect of quitting in tables 2 to 4. The disease risks after quitting are constructed by adjusting the smokers' risks by the remaining risk. The remaining risk is modelled as fractions of risk, given the number of years since quitting. The annual remaining risks are estimated by linear interpolation. The effects on the risk for CHD and stroke are modelled on the dummy variable smoking, adjusting the value of 1 by the remaining risk fraction.

The societal costs
The model is reflecting the societal perspective, including disease-related costs for all sectors of the Swedish society. The costs included are medical treatment costs, costs for institutional care and technical aids, pharmaceutical costs, informal care and other patient and relatives' costs, and morbidity productivity costs.
Most of the data on societal costs are taken from Swedish studies published during the 2010s. Data reported as distributions, i.e. with the Gamma parameters for costs, or bootstrapped 95 percent confidence interval were preferred and used in the model to  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y 13 enable stochastic estimation. If data was reported as mean and standard deviation, the Gamma distribution was simulated employing the Treeage function. In one case, data was reported as fraction of patients consuming a specific resource, which was used for sampling within the model. Otherwise the reported point estimate, usually the average cost across the patient group, was used. If no Swedish data on a cost item was found, the cost was taken from studies reporting data from settings assumed similar to the Swedish.
All costs are reported in SEK year 2014 (USD 1=SEK 6.86; Euro 1=SEK 9.10), adjusted when necessary with the Swedish CPI. To adjust reported Gamma distributed parameters to the price level, only the second parameter, i.e. the scale parameter, was adjusted.

Pharmaceutical costs
Costs for pharmaceuticals in Sweden ought to be divided between the county councils and the patients, as patients pay a considerable share in co-payment. This is however not possible, given the data available. Table 15 therefore presents the drug costs to the regional healthcare authorities. The costs of pharmaceuticals dispensed during hospital stays are included in the medical treatment costs.

Informal care and other patient and relatives' costs
These costs include the value of care given to patients by relatives and other costs for patients or relatives, such as time, co-payments paid for health care and drugs as well as costs for transportation, modifications at home etc. Complete estimates could not be obtained for any disease, see table 16, except IHD which however might be outdated. Informal care in present-day Sweden probably constitute a sizeable part of total societal costs.  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y 15

Productivity costs
The productivity costs only value the lost production because of morbidity before the age of 66 years, not mortality. The productivity costs for lung cancer is simulated within the model, via sampling from the fraction of patients on sick leave and combined with ageand gender-specific average monthly wages, including 40% employer taxes. Remaining data is taken from the literature, see table 17, and most estimates are recent. The costs are valued by the human capital method, and thus only include losses in salaried work before the official age of retirement.

Life years lost
The number of life years lost (YLS) are calculated until the age of 95 years, and only for individuals dead in the modelled diseases. Life years lost are presented both discounted 3% and undiscounted.

QALYs
The number of quality-adjusted life years (QALYs) are calculated during healthy years and years spent diseased, until death or the age of 95 years.
The disease-specific QoL used in the health states are all, except one, modelled as decrements from the average population age-group and gender-specific QoL, see table 19. For lung cancer no data was available on the marginal effect of the disease on the population average QoL, so a fixed value over the ages and genders had to be used.

Sensitivity analyses
Several univariate and multivariate sensitivity analyses have been performed. Analyses on some methodological issues, as well as a probabilistic sensitivity analysis, have also been performed. The analyses are reported for men and women aged 50 years.
To give another measure of the uncertainty surrounding the cost-effectiveness ratio, the 95% confidence interval for the difference between smokers and quitters is reported, calculated from the standard deviation of outcomes.

Probabilistic analysis
A bootstrap sampling was performed using the smoker and quitter Monte Carlo simulations of 10 000 runs. A sample of 1 000 from each simulation was drawn, with replacement, performed in Microsoft Excel. The mean of the difference in costs and QALYs between smokers and quitters was then calculated. This was replicated 1 000 times. The bootstrap is represented as a scatterplot in the cost-effectiveness plane.

Results
In this chapter, the model estimates of QALYs, YLS and societal costs are presented for men and women in some selected ages, mainly for validation purposes. More detailed simulation outcomes as well as the results of the sensitivity analyses are presented for men and women at age 50 years. Model estimates can be obtained for men and women for all ages between 15 and 95 years.

The model estimates
In table 20 the simulation results for QALYs (quality-adjusted life-years) experienced until the age of 95 years are presented, for the selected ages 15,30,50 and 70 years at the start of the simulations. As can be expected, the number of QALYs are highest in the younger age groups, and somewhat higher for women in most age groups. In the selected age groups, the differences between smokers and quitters are at a maximum at age 30; 0.68 for females and 0.81 for males. The confidence intervals, calculated via the mean and standard deviation (sd) from the 10 000 model runs, indicate that there are differences in QALYs between smokers and quitters.
The YLS (life-years saved) lost before the age of 95 years are presented in tables 21 and 22, discounted 3% and undiscounted. The differences in discounted YLS between smokers and quitters are somewhat higher than the differences in QALYs. The undiscounted YLS in table 22 show the number of years that smokers and quitters are expected to lose before the age of 95 years. For the ages 15, 30, and 50 the number of lost life-years is estimated at around 6 years for women smokers and 9 years for men, implying that the female smokers are estimated to live until age 89 and the male until age 86. In the oldest age group presented here, age 70, the number of lost life-years are only 1-2 years. The quitters are estimated to lose considerably fewer life-years; 1-4 years for the women and 3-5 years for the men. As expected, the difference between smokers and quitters diminish with age, with a maximum at around 5 years for the females and around 6 years for the males at age 15.
The societal costs estimated for the smokers and quitters for the selected age groups are presented in table 23. The highest costs are found for age 50; 200 000 SEK and 250 000 SEK for the smokers and 130 000 and 170 000 for the quitters, in both cases higher among the men. The highest difference between smokers and quitters is however found at age 30, with a difference of 100 000 among the females and 120 000 among the males. The difference among the eldest, at age 70, is around 20 000 SEK. These cost differences reflect the amount that tobacco cessation interventions could spend on achieving one quitter and still be cost-saving.

Selected model outcomes
The underlying estimated disease outcome is presented in figures 2 and 3, for the age 50 years. For both women and men, there is a marked decrease for quitters in the number of diseased and dead in the model diseases, which is somewhat offset by an increase in the number of deaths in unrelated diseases. The number of diseased and deaths are higher for    men, mainly originating from CHD. The model disease with the highest smoking-related incidence is COPD, for both genders. The increase in unrelated deaths for the quitters is an example of competing risks, which decreases the difference in life-years and QALYs between smokers and quitters. Table 24 and 25 shows the full model simulation results of the societal cost savings because of tobacco quitting at age 50 years. For women, the highest estimated savings are found in lung cancer, COPD and stroke at around 15-20 000 SEK per quitter. For men the cost savings because of lung cancer are considerable higher, at around 35 000, due to the higher incidence among the men. The cost item with the largest cost savings are medical treatment costs for both genders, at around 30 000 SEK. Most of the difference in savings between men and women originate from the productivity costs, possibly reflecting disease onset at younger ages among men. Note that a cost saving of zero means that no cost is being modelled, as cost data was lacking.

Sensitivity analyses
The results of the sensitivity analyses are presented on women and men at starting age 50 years. Figure 4 shows the results for women and figure 5 for men.
All analyses show a similar pattern between men and women, and also similar ranges. The univariate sensitivity analyses on the model parameters, analyses A to E, result in small changes in costs and QALYs. Also the multivariate analyses F and G, which are constructed as scenarios that allow the risk parameters to vary consistently upwards or downwards, and along with the costs in analysis G, show moderate changes from the base case estimates. The methodological choices have a more pronounced effect, as the largest difference in QALYs is achieved by varying the discount rate (analysis H) between 0 and 5%, which also affects the costs substantially. The two analyses that reflect the choices of which costs to include in the estimates, analysis I that reflects the UK NICE health care and social services perspective and analysis J that only include Swedish data published since the year 2005, both decrease the cost differences between smokers and quitters.  The scatter plot of the bootstrap analysis based on the microsimulation results for women and men aged 50 are shown in figures 6 and 7. The uncertainty is higher for the men, as the plots are more scattered. All plots are however situated in the cost decrease and QALY increase quadrant, with costs below -20 000 SEK and QALYs over 0.2.

Discussion: Model validity
The discussion of the model validity is structured around four aspects as proposed by : the structure of the model, the inputs to the model, the results of the model and the value of the model to the decision-maker.
The structure of the model The structure of the model is a Markov model constructed for microsimulations, on the three most smoking-related disease groups; lung cancer, COPD, and CVD including stroke and CHD. The present updated version of the model includes one less CHD disease compared to the first version of the model, as unrecognized acute myocardial infarction now is included in the IHD disease, mainly because the disease definition is rarely used nowadays. Choosing only three disease groups is a clear simplification as smoking is known to cause hundreds of different diseases. The effects from smoking, and thus quitting, are furthermore confined to the individuals themselves; no side-effects on other individuals such as environmental tobacco smoke or smoking uptake are included. These two features leads to an underestimate of the true effects of tobacco quitting.
The same disease-specific approach has been taken by most other tobacco cessation models , even though some of them include more diseases, such as asthma. Another approach would be to use the overall differences in mortality between current, former, and never-smokers taken from large US studies, as some early tobacco cessation models did Tengs et al, 2001). In order not to overestimate the effects of quitting tobacco, we chose to model the smoking-related risk for certain diseases instead, as it is improbable that all differences in mortality and morbidity between smokers and former smokers are due to the smoking habit (Doll et al, 1994).
The model aims to reflect disease onset related to smoking tobacco. As disease in all the three disease groups included in model may be caused by other factors than smoking only the excess risks for smokers are modelled. For the diseases lung cancer and COPD this implies that the risk for smokers found in epidemiological studies is adjusted by the risk found for non-smokers. For the disease group CHD and stroke, where a large fraction of disease onset is caused by other factors than smoking, this adjustment for smokers´ excess risk was performed by setting the other risk factors in the risk function at minimal risk levels. This is an underestimate, as the risk factor levels among smokers can be expected to be at least as elevated as among the general population. The underestimate is aggravated by the fact that the functional form of the risk function results in a multiplier effect of the risk factors.
The present version of the model includes seven health states: lung cancer, COPD, stroke, and CHD divided into four diseases. This is a clear simplification, as the costs and QoL can be expected to vary considerable between patients with different severity levels within the diseases. This is particularly true for COPD which is a chronic progressive disease, i.e. the An obvious problem with the model, inherent in all Markov models, are the mutually exclusive health states; any individual can only contract one disease, and once diseased the individual never recovers (apart from the very rare 5 year survivors in lung cancer). This feature implies both an overestimate and an underestimate of the true effects. The underestimate stems from the fact that co-morbidity is very common, especially among the individuals with the chronic diseases COPD, CHD, and stroke. The overestimate of costs and effects arise as individuals stay in the health states until death. If the costs and outcomes associated with the health states are taken from severely ill individuals, then these become grossly overestimated. This overestimate is partly offset by the use of separate costs for the first and subsequent years, for all societal costs due to AMI and stroke. In order not to overestimate the numbers of years spent in disease states, the possibility of dying in unrelated diseases is present in all health states. This feature is also included in the CHD Policy Model (Weinstein et al, 1987).
Most tobacco cessation models are built for cohort estimation , but this model is constructed for individual-level estimation using the microsimulation methodology. As the data available admitted a microsimulation structure, e.g. the risk functions, the methodology was chosen as the advantages to model and to obtain a richer data set with results that reflect the heterogeneity of outcomes between individuals was deemed to offset the disadvantages of calculation burden. The use of the software Treeage also facilitates the use of microsimulation. Age-and gender-specific estimates can thus be obtained from the model, between ages 15 and 95 years.
The model stages are one-year long, which seems accurate given the risk estimates and the long time horizon of the model. The reason for the model maximum age of 95 years is the lack of risk estimates for older ages. Some extrapolations of risk estimates to the age of 95 years indeed resulted problematic, as some disease-specific death risks expressed as multipliers of the average age-specific death risk resulted in risks above 1. Further extrapolations beyond the age of 95 years were deemed unnecessary, as most of the relevant differences between smokers and quitters would have arisen by that age.

The inputs of the model
The second aspect of model validity is the inputs of the model. The model contains a large number of data taken from different sources. This is of course a threat to the internal validity of the model, shared with most models. However, the data have been chosen to reflect current Swedish circumstances. The current updated version of the model has exchanged almost all cost data, if more recent estimates were available, and all death risks to recent Swedish register data. As the number of studies on any particular data items are few, no meta-analysis or any other synthesis of data was carried out. The disease risks are of course are pivotal for the result. The lung cancer disease risks are probably the best that can be obtained, from a large epidemiological study . The risk equation used for CHD and stroke is taken from the Framingham studies, and even though there are more recent risk scores developed from the study (D'Agostini et al, 2008), the  risk functions are still frequently employed. The disease COPD has been the subject of a large long term epidemiological study in Sweden, The Obstructive Lung Disease in Northern Sweden (OLIN) , which is thus the most relevant data source for the model.
In the model, there is an increased risk for a smoking-related disease remaining for some years after the tobacco cessation, in accordance with epidemiological evidence (Surgeon General, 1990;. The feature is also considered a marker of high quality tobacco cessation models .
The majority of the cost data are taken from Swedish studies published during the 2010s. To take fully advantage of the microsimulation structure and to obtain stochastic estimates, the preferred data sources were the ones reported as distributions, i.e. as Gamma parameters or bootstrapped 95 percent confidence intervals. If no Swedish data was found, an international estimate was instead used in order to seek to represent the full societal costs. However, apart from certain cost items and for some of the diseases, the lack of data results in considerable underestimates of the true societal costs. This is particularly true in the cases of the costs for care, both institutional and informal. The institutional care could amount to considerable costs, exemplified by the costs for stroke and AMI patients, see table 14. In particular for lung cancer the lack of data results in considerable underestimates of the true disease-related costs. This is why the possible overestimate of the informal care for the disease, obtained from an Italian study, probably does not bias the overall result. To investigate the issue, one sensitivity analysis only included recent Swedish data. The analysis lead to decreases in cost savings for quitters aged 50 years of around 30%.
The QoL estimates are constructed as disease-specific decrements from the average ageand gender-specific QoL, except for lung cancer for which no QoL decrement could be found . The average population age-and gender-specific QoL weights, which are certainly not 1, are also used during healthy years for the base case estimates. This means that the model assumes that an individual that avoids the smoking-related diseases is not having perfect health, but the health of an average Swede at the same age, as recommended .
The stated purpose of the model is to reflect the societal perspective, which for Sweden includes the morbidity productivity costs, but not the productivity costs resulting from mortality. All the model data on productivity costs value them according to the human capital approach for individuals under the age of 65, the customary Swedish age of retirement.
A full societal perspective might also include other aspects, considering that this is a model on individuals that are participating in an intervention that aims to change their lifestyle. The previous version of the tobacco cessation model, version 1 , reported 28 sensitivity analyses that modelled some effects on the tobacco quitters, by including savings from cigarette purchases and a decreased QoL because of withdrawal effects during the first year. When that analysis was applied to an intervention, a decreased QoL during the first year was also deducted for the smokers that failed to quit, as the failure to achieve a personal goal might to lead to a decrease in QoL.
The results of the model The third aspect of model validity is the results of the model, e.g. a comparison with reality or with other study results. A direct comparison with reality is not possible, since the model covers the ages 15-95 years, with a follow-up time of 80 years for the youngest age group.
The model estimates that around 60% of the women and 70% of the men aged 50 at the start of the simulations will contract one of the modelled diseases, and that around 50% of those will die in the diseases before the age of 95 years. The disease risks for the quitters at age 50 are not eliminated; 30-40% of them will still contract the smoking-related diseases because of remaining disease risks after quitting. As expected, the unrelated deaths increase among the quitters, in sum leading to an increase in YLS (undiscounted) of 2-3 years for those quitting at age 50, compared with continuing smokers. The increases in QALYs (discounted 3%) are smaller because of less-than-perfect health among those aged 50 years and above; 0.61 for women and 0.77 for men. The disease outcomes are fairly similar to the estimates from the previous versions of the model, but because of decreased death risks, the outcomes in terms of YLS and QALYs are considerably higher. The 2004 version of the model estimated an increased YLS of 0.93 and of 1.66 for women and men aged 50-54 years, and QALY gains of 0.36 and 0.71, respectively. The differences are due to the longer time perspective of the present version, 95 years versus 85 years, and the somewhat decreased case-fatality risk (i.e. the mortality risk among those with disease) because of improvements in medical technologies during the past decade.
Apart from increases in health, the societal cost savings because of quitting smoking are considerable. For men, the cost savings amount to around 100 000 SEK for quitters aged between 15 and 50 years, and around 70-90 000 SEK for women. Even in the age group 70 years there are estimated cost savings of around 20 000 SEK per individual quitter. This implies that substantial funds could be invested in smoking cessation interventions, and the interventions would still be cost-effective, or even cost-saving. The cost savings in the present model are considerably higher than those of the previous model, in part due to changes in price year.
Comparisons of model estimates with other models' are difficult to perform, as the time horizon, costs included, jurisdiction, and the diseases included differ. Among the recently reported model estimates , there are two Australian models. The model developed within ACE  report estimates of life-years saved that are considerable higher than the present model's; 5.7 years for men and 6.6 years for women in age group 50-54 years. That model time horizon is however 100 year, but it is unlikely There have been two, to my knowledge, reports of tobacco cessation model estimates for Sweden, one using the Benesco model  and one using an extended version of the HECOS model . Comparison with those model estimates are unfortunately not possible, due to lack of reporting detail. However, estimates from the previous version of this model were fairly consistent with the HECOS model estimates  for Sweden, available at the time .

The value of the model to the decision-maker
The fourth aspect of validity is the value of the model to the decision-maker. There are several models on tobacco cessation that conforms to international recommendations on how to perform cost-effectiveness analyses . This model however reflect Swedish circumstances, with Swedish cost and QoL data, why the model might be useful for Swedish decision-makers.
We hope that the model will be used to perform economic evaluations of a range of tobacco cessation interventions. For tobacco prevention interventions, i.e. prevention of initiation of smoking, another model version, version 2, has been constructed and is available for analyses. The use of these models will in time enable incremental and marginal calculations of the cost-effectiveness of different tobacco interventions and their components and suitable target groups. The basis for decisions on which tobacco cessation and prevention interventions to implement will then be more comprehensive.
Another frequent use of models is to forecast future events. This model is not suitable for estimating what the costs of smoking will be in the future. The reason is that the model does not incorporate any adjustments of possible future developments. The risk of smoking is based on studies with follow-up periods of sometimes 30 years, which means that the risks are reflecting the smoking behaviour among smokers 30 years ago. The changes in cigarette content and in the frequency of smoking might lead to changes in disease risk in the future. Also the costs for the smoking-related diseases might change in the future, because of changes in health care technology. Another example would be the value of the morbidity productivity costs, as well as informal care, as wages and productivity often are expected to increase in the future.
Nevertheless, the model actually forecasts what the costs for smokers and quitters will be in 80 years' time, for the youngest age group. That implies that we know that the model forecasts will be wrong, but it is of minor significance as the model is constructed to be used for comparisons between two groups, smokers and quitters, thus eliminating some

The uncertainty
Another aspect of model validity is the uncertainty surrounding the model estimates.
The univariate sensitivity analyses on the model parameters (analyses A-F in figures 4 and 5 for men and women aged 50) show minor deviations from the base case result, while the multivariate analysis on costs and risks combined (analysis G) affects in particular the cost estimates. The methodological choices affect the results to a greater extent, with the discount rate (H) heavily influencing the QALYs and the more restricted perspective (I) decreasing the cost-savings. The multivariate analysis that only include higher-quality data (J) also imply decreases in the cost differences between smokers and quitters, but the difference remains substantial; around 50 000 SEK for females aged 50 years and 60 000 SEK for men, respectively. The overall conclusion from the parameter sensitivity analyses is that the QALY gains are at least 0.35 and 0.40 and the cost savings at least SEK 35 000, for female and male quitters aged 50, respectively.
The probabilistic analysis shows no uncertainty whether quitting tobacco leads to costsavings and increases in QALYs, as all bootstraps are placed in the southeast quadrant of the cost-effectiveness plane. The bootstrap results exhibit a mixture of first and second order uncertainty, as it reflects both the probabilistic structure of the Markov model and the simulation of some parameter values .
Another measure of uncertainty is the confidence intervals around the estimated mean differences, reported in tables 20-23. However, that measure is not fully appropriate as the large sample sizes of the Monte Carlo simulation (10 000 runs) diminishes the standard error of the mean .
The structural uncertainty of the model, i.e. whether the results would be different if the model would have been constructed in another way, have not been studied. Alternatives to the chosen model structure could have been deterministic or discreet event simulations, more or less health states, other functional forms of risk functions, and other subgroups than men and women and five-year age-groups model results. The flaw is however shared with most tobacco quitting models .

Checking for technical errors
The model contains a large number of trackers, i.e. variables that count events, to enable checking for technical errors. Tentative runs were executed after the introduction of every new variable, with cost items undiscounted, and the simulation results examined manually. Thus, the model has been thoroughly checked for technical errors.

Conclusions
The aim of this study is to develop a model predicting health and economic consequences of smoking cessation, to be used for cost-effectiveness analyses of smoking cessation interventions. The updated model strives to incorporate data that is recent, accurate and appropriate for Sweden in year 2015. The model also adhere to Swedish recommendations on how to perform cost-effectiveness analyses within the health care sector. Data is however lacking to completely fulfil these requirements. Many model parameters are based on very few studies. Some information just does not exist, at least not accessible to us.
These are issues shared with most model, however. The purpose of modelling is to assemble the most accurate information at a point of time, to enable decision-making at that particular point of time. This is in accordance with one of the fundamentals of economics: decision-making under uncertainty, which implies that decisions have to be made even if there is no full information. We hope that the model will be applied to a range of different tobacco cessation interventions, which in time will enable a more comprehensive basis for decision-making. For each intervention, report mean values for the main categories of estimated costs and outcomes of interest, as well as mean differences between the comparator groups. If applicable, report incremental cost-effectiveness ratios.
Page 12-13, Table 2 Page 16, Table 3 20a Single study-based economic evaluation: Describe the effects of sampling uncertainty for the estimated incremental cost and incremental effectiveness parameters, together with the impact of methodological assumptions (such as discount rate, study perspective).

Not applicable Characterising uncertainty 20b
Model-based economic evaluation: Describe the effects on the results of uncertainty for all input parameters, and uncertainty related to the structure of the model and assumptions.

21
If applicable, report differences in costs, outcomes, or cost-effectiveness that can be explained by variations between subgroups of patients with different baseline characteristics or other observed variability in effects that are not reducible by more information.

Discussion
Study findings, limitations, generalisability, and current knowledge 22 Summarise key study findings and describe how they support the conclusions reached. Discuss limitations and the generalisability of the findings and how the findings fit with current knowledge.

Pages 17-21
Other Source of funding 23 Describe how the study was funded and the role of the funder in the identification, design, conduct, and reporting of the analysis. Describe other non-monetary sources of support.

Conflicts of interest 24
Describe any potential for conflict of interest of study contributors in accordance with journal policy. In the absence of a journal policy, we recommend authors comply with International Committee of Medical Journal Editors recommendations.
Page 21 "Competing interests" The CHEERS statement checklist format is based on the format of the CONSORT statement checklist

Cost-effectiveness of a high-vs a low-intensity smoking
3 cessation intervention in a dental setting: long-term 4 follow up. 5  Objectives. The aim of this study was to conduct a cost-effectiveness analysis of a high-and 5 a low-intensity smoking cessation treatment programme (HIT and LIT) using long-term 6 follow-up effectiveness data and to validate the cost-effectiveness results based on short-term 7 follow-up. analysis, and the results were presented as an incremental cost-effectiveness ratio (ICER).
14 Setting. Dental clinics in Sweden.

16
Interventions. Behaviour therapy, coaching and pharmacological advice (HIT) was compared 17 with one counselling session introducing a conventional self-help programme (LIT).

18
Results. The more costly HIT led to higher number of 6-month continuous abstinent 19 participants after 1 year and higher number of sustained abstinent participants after 5-8 years, 20 which translates into larger societal costs avoided and health gains than LIT. The incremental 21 cost/QALY of HIT compared to LIT amounted to €918 and €3,786 using short-and long-term 22 effectiveness respectively, which is considered very cost-effective in Sweden.

23
Conclusion. Cost-effectiveness analysis favours the more costly HIT if decision-makers are 24 willing to spend at least €4,000/QALY for tobacco cessation treatment.   Smoking is likely to remain the single most important preventable health risk in the world. 5 Despite continuously declining prevalence in recent decades, one in ten adults in Sweden still 6 smokes daily 1 . Cigarette smoking contributes to 7.5% of the burden of disease in Sweden 2   7 and was estimated to account for approximately €3,000,000 (31.5 billion Swedish krona, Several smoking cessation interventions, targeted at current smokers, are available; 14 furthermore, evaluations so far have confirmed the effectiveness of the majority of them. 15 Additionally, some recent studies emphasise that higher level of intervention intensity, such 16 as additional counselling sessions 4 and intensive support through a mobile application 5 , 17 resulted in the highest smoking cessation rates. However, due to increasing number of  approaches of different level of intensity for cancer patients 11 . The results suggested that the 2 higher intensive interventions are preferable from health economics point of view, but all 3 those evaluations were based on 6-or 12-months follow-up, long-term follow-ups are scarce 4 in randomised controlled trials.

5
The effects of smoking on health occur during many years because current smoking 6 influences future health risks; similarly, a smoking cessation today will cause smoking related 7 health risks to tail off gradually. Thus, in order to estimate cost-effectiveness of smoking  Our previous economic evaluation of high-and low-intensity programmes (HIT and LIT) for 4 smoking cessation in a dental setting was based on the reported number of quitters measured 5 as point prevalence abstinent (not one puff of smoke during the past seven days prior to [1][2][3][4][5][6] year follow-up). The conclusion was that high-intensity treatment support is the preferred 7 option if the decision-makers' willingness-to-pay exceeds €5,100 (50,000 SEK) per QALY.

8
The base-case scenario of the analysis assumed a sustained abstinence for the quitters 16 . The 9 long-term follow-up of the programmes was performed five to eight years later 17 . In this 10 study, we used a unique opportunity to compare cost-effectiveness analyses of a high-and a 11 low-intensity smoking cessation intervention in a dental setting, using data from short-term 12 (1-year) and long-term (5-8 years) follow-up.

14
We set out to: 1) perform a cost-effectiveness analysis of a high-and a low-intensity smoking 15 cessation programme in a dental setting using long-term (5-8 years) follow-up data and 2) 16 compare the cost-effectiveness results with the previous study based on short-term (1-year) 17 follow-up. In the smoking cessation intervention study 18 , between August 2003 andFebruary 2005, 300 25 adult smokers recruited via direct inquiry or advertising in dental or general health care were 26 offered smoking cessation support performed in a dental setting. Inclusion criteria were daily 27 smokers over 20 years of age, while exclusion criteria were reading difficulties and problems with Swedish language. The participants were randomly assigned to two interventions; one 2 received high-intensity and one low-intensity treatment support.

3
The high-intensity smoking cessation treatment, the HIT programme, comprised eight 4 individual sessions, of in total 3.5 hr over a period of 4 months, and was based on behaviour 5 therapy, coaching and pharmacological advice. The low-intensity smoking cessation 6 treatment, the LIT programme, comprised one counselling session, of up to 45 min, 7 introducing a conventional self-help programme running over 8 weeks. Both programmes 8 were free of charge.

9
The participants answered a baseline questionnaire and a short-term (one year after the 10 planned smoking cessation date) follow-up questionnaire. Demographic characteristics such 11 as gender, age and education level were also collected. The effectiveness of the trial was 12 reported elsewhere 18 . The analysis concluded that the more extensive and expensive HIT 13 programme was more effective and cost-effective, in terms of proportion of smokers who 14 were still smoke-free after one year 16 18 . The long-term follow-up was performed 5-8 years 15 after the planned smoking cessation date. The effectiveness analysis showed that the 16 difference in outcome between the HIT and LIT programmes remained relatively constant and 17 significant in favour of HIT, and that abstinence at 1-year follow-up was a good predictor for 18 long-term abstinence 17 . All analyses were done using the "intention to treat" approach where 19 non-responders were considered as smokers. Mortality and morbidity data for the participants 20 were not collected either by questionnaire or through the registers. The original study, as well 21 as the long-term follow-up, was approved by the ethical committee at Uppsala University

23
The mean age of the participants was 49 years, and 78% were women. Short-term follow-up

24
(one year) questionnaire was answered by 84% of the randomised participants (88% for HIT 25 vs 81% for LIT). Fourteen per cent (41 of the 300 participants) reported 6-month continuous since the planned smoking cessation date. Relapse rate was 26% and 50% for participants 5 reported 6-month continuous abstinence at 1-year follow-up in HIT and LIT respectively, but 6 the difference was not statistically significant. Characteristics of the study participants as well 7 as abstinence at the 1-year and at the long-term follow-up are presented in Table 1.  14 Both analyses used the same methodology described below. 15 Economic evaluations were based on the costs to implement the programmes, the number and  Intervention costs 5 6 The intervention costs were collected prospectively by interviewing the three dental 7 hygienists who carried out the patient work as well as the project leader and the project 8 coordinator. The costs were divided into joint costs for the two programmes and programme-9 specific costs, and undiscounted because of the short 3-year project time. The joint costs were 10 assumed, divided equally between the programmes while the programme-specific costs 11 included staff time for patient work, material, and participant costs. Estimation of the 12 intervention costs has been described in detail previously 16  LIT. 15 Intervention effectiveness 16 17 For CEA short-term, we used 6-month continuous abstinence at 1-year follow-up reported by 18 41 participants (14 from HIT and 27 from LIT). For CEA long-term, we used sustained 19 abstinence at 5-8 years reported by 24 participants (17 from HIT and 7 from LIT), see Table   20 1. Both measures were statistically significant different between the treatment programs. In 21 order to generalize the long-term effectiveness of our study, we performed a logistic 22 regression analysis to calculate the probability of sustained abstinence depending on 23 programme (HIT vs LIT), participant's gender and age, see Table 2.  The type of the programme (HIT vs LIT) was significantly associated with sustained 6 abstinence while gender and age were not. The regression equation [1] demonstrates 7 dependence between "abstinence" (1 -abstinence, 0 -no abstinence) and "programme" (1 -8 HIT, 0 -LIT), "gender" (1 -male, 0 -female) and "age" (19-71): decreased by 25%, the disease risks by 50%, the death risks by 10%, and the risk fractions 2 after quitting by 0.1. This low cost/low risk analysis led to substantial decreases in cost and 3 QALY differences between quitters and smokers. This sensitivity analysis was applied to 4 compare costs and effects between HIT and LIT, to validate the results of the CEA long-term. 5 To increase the generalizability of the cost-effectiveness results, we have also applied the 6 probabilities of long-term abstinence depending on programme (HIT vs LIT), participant's 7 gender and age on the modelling results. We estimated the avoided social costs and gained 8 QALYs for a random quitter from our sample and then adjusted the results to the probability 9 to quit (Abstinence), calculated in [1]. Cost-effectiveness was estimated for men and women 10 separately.

11
Further, a probabilistic sensitivity analysis (PSA) was conducted, based on the uncertainty of 12 the difference in sustained abstinent participants in the two programmes. The effectiveness of

13
LIT was fixed at the 7% quit rate, but the HIT quit rate was sampled from the 95% confidence 14 interval (9% -22% Model estimates for the CEA short-term and CEA long-term are presented in Table 3 (societal   6 costs and QALYs). The second and third columns in Table 3    1-year follow-up 7 f NLp -number of 6-month continuous abstinent participants LIT treatment programme according to [1][2][3][4][5][6][7][8] year follow-up 9 g NHs -number of sustained abstinent participants HIT treatment according to [5][6][7][8] year follow-up 10 h NLs -number of sustained abstinent participants LIT treatment according to 5-8 year follow-up 11 N*-difference in numbers of 6-month continuous abstinent participants between the treatment 12 programmes according to 1-year follow-up 13 N**-difference in number of sustained abstinent participants between the treatment programmes 14 according to 5-8 year follow-up 15 16 17 The CEA short-term indicated that HIT led to additional avoided societal costs of €82,253 and The more costly HIT programme led to a higher number of 6-month continuous abstinent 25 participants at 1-year follow-up (CEA short-term) as well as higher number of sustained 26 abstinent participants at 5-8 year follow-up (CEA long-term), which translates into larger 27 costs avoided and health gains than LIT, see Table 4. However, the difference in intervention 28 costs were not fully balanced by the societal costs avoided, so HIT implied an incremental net 29 cost of about €6,832 in CEA short-term and €21,619 in CEA long-term, compared with LIT.

30
HIT was estimated to lead to more QALYs, so the incremental cost per QALY of HIT 31 compared with LIT amounted to €918 for CEA short-term and €3,786 for CEA long-term, 32 which is considered to be very cost-effective in Sweden 20 . The incremental analysis favours 33 the more costly HIT, if decision-makers are willing to spend at least €4,000/QALY for 34 tobacco cessation programmes.   Sensitivity analyses 15 16 The most conservative sensitivity analysis, a multivariate low cost/low risk analysis, was 17 applied to CEA long-term. This analysis led to substantial decreases in avoided social costs 18 and QALY gains for both HIT and LIT. At the same time, the incremental costs increased and 19 incremental QALYs slightly decreased which resulted in higher incremental cost of €11,746 20 per QALY see Table 4.

21
The probability of sustained abstinence varies between 0.11 and 0.13 for men and between  of abstinence, the difference in avoided societal costs per person between HIT and LIT was 2 estimated as €779 for men and €502 for women and the correspondent difference in QALYs 3 gained was 0.0664 for men and 0.0462 for women. The incremental cost-effectiveness ration 4 (ICER) was negative for men (HIT was cost saving and entailed positive health outcomes in 5 comparison to LIT) but amounted to €2,391 for women, which is close to our base-case 6 analysis, see Table 4.

7
At all values of willingness-to-pay for a QALY, including zero, the HIT was more cost-8 effective than the LIT, see the probabilistic sensitivity analysis on the HIT quit rate in Figure   9 1. Main results

22
In this study, we performed a cost-effectiveness analysis using the long-term follow-up data 23 from a RCT of a high-and a low-intensity treatment programme (HIT and LIT) for smoking 24 cessation in a dental setting. We also validated the cost-effectiveness results of the previous 25 study based on short-term follow-up 16  conclusions of the previous cost-effectiveness analyses based on short-term follow-up data, 2 and we would recommend the use of the HIT programme as a cost-effective option for 3 smoking cessation.

Introduction
This is a technical report on an updated version of a model, originally developed in year 2004 , to enable systematic cost-effectiveness analyses of tobacco cessation interventions in Sweden. It aims to follow international and Swedish recommendations of cost-effectiveness analyses in health and medicine. The model holds a societal perspective, aiming to incorporate available disease-specific costs for all sectors of Swedish society. The updated model contains more recent data on societal costs, disease and death risks, and quality-of life-estimates, to enable estimates that reflects current Swedish conditions.
The model simulates the lifetime societal effects of quitting smoking on three diseases: lung cancer, chronic obstructive pulmonary disease (COPD), and cardiovascular disease (CVD) including coronary heart disease (CHD) and stroke. The model incorporates the smoking-related disease risks, the remaining disease risks after tobacco quitting, the death risks in the diseases and unrelated diseases, as well as the societal effects of the diseases. The societal effects include medical treatment costs, costs for institutional care, drug costs, costs for informal care and other costs for patients and relatives, and morbidity productivity costs, as well as loss of life-years and quality-adjusted life-years (QALYs).
This technical report contains a description of the model structure, of all the data sources used and of the assumptions made. For validation purposes, it also reports model estimates for some selected age-groups and more detailed outcomes and sensitivity analyses for one age-group, men and women aged 50 years at the start of the simulations. To investigate model uncertainty, univariate and multivariate sensitivity analyses are reported, as well as a probabilistic analysis. The model validity is discussed in the final section of the report.

Method The diseases
The model incorporates the three most smoking-related diseases: lung cancer, chronic obstructive pulmonary disease (COPD), and cardiovascular disease (CVD) including coronary heart disease (CHD) and stroke, see table 1. The model is restricted to the effects on the individual smoker/quitter, thus not incorporating any side-effects on other people.

The model
The stochastic model simulates the societal effects of smoking cessation on three smokingrelated diseases. It is constructed as a Markov-cycle tree model appropriate for microsimulations.
The Markov model is a health state-transition model  using probabilities for transitions between health states. These probabilities are the age-and gender-specific disease risks, conditional on smoking status and years since quitting, and age-, gender-and disease-specific death risks. The states are mutually exclusive and collectively exhaustive, and transitions between disease states are not allowed. The only exits from disease states are death, in the disease in question or in unrelated diseases, except for 5-year survivors in lung cancer which are assumed to recover to complete health. All other disease states are assumed to last life-long. See figure  1 for the state-transition diagram.
The Markov stages are one year-long, with no half-cycle correction. The starting point is the state healthy. The model covers the ages 15 to 95 years. The Markov-cycle tree has been created in Treeage Pro (Treeage Inc., 2015).  The Markov-cycle tree is run as a microsimulation with 10 000 repetitions. The simulation ends at death or age 95 years. The model is run separately for age and gender groups. The result of each simulation is expected value, with accompanying distributions. The two simulations, the continuing smoker and the quitter, are compared outside the model. The results are presented as expected value per individual, specific for gender, age and smoking status.

Material
The model is based on principles for cost-effectiveness analysis in health and medicine  and Swedish methodological recommendations (TLV, 2004). The model holds the societal perspective, aiming to incorporate diseasespecific costs for all sectors of Swedish society.
The model uses Swedish register data and secondary data from previously published scientific articles. The secondary data was found through searches in the database MEDLINE and the reference lists of retrieved articles, choosing the data that is considered most relevant to present-day Swedish circumstances and the target group. No metaanalysis nor other synthesis of data was performed.
All costs are expressed in year 2014 SEK (USD 1=SEK 6.86; Euro 1=SEK 9.10), converted if necessary by the Swedish CPI (consumer price index). The annual discount rate is 3% for both costs and health effects.

The risks
Disease risks All disease risks are annual age-and gender-specific excess incidence risk until the age of 95 years, see tables 2 to 5.
The COPD disease risk is taken from the Swedish population-based study Obstructive Lung Disease in Northern Sweden (OLIN), which was started in year 1985 . The risk is the reported average excess seven-year incidence among smokers in three age groups, of which the youngest was 45 years at baseline, see table 2. COPD was defined according to the spirometer GOLD definition.  The lung cancer disease risk is estimated from reports on lung cancer deaths until age 75 for smokers (15-24 cigarettes/day) and non-smokers, see table 3. The annual excess death risk is estimated by a quadratic function of the accumulated risk until age 75 years. The lung cancer death risk is assumed 0 until the age of 40 years, and assumed constant between ages 75 and 95. The disease risk is obtained by adjusting the annual death risk by the annual crude survival rate of lung cancer in Sweden for a similar time period as the Peto data, from .  Source:    Source:  The CHD and stroke disease risk estimates are based on the Framingham CVD risk function, see table 4 and tables 5-6. As the Framingham CHD risk function only calculates CHD events, the division of these events into the particular diseases are based on recent Swedish register data, see table 7. To avoid over-estimation of risks, the risk factors for CHD and stroke are evaluated at minimal-risk levels; 120 mmHg for systolic blood pressure (SBP), HDL-cholesterol (HDL) at 1.5 and cholesterol (Chol) at 4. Diabetes is set at 0, while the variable smoking (smok) is set at 1 for the smokers.  Death risks All death risks are age-and gender disease-specific conditional risks; in some cases estimated as fractions of the general population age-and gender-specific mortality risk, see tables 2 to 4, and in some cases based on Swedish register data, see tables 8 to 11.
The COPD death risk is estimated from the study Obstructive Lung Disease in Northern Sweden (OLIN), which reported the 20-year mortality in three age groups. Comparison with the general age-specific mortality risks revealed no excess risk of death among those younger than 58 years and older than 70 years, but a considerable increased risk among those aged 58-70 years at follow-up. The excess risk was estimated at on average around 5 times the age-and gender-specific general population death risk, see table 2.
The lung cancer death risk is based on survival data from the Swedish National Cancer Registry, see table 8. The death risks for year 3 and 4 after diagnosis are estimated by linear interpolation between years 2 to 5. Lung cancer survivors at 5 years are assumed recovered, and returned to the health state healthy.
The death risks from CHD and stroke are taken from Swedish registers, see tables 9 to 11, or published scientific reports, see table 5. The death risks for AMI, stroke and IHD are divided into risks the first year after the first event and the second and following years after first event.   The model also incorporates the possibility of dying in unrelated diseases. The death risk in the health state Healthy is the average 5-year age group-and gender-specific risk adjusted for all model disease deaths, the last column in table 12. In disease health states, the risk of dying in unrelated disease is the average 5-year age group-and gender-specific   risk adjusted for the deaths in each respective disease. For ages below 39 years the risk in the age group 35-39 years is used, and for ages 80-84 years the risk >79 years. For ages above 84 years, the general population age-and gender specific death risk is used for the unrelated death risk. As the lung cancer death risks are so high, the unrelated death risks for lung cancer among individuals aged above 84 years had to be adjusted, by deducting 0.05. For those aged below 85 years, the age-and gender-specific general population risk of death is only used for calculating some disease-specific death risks, see tables 2 and 4. The risk is taken from the Swedish national mortality statistics for the year 2014 (Statistics Sweden, 2015).

Changes in risk after quitting smoking
The excess disease risks for smokers are not eliminated immediately after quitting smoking. This "lead time" is 36 years for lung cancer, 16 years for CHD, and 11 years for stroke, while for COPD some excess risk remain life-long, see heading effect of quitting in tables 2 to 4. The disease risks after quitting are constructed by adjusting the smokers' risks by the remaining risk. The remaining risk is modelled as fractions of risk, given the number of years since quitting. The annual remaining risks are estimated by linear interpolation. The effects on the risk for CHD and stroke are modelled on the dummy variable smoking, adjusting the value of 1 by the remaining risk fraction.

The societal costs
The model is reflecting the societal perspective, including disease-related costs for all sectors of the Swedish society. The costs included are medical treatment costs, costs for institutional care and technical aids, pharmaceutical costs, informal care and other patient and relatives' costs, and morbidity productivity costs.
Most of the data on societal costs are taken from Swedish studies published during the 2010s. Data reported as distributions, i.e. with the Gamma parameters for costs, or bootstrapped 95 percent confidence interval were preferred and used in the model to 13 enable stochastic estimation. If data was reported as mean and standard deviation, the Gamma distribution was simulated employing the Treeage function. In one case, data was reported as fraction of patients consuming a specific resource, which was used for sampling within the model. Otherwise the reported point estimate, usually the average cost across the patient group, was used. If no Swedish data on a cost item was found, the cost was taken from studies reporting data from settings assumed similar to the Swedish.
All costs are reported in SEK year 2014 (USD 1=SEK 6.86; Euro 1=SEK 9.10), adjusted when necessary with the Swedish CPI. To adjust reported Gamma distributed parameters to the price level, only the second parameter, i.e. the scale parameter, was adjusted.

Medical treatment costs
Recent Swedish estimates on medical treatment costs were possible to obtain for all model diseases, see table 13. The costs are paid by the regional healthcare authorities.

Institutional care and technical aids costs
These costs include rehabilitation, terminal care, old age homes, support for individuals living at home, transportation and technical aids. In Sweden, institutional care and technical aids used by patients in their homes are the responsibility of the local authorities (municipalites, in Swedish: kommuner). The costs are not fully represented for any disease, see table 14. Estimates are not available for lung cancer and the only available costs for IHD are outdated, so the institutional care costs are probably underestimated.

Pharmaceutical costs
Costs for pharmaceuticals in Sweden ought to be divided between the county councils and the patients, as patients pay a considerable share in co-payment. This is however not possible, given the data available. Table 15 therefore presents the drug costs to the regional healthcare authorities. The costs of pharmaceuticals dispensed during hospital stays are included in the medical treatment costs.

Informal care and other patient and relatives' costs
These costs include the value of care given to patients by relatives and other costs for patients or relatives, such as time, co-payments paid for health care and drugs as well as costs for transportation, modifications at home etc. Complete estimates could not be obtained for any disease, see table 16, except IHD which however might be outdated. Informal care in present-day Sweden probably constitute a sizeable part of total societal costs.

Productivity costs
The productivity costs only value the lost production because of morbidity before the age of 66 years, not mortality. The productivity costs for lung cancer is simulated within the model, via sampling from the fraction of patients on sick leave and combined with ageand gender-specific average monthly wages, including 40% employer taxes. Remaining data is taken from the literature, see table 17, and most estimates are recent. The costs are valued by the human capital method, and thus only include losses in salaried work before the official age of retirement.

Life years lost
The number of life years lost (YLS) are calculated until the age of 95 years, and only for individuals dead in the modelled diseases. Life years lost are presented both discounted 3% and undiscounted.

QALYs
The number of quality-adjusted life years (QALYs) are calculated during healthy years and years spent diseased, until death or the age of 95 years.
The QoL weights used during healthy years are mean age group-and gender-specific population weights, see table 18. The data is somewhat dated, but it is the only general population QoL weights available in Sweden. The QoL of the age group 20-29 years is used  also for younger ages, and the QoL of the age group 80-88 years is used for those aged 89-95 years. This last assumption is probably an overestimate.
The disease-specific QoL used in the health states are all, except one, modelled as decrements from the average population age-group and gender-specific QoL, see table 19. For lung cancer no data was available on the marginal effect of the disease on the population average QoL, so a fixed value over the ages and genders had to be used.

Sensitivity analyses
Several univariate and multivariate sensitivity analyses have been performed. Analyses on some methodological issues, as well as a probabilistic sensitivity analysis, have also been performed. The analyses are reported for men and women aged 50 years.
To give another measure of the uncertainty surrounding the cost-effectiveness ratio, the 95% confidence interval for the difference between smokers and quitters is reported, calculated from the standard deviation of outcomes.

Probabilistic analysis
A bootstrap sampling was performed using the smoker and quitter Monte Carlo simulations of 10 000 runs. A sample of 1 000 from each simulation was drawn, with replacement, performed in Microsoft Excel. The mean of the difference in costs and QALYs between smokers and quitters was then calculated. This was replicated 1 000 times. The bootstrap is represented as a scatterplot in the cost-effectiveness plane.

Results
In this chapter, the model estimates of QALYs, YLS and societal costs are presented for men and women in some selected ages, mainly for validation purposes. More detailed simulation outcomes as well as the results of the sensitivity analyses are presented for men and women at age 50 years. Model estimates can be obtained for men and women for all ages between 15 and 95 years.

The model estimates
In table 20 the simulation results for QALYs (quality-adjusted life-years) experienced until the age of 95 years are presented, for the selected ages 15, 30, 50 and 70 years at the start of the simulations. As can be expected, the number of QALYs are highest in the younger age groups, and somewhat higher for women in most age groups. In the selected age groups, the differences between smokers and quitters are at a maximum at age 30; 0.68 for females and 0.81 for males. The confidence intervals, calculated via the mean and standard deviation (sd) from the 10 000 model runs, indicate that there are differences in QALYs between smokers and quitters.
The YLS (life-years saved) lost before the age of 95 years are presented in tables 21 and 22, discounted 3% and undiscounted. The differences in discounted YLS between smokers and quitters are somewhat higher than the differences in QALYs. The undiscounted YLS in table 22 show the number of years that smokers and quitters are expected to lose before the age of 95 years. For the ages 15, 30, and 50 the number of lost life-years is estimated at around 6 years for women smokers and 9 years for men, implying that the female smokers are estimated to live until age 89 and the male until age 86. In the oldest age group presented here, age 70, the number of lost life-years are only 1-2 years. The quitters are estimated to lose considerably fewer life-years; 1-4 years for the women and 3-5 years for the men. As expected, the difference between smokers and quitters diminish with age, with a maximum at around 5 years for the females and around 6 years for the males at age 15.
The societal costs estimated for the smokers and quitters for the selected age groups are presented in table 23. The highest costs are found for age 50; 200 000 SEK and 250 000 SEK for the smokers and 130 000 and 170 000 for the quitters, in both cases higher among the men. The highest difference between smokers and quitters is however found at age 30, with a difference of 100 000 among the females and 120 000 among the males. The difference among the eldest, at age 70, is around 20 000 SEK. These cost differences reflect the amount that tobacco cessation interventions could spend on achieving one quitter and still be cost-saving.

Selected model outcomes
The underlying estimated disease outcome is presented in figures 2 and 3, for the age 50 years. For both women and men, there is a marked decrease for quitters in the number of diseased and dead in the model diseases, which is somewhat offset by an increase in the number of deaths in unrelated diseases. The number of diseased and deaths are higher for    men, mainly originating from CHD. The model disease with the highest smoking-related incidence is COPD, for both genders. The increase in unrelated deaths for the quitters is an example of competing risks, which decreases the difference in life-years and QALYs between smokers and quitters. Table 24 and 25 shows the full model simulation results of the societal cost savings because of tobacco quitting at age 50 years. For women, the highest estimated savings are found in lung cancer, COPD and stroke at around 15-20 000 SEK per quitter. For men the cost savings because of lung cancer are considerable higher, at around 35 000, due to the higher incidence among the men. The cost item with the largest cost savings are medical treatment costs for both genders, at around 30 000 SEK. Most of the difference in savings between men and women originate from the productivity costs, possibly reflecting disease onset at younger ages among men. Note that a cost saving of zero means that no cost is being modelled, as cost data was lacking.

Sensitivity analyses
The results of the sensitivity analyses are presented on women and men at starting age 50 years. Figure 4 shows the results for women and figure 5 for men.
All analyses show a similar pattern between men and women, and also similar ranges. The univariate sensitivity analyses on the model parameters, analyses A to E, result in small changes in costs and QALYs. Also the multivariate analyses F and G, which are constructed as scenarios that allow the risk parameters to vary consistently upwards or downwards, and along with the costs in analysis G, show moderate changes from the base case estimates. The methodological choices have a more pronounced effect, as the largest difference in QALYs is achieved by varying the discount rate (analysis H) between 0 and 5%, which also affects the costs substantially. The two analyses that reflect the choices of which costs to include in the estimates, analysis I that reflects the UK NICE health care and social services perspective and analysis J that only include Swedish data published since the year 2005, both decrease the cost differences between smokers and quitters.  The scatter plot of the bootstrap analysis based on the microsimulation results for women and men aged 50 are shown in figures 6 and 7. The uncertainty is higher for the men, as the plots are more scattered. All plots are however situated in the cost decrease and QALY increase quadrant, with costs below -20 000 SEK and QALYs over 0.2.

Discussion: Model validity
The discussion of the model validity is structured around four aspects as proposed by : the structure of the model, the inputs to the model, the results of the model and the value of the model to the decision-maker.
The structure of the model The structure of the model is a Markov model constructed for microsimulations, on the three most smoking-related disease groups; lung cancer, COPD, and CVD including stroke and CHD. The present updated version of the model includes one less CHD disease compared to the first version of the model, as unrecognized acute myocardial infarction now is included in the IHD disease, mainly because the disease definition is rarely used nowadays. Choosing only three disease groups is a clear simplification as smoking is known to cause hundreds of different diseases. The effects from smoking, and thus quitting, are furthermore confined to the individuals themselves; no side-effects on other individuals such as environmental tobacco smoke or smoking uptake are included. These two features leads to an underestimate of the true effects of tobacco quitting.
The same disease-specific approach has been taken by most other tobacco cessation models , even though some of them include more diseases, such as asthma. Another approach would be to use the overall differences in mortality between current, former, and never-smokers taken from large US studies, as some early tobacco cessation models did Tengs et al, 2001). In order not to overestimate the effects of quitting tobacco, we chose to model the smoking-related risk for certain diseases instead, as it is improbable that all differences in mortality and morbidity between smokers and former smokers are due to the smoking habit (Doll et al, 1994).
The model aims to reflect disease onset related to smoking tobacco. As disease in all the three disease groups included in model may be caused by other factors than smoking only the excess risks for smokers are modelled. For the diseases lung cancer and COPD this implies that the risk for smokers found in epidemiological studies is adjusted by the risk found for non-smokers. For the disease group CHD and stroke, where a large fraction of disease onset is caused by other factors than smoking, this adjustment for smokers´ excess risk was performed by setting the other risk factors in the risk function at minimal risk levels. This is an underestimate, as the risk factor levels among smokers can be expected to be at least as elevated as among the general population. The underestimate is aggravated by the fact that the functional form of the risk function results in a multiplier effect of the risk factors.
The present version of the model includes seven health states: lung cancer, COPD, stroke, and CHD divided into four diseases. This is a clear simplification, as the costs and QoL can be expected to vary considerable between patients with different severity levels within the diseases. This is particularly true for COPD which is a chronic progressive disease, i.e. the diseased get more severely ill over time. However, a model with 7 health states with accompanying disease-specific death risks, costs and QoL weights is fairly complex as well as data-demanding. For the purposes of this study´s model, the division of diseases into severity levels was not deemed necessary.
An obvious problem with the model, inherent in all Markov models, are the mutually exclusive health states; any individual can only contract one disease, and once diseased the individual never recovers (apart from the very rare 5 year survivors in lung cancer). This feature implies both an overestimate and an underestimate of the true effects. The underestimate stems from the fact that co-morbidity is very common, especially among the individuals with the chronic diseases COPD, CHD, and stroke. The overestimate of costs and effects arise as individuals stay in the health states until death. If the costs and outcomes associated with the health states are taken from severely ill individuals, then these become grossly overestimated. This overestimate is partly offset by the use of separate costs for the first and subsequent years, for all societal costs due to AMI and stroke. In order not to overestimate the numbers of years spent in disease states, the possibility of dying in unrelated diseases is present in all health states. This feature is also included in the CHD Policy Model (Weinstein et al, 1987).
Most tobacco cessation models are built for cohort estimation , but this model is constructed for individual-level estimation using the microsimulation methodology. As the data available admitted a microsimulation structure, e.g. the risk functions, the methodology was chosen as the advantages to model and to obtain a richer data set with results that reflect the heterogeneity of outcomes between individuals was deemed to offset the disadvantages of calculation burden. The use of the software Treeage also facilitates the use of microsimulation. Age-and gender-specific estimates can thus be obtained from the model, between ages 15 and 95 years.
The model stages are one-year long, which seems accurate given the risk estimates and the long time horizon of the model. The reason for the model maximum age of 95 years is the lack of risk estimates for older ages. Some extrapolations of risk estimates to the age of 95 years indeed resulted problematic, as some disease-specific death risks expressed as multipliers of the average age-specific death risk resulted in risks above 1. Further extrapolations beyond the age of 95 years were deemed unnecessary, as most of the relevant differences between smokers and quitters would have arisen by that age.

The inputs of the model
The second aspect of model validity is the inputs of the model. The model contains a large number of data taken from different sources. This is of course a threat to the internal validity of the model, shared with most models. However, the data have been chosen to reflect current Swedish circumstances. The current updated version of the model has exchanged almost all cost data, if more recent estimates were available, and all death risks to recent Swedish register data. As the number of studies on any particular data items are few, no meta-analysis or any other synthesis of data was carried out. The disease risks are of course are pivotal for the result. The lung cancer disease risks are probably the best that can be obtained, from a large epidemiological study . The risk equation used for CHD and stroke is taken from the Framingham studies, and even though there are more recent risk scores developed from the study (D'Agostini et al, 2008), the  risk functions are still frequently employed. The disease COPD has been the subject of a large long term epidemiological study in Sweden, The Obstructive Lung Disease in Northern Sweden (OLIN) , which is thus the most relevant data source for the model.
In the model, there is an increased risk for a smoking-related disease remaining for some years after the tobacco cessation, in accordance with epidemiological evidence (Surgeon General, 1990;. The feature is also considered a marker of high quality tobacco cessation models .
The majority of the cost data are taken from Swedish studies published during the 2010s. To take fully advantage of the microsimulation structure and to obtain stochastic estimates, the preferred data sources were the ones reported as distributions, i.e. as Gamma parameters or bootstrapped 95 percent confidence intervals. If no Swedish data was found, an international estimate was instead used in order to seek to represent the full societal costs. However, apart from certain cost items and for some of the diseases, the lack of data results in considerable underestimates of the true societal costs. This is particularly true in the cases of the costs for care, both institutional and informal. The institutional care could amount to considerable costs, exemplified by the costs for stroke and AMI patients, see table 14. In particular for lung cancer the lack of data results in considerable underestimates of the true disease-related costs. This is why the possible overestimate of the informal care for the disease, obtained from an Italian study, probably does not bias the overall result. To investigate the issue, one sensitivity analysis only included recent Swedish data. The analysis lead to decreases in cost savings for quitters aged 50 years of around 30%.
The QoL estimates are constructed as disease-specific decrements from the average ageand gender-specific QoL, except for lung cancer for which no QoL decrement could be found . The average population age-and gender-specific QoL weights, which are certainly not 1, are also used during healthy years for the base case estimates. This means that the model assumes that an individual that avoids the smoking-related diseases is not having perfect health, but the health of an average Swede at the same age, as recommended .
The stated purpose of the model is to reflect the societal perspective, which for Sweden includes the morbidity productivity costs, but not the productivity costs resulting from mortality. All the model data on productivity costs value them according to the human capital approach for individuals under the age of 65, the customary Swedish age of retirement.
A full societal perspective might also include other aspects, considering that this is a model on individuals that are participating in an intervention that aims to change their lifestyle. The previous version of the tobacco cessation model, version 1 , reported 28 sensitivity analyses that modelled some effects on the tobacco quitters, by including savings from cigarette purchases and a decreased QoL because of withdrawal effects during the first year. When that analysis was applied to an intervention, a decreased QoL during the first year was also deducted for the smokers that failed to quit, as the failure to achieve a personal goal might to lead to a decrease in QoL.
The results of the model The third aspect of model validity is the results of the model, e.g. a comparison with reality or with other study results. A direct comparison with reality is not possible, since the model covers the ages 15-95 years, with a follow-up time of 80 years for the youngest age group.
The model estimates that around 60% of the women and 70% of the men aged 50 at the start of the simulations will contract one of the modelled diseases, and that around 50% of those will die in the diseases before the age of 95 years. The disease risks for the quitters at age 50 are not eliminated; 30-40% of them will still contract the smoking-related diseases because of remaining disease risks after quitting. As expected, the unrelated deaths increase among the quitters, in sum leading to an increase in YLS (undiscounted) of 2-3 years for those quitting at age 50, compared with continuing smokers. The increases in QALYs (discounted 3%) are smaller because of less-than-perfect health among those aged 50 years and above; 0.61 for women and 0.77 for men. The disease outcomes are fairly similar to the estimates from the previous versions of the model, but because of decreased death risks, the outcomes in terms of YLS and QALYs are considerably higher. The 2004 version of the model estimated an increased YLS of 0.93 and of 1.66 for women and men aged 50-54 years, and QALY gains of 0.36 and 0.71, respectively. The differences are due to the longer time perspective of the present version, 95 years versus 85 years, and the somewhat decreased case-fatality risk (i.e. the mortality risk among those with disease) because of improvements in medical technologies during the past decade.
Apart from increases in health, the societal cost savings because of quitting smoking are considerable. For men, the cost savings amount to around 100 000 SEK for quitters aged between 15 and 50 years, and around 70-90 000 SEK for women. Even in the age group 70 years there are estimated cost savings of around 20 000 SEK per individual quitter. This implies that substantial funds could be invested in smoking cessation interventions, and the interventions would still be cost-effective, or even cost-saving. The cost savings in the present model are considerably higher than those of the previous model, in part due to changes in price year.
Comparisons of model estimates with other models' are difficult to perform, as the time horizon, costs included, jurisdiction, and the diseases included differ. Among the recently reported model estimates , there are two Australian models. The model developed within ACE  report estimates of life-years saved that are considerable higher than the present model's; 5.7 years for men and 6.6 years for women in age group 50-54 years. That model time horizon is however 100 year, but it is unlikely There have been two, to my knowledge, reports of tobacco cessation model estimates for Sweden, one using the Benesco model  and one using an extended version of the HECOS model . Comparison with those model estimates are unfortunately not possible, due to lack of reporting detail. However, estimates from the previous version of this model were fairly consistent with the HECOS model estimates  for Sweden, available at the time .

The value of the model to the decision-maker
The fourth aspect of validity is the value of the model to the decision-maker. There are several models on tobacco cessation that conforms to international recommendations on how to perform cost-effectiveness analyses . This model however reflect Swedish circumstances, with Swedish cost and QoL data, why the model might be useful for Swedish decision-makers.
We hope that the model will be used to perform economic evaluations of a range of tobacco cessation interventions. For tobacco prevention interventions, i.e. prevention of initiation of smoking, another model version, version 2, has been constructed and is available for analyses. The use of these models will in time enable incremental and marginal calculations of the cost-effectiveness of different tobacco interventions and their components and suitable target groups. The basis for decisions on which tobacco cessation and prevention interventions to implement will then be more comprehensive.
Another frequent use of models is to forecast future events. This model is not suitable for estimating what the costs of smoking will be in the future. The reason is that the model does not incorporate any adjustments of possible future developments. The risk of smoking is based on studies with follow-up periods of sometimes 30 years, which means that the risks are reflecting the smoking behaviour among smokers 30 years ago. The changes in cigarette content and in the frequency of smoking might lead to changes in disease risk in the future. Also the costs for the smoking-related diseases might change in the future, because of changes in health care technology. Another example would be the value of the morbidity productivity costs, as well as informal care, as wages and productivity often are expected to increase in the future.
Nevertheless, the model actually forecasts what the costs for smokers and quitters will be in 80 years' time, for the youngest age group. That implies that we know that the model forecasts will be wrong, but it is of minor significance as the model is constructed to be used for comparisons between two groups, smokers and quitters, thus eliminating some

The uncertainty
Another aspect of model validity is the uncertainty surrounding the model estimates.
The univariate sensitivity analyses on the model parameters (analyses A-F in figures 4 and 5 for men and women aged 50) show minor deviations from the base case result, while the multivariate analysis on costs and risks combined (analysis G) affects in particular the cost estimates. The methodological choices affect the results to a greater extent, with the discount rate (H) heavily influencing the QALYs and the more restricted perspective (I) decreasing the cost-savings. The multivariate analysis that only include higher-quality data (J) also imply decreases in the cost differences between smokers and quitters, but the difference remains substantial; around 50 000 SEK for females aged 50 years and 60 000 SEK for men, respectively. The overall conclusion from the parameter sensitivity analyses is that the QALY gains are at least 0.35 and 0.40 and the cost savings at least SEK 35 000, for female and male quitters aged 50, respectively.
The probabilistic analysis shows no uncertainty whether quitting tobacco leads to costsavings and increases in QALYs, as all bootstraps are placed in the southeast quadrant of the cost-effectiveness plane. The bootstrap results exhibit a mixture of first and second order uncertainty, as it reflects both the probabilistic structure of the Markov model and the simulation of some parameter values .
Another measure of uncertainty is the confidence intervals around the estimated mean differences, reported in tables 20-23. However, that measure is not fully appropriate as the large sample sizes of the Monte Carlo simulation (10 000 runs) diminishes the standard error of the mean .
The structural uncertainty of the model, i.e. whether the results would be different if the model would have been constructed in another way, have not been studied. Alternatives to the chosen model structure could have been deterministic or discreet event simulations, more or less health states, other functional forms of risk functions, and other subgroups than men and women and five-year age-groups model results. The flaw is however shared with most tobacco quitting models .

Checking for technical errors
The model contains a large number of trackers, i.e. variables that count events, to enable checking for technical errors. Tentative runs were executed after the introduction of every new variable, with cost items undiscounted, and the simulation results examined manually. Thus, the model has been thoroughly checked for technical errors.

Conclusions
The aim of this study is to develop a model predicting health and economic consequences of smoking cessation, to be used for cost-effectiveness analyses of smoking cessation interventions. The updated model strives to incorporate data that is recent, accurate and appropriate for Sweden in year 2015. The model also adhere to Swedish recommendations on how to perform cost-effectiveness analyses within the health care sector. Data is however lacking to completely fulfil these requirements. Many model parameters are based on very few studies. Some information just does not exist, at least not accessible to us.
These are issues shared with most model, however. The purpose of modelling is to assemble the most accurate information at a point of time, to enable decision-making at that particular point of time. This is in accordance with one of the fundamentals of economics: decision-making under uncertainty, which implies that decisions have to be made even if there is no full information. We hope that the model will be applied to a range of different tobacco cessation interventions, which in time will enable a more comprehensive basis for decision-making. For each intervention, report mean values for the main categories of estimated costs and outcomes of interest, as well as mean differences between the comparator groups. If applicable, report incremental cost-effectiveness ratios.
Page 15, Table 3 Page 17, Table 4 20a Single study-based economic evaluation: Describe the effects of sampling uncertainty for the estimated incremental cost and incremental effectiveness parameters, together with the impact of methodological assumptions (such as discount rate, study perspective).

Not applicable Characterising uncertainty 20b
Model-based economic evaluation: Describe the effects on the results of uncertainty for all input parameters, and uncertainty related to the structure of the model and assumptions.

21
If applicable, report differences in costs, outcomes, or cost-effectiveness that can be explained by variations between subgroups of patients with different baseline characteristics or other observed variability in effects that are not reducible by more information.

Cost-effectiveness of a high-vs a low-intensity smoking
3 cessation intervention in a dental setting: long-term 4 follow up. 5 6 7 Inna Feldman  Objectives. The aim of this study was to conduct a cost-effectiveness analysis of a high-and 5 a low-intensity smoking cessation treatment programme (HIT and LIT) using long-term 6 follow-up effectiveness data and to validate the cost-effectiveness results based on short-term 7 follow-up. willing to spend at least €4,000/QALY for tobacco cessation treatment.   1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  Smoking is likely to remain the single most important preventable health risk in the world. 5 Despite continuously declining prevalence in recent decades, one in ten adults in Sweden still 6 smokes daily 1 . Cigarette smoking contributes to 7.5% of the burden of disease in Sweden 2   7 and was estimated to account for approximately €3,000,000 (31.5 billion Swedish krona,  approaches of different level of intensity for cancer patients 11 . The results suggested that the 2 higher intensive interventions are preferable from health economics point of view, but all 3 those evaluations were based on 6-or 12-months follow-up, long-term follow-ups are scarce 4 in randomised controlled trials.

5
The effects of smoking on health occur during many years because current smoking 6 influences future health risks; similarly, a smoking cessation today will cause smoking related 7 health risks to tail off gradually. Thus, in order to estimate cost-effectiveness of smoking year follow-up). The conclusion was that high-intensity treatment support is the preferred 7 option if the decision-makers' willingness-to-pay exceeds €5,100 (50,000 SEK) per QALY.

8
The base-case scenario of the analysis assumed a sustained abstinence for the quitters 16 . The 9 long-term follow-up of the programmes was performed five to eight years later 17 . In this 10 study, we used a unique opportunity to compare cost-effectiveness analyses of a high-and a 11 low-intensity smoking cessation intervention in a dental setting, using data from short-term

14
We set out to: 1) perform a cost-effectiveness analysis of a high-and a low-intensity smoking 15 cessation programme in a dental setting using long-term (5-8 years) follow-up data and 2) 16 compare the cost-effectiveness results with the previous study based on short-term (1-year) 17 follow-up. In the smoking cessation intervention study 18 , between August 2003 andFebruary 2005, 300 25 adult smokers recruited via direct inquiry or advertising in dental or general health care were 26 offered smoking cessation support performed in a dental setting. Inclusion criteria were daily 27 smokers over 20 years of age, while exclusion criteria were reading difficulties and problems with Swedish language. The participants were randomly assigned to two interventions; one 2 received high-intensity and one low-intensity treatment support.

3
The high-intensity smoking cessation treatment, the HIT programme, comprised eight 4 individual sessions, of in total 3.5 hr over a period of 4 months, and was based on behaviour 5 therapy, coaching and pharmacological advice. The low-intensity smoking cessation 6 treatment, the LIT programme, comprised one counselling session, of up to 45 min, 7 introducing a conventional self-help programme running over 8 weeks. Both programmes 8 were free of charge.

9
The participants answered a baseline questionnaire and a short-term (one year after the 10 planned smoking cessation date) follow-up questionnaire. Demographic characteristics such 11 as gender, age and education level were also collected. The effectiveness of the trial was 12 reported elsewhere 18 . The analysis concluded that the more extensive and expensive HIT 13 programme was more effective and cost-effective, in terms of proportion of smokers who 14 were still smoke-free after one year 16 18 . The long-term follow-up was performed 5-8 years 15 after the planned smoking cessation date. The effectiveness analysis showed that the 16 difference in outcome between the HIT and LIT programmes remained relatively constant and 17 significant in favour of HIT, and that abstinence at 1-year follow-up was a good predictor for 18 long-term abstinence 17 . All analyses were done using the "intention to treat" approach where 19 non-responders were considered as smokers. Mortality and morbidity data for the participants 20 were not collected either by questionnaire or through the registers. The original study, as well 21 as the long-term follow-up, was approved by the ethical committee at Uppsala University

23
The mean age of the participants was 49 years, and 78% were women. Short-term follow-up

15
Intervention effectiveness 16 17 For CEA short-term, we used 6-month continuous abstinence at 1-year follow-up reported by 18 41 participants (14 from HIT and 27 from LIT). For CEA long-term, we used sustained 19 abstinence at 5-8 years reported by 24 participants (17 from HIT and 7 from LIT), see Table   20 1. Both measures were statistically significant different between the treatment programs. In 21 order to generalize the long-term effectiveness of our study, we performed a logistic 22 regression analysis to calculate the probability of sustained abstinence depending on 23 programme (HIT vs LIT), participant's gender and age, see Table 2.  The type of the programme (HIT vs LIT) was significantly associated with sustained 6 abstinence while gender and age were not. The regression equation [1] demonstrates 7 dependence between "abstinence" (1 -abstinence, 0 -no abstinence) and "programme" (1 -8 HIT, 0 -LIT), "gender" (1 -male, 0 -female) and "age" (19-71):   1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59    decreased by 25%, the disease risks by 50%, the death risks by 10%, and the risk fractions 2 after quitting by 0.1. This low cost/low risk analysis led to substantial decreases in cost and 3 QALY differences between quitters and smokers. This sensitivity analysis was applied to 4 compare costs and effects between HIT and LIT, to validate the results of the CEA long-term. 5 To increase the generalizability of the cost-effectiveness results, we have also applied the 6 probabilities of long-term abstinence depending on programme (HIT vs LIT), participant's 7 gender and age on the modelling results. We estimated the avoided social costs and gained 8 QALYs for a random quitter from our sample and then adjusted the results to the probability 9 to quit (Abstinence), calculated in [1]. Cost-effectiveness was estimated for men and women 10 separately.

11
Further, a probabilistic sensitivity analysis (PSA) was conducted, based on the uncertainty of 12 the difference in sustained abstinent participants in the two programmes. The effectiveness of

17 18
The CEA short-term indicated that HIT led to additional avoided societal costs of €82,253 and The more costly HIT programme led to a higher number of 6-month continuous abstinent 26 participants at 1-year follow-up (CEA short-term) as well as higher number of sustained 27 abstinent participants at 5-8 year follow-up (CEA long-term), which translates into larger 28 costs avoided and health gains than LIT, see Table 4. However, the difference in intervention 29 costs were not fully balanced by the societal costs avoided, so HIT implied an incremental net 30 cost of about €6,832 in CEA short-term and €21,619 in CEA long-term, compared with LIT.

31
HIT was estimated to lead to more QALYs, so the incremental cost per QALY of HIT 32 compared with LIT amounted to €918 for CEA short-term and €3,786 for CEA long-term, 33 which is considered to be very cost-effective in Sweden 20 . The incremental analysis favours 34 the more costly HIT, if decision-makers are willing to spend at least €4,000/QALY for    Sensitivity analyses 16 17 The most conservative sensitivity analysis, a multivariate low cost/low risk analysis, was 18 applied to CEA long-term. This analysis led to substantial decreases in avoided social costs 19 and QALY gains for both HIT and LIT. At the same time, the incremental costs increased and 20 incremental QALYs slightly decreased which resulted in higher incremental cost of €11,746 21 per QALY see Table 4.

22
The probability of sustained abstinence varies between 0.11 and 0.13 for men and between of abstinence, the difference in avoided societal costs per person between HIT and LIT was 3 estimated as €779 for men and €502 for women and the correspondent difference in QALYs 4 gained was 0.0664 for men and 0.0462 for women. The incremental cost-effectiveness ration 5 (ICER) was negative for men (HIT was cost saving and entailed positive health outcomes in 6 comparison to LIT) but amounted to €2,391 for women, which is close to our base-case 7 analysis, see Table 4.

8
At all values of willingness-to-pay for a QALY, including zero, the HIT was more cost-9 effective than the LIT, see the probabilistic sensitivity analysis on the HIT quit rate in Figure   10 1. Main results

23
In this study, we performed a cost-effectiveness analysis using the long-term follow-up data 24 from a RCT of a high-and a low-intensity treatment programme (HIT and LIT) for smoking 25 cessation in a dental setting. We also validated the cost-effectiveness results of the previous 26 study based on short-term follow-up 16 . HIT was more effective in getting participants to quit 27 smoking and to keep sustained abstinent, resulted in higher societal costs avoided and more 28 QALYs gained among both men and women, compared with LIT and thus can be considered 29 cost-effective. The incremental cost-effectiveness ratios (ICERs) were €918 and €3,786 using 30 short-and long-term effectiveness, respectively, which are below the Swedish willingness-to-31 pay threshold of €50,000 per QALY 25 , thus, indicating that the resource intensive HIT was cost-effective compared to the less resource demanding LIT. The results also confirm the 2 conclusions of the previous cost-effectiveness analysis based on short-term follow-up data 3 and suggest its sustainability. We would recommend the use of the HIT programme as a cost-4 effective option for smoking cessation. 5 Notably, the usage of both the HIT and LIT programmes is not limited to dental settings and 6 can be implemented in other healthcare sectors and delivered by trained nurses instead of 7 dental hygienists. Since the salaries of registered nurses and dental hygienists are comparable, 8 the conclusion of high cost-effectiveness of the HIT programme remains.

10
However, although HIT was shown to be cost-effective in comparison with LIT, the 11 sensitivity analysis using the probability of abstinence suggested that HIT dominated over 12 LIT for men (saved societal costs and generated more QALYs). In our sample the majority of 13 study participants were women, that is why the results of the sensitivity analysis for women 14 was very close to our base-case analysis.  LIT contributed to a low ICER for the long-term cost-effectiveness analyses.

5
The effects of smoking cessation are certainly underestimated in the model estimates since 6 only three disease groups including lung cancer are modelled and no effects of passive 7 smoking are included, but smoking is causally related to at least 15 other types of cancer 33 . 8 In addition, quitting smoking reduced the rate of incidence diabetes to that of non-smokers 9 after five years in women and after 10 years in men 27 . We could not find any cost-effectiveness analyses based on more than 1-year follow-up, and 5 therefore we compared our results with other studies estimating cost-effectiveness of 6 interventions with different level of intensity using 6-or 12-month follow-up. Thus, a cost-  In conclusion, the more costly HIT smoking cessation programme is an economically 4 attractive option when compared to the LIT programme over a broad range of assumptions, 5 using shot-and long-term outcomes. Cost-effectiveness analysis favours the more costly HIT 6 if decision-makers are willing to spend at least €4,000/QALY for tobacco cessation treatment.

Introduction
This is a technical report on an updated version of a model, originally developed in year 2004 , to enable systematic cost-effectiveness analyses of tobacco cessation interventions in Sweden. It aims to follow international and Swedish recommendations of cost-effectiveness analyses in health and medicine. The model holds a societal perspective, aiming to incorporate available disease-specific costs for all sectors of Swedish society. The updated model contains more recent data on societal costs, disease and death risks, and quality-of life-estimates, to enable estimates that reflects current Swedish conditions. The model simulates the lifetime societal effects of quitting smoking on three diseases: lung cancer, chronic obstructive pulmonary disease (COPD), and cardiovascular disease (CVD) including coronary heart disease (CHD) and stroke. The model incorporates the smoking-related disease risks, the remaining disease risks after tobacco quitting, the death risks in the diseases and unrelated diseases, as well as the societal effects of the diseases. The societal effects include medical treatment costs, costs for institutional care, drug costs, costs for informal care and other costs for patients and relatives, and morbidity productivity costs, as well as loss of life-years and quality-adjusted life-years (QALYs).
This technical report contains a description of the model structure, of all the data sources used and of the assumptions made. For validation purposes, it also reports model estimates for some selected age-groups and more detailed outcomes and sensitivity analyses for one age-group, men and women aged 50 years at the start of the simulations. To investigate model uncertainty, univariate and multivariate sensitivity analyses are reported, as well as a probabilistic analysis. The model validity is discussed in the final section of the report.

Method The diseases
The model incorporates the three most smoking-related diseases: lung cancer, chronic obstructive pulmonary disease (COPD), and cardiovascular disease (CVD) including coronary heart disease (CHD) and stroke, see table 1. The model is restricted to the effects on the individual smoker/quitter, thus not incorporating any side-effects on other people.

The model
The stochastic model simulates the societal effects of smoking cessation on three smokingrelated diseases. It is constructed as a Markov-cycle tree model appropriate for microsimulations.
The Markov model is a health state-transition model  using probabilities for transitions between health states. These probabilities are the age-and gender-specific disease risks, conditional on smoking status and years since quitting, and age-, gender-and disease-specific death risks. The states are mutually exclusive and collectively exhaustive, and transitions between disease states are not allowed. The only exits from disease states are death, in the disease in question or in unrelated diseases, except for 5-year survivors in lung cancer which are assumed to recover to complete health. All other disease states are assumed to last life-long. See figure  1 for the state-transition diagram.
The Markov stages are one year-long, with no half-cycle correction. The starting point is the state healthy. The model covers the ages 15 to 95 years. The Markov-cycle tree has been created in Treeage Pro (Treeage Inc., 2015).  The Markov-cycle tree is run as a microsimulation with 10 000 repetitions. The simulation ends at death or age 95 years. The model is run separately for age and gender groups. The result of each simulation is expected value, with accompanying distributions. The two simulations, the continuing smoker and the quitter, are compared outside the model. The results are presented as expected value per individual, specific for gender, age and smoking status.

Material
The model is based on principles for cost-effectiveness analysis in health and medicine  and Swedish methodological recommendations (TLV, 2004). The model holds the societal perspective, aiming to incorporate diseasespecific costs for all sectors of Swedish society.
The model uses Swedish register data and secondary data from previously published scientific articles. The secondary data was found through searches in the database MEDLINE and the reference lists of retrieved articles, choosing the data that is considered most relevant to present-day Swedish circumstances and the target group. No metaanalysis nor other synthesis of data was performed.
All costs are expressed in year 2014 SEK (USD 1=SEK 6.86; Euro 1=SEK 9.10), converted if necessary by the Swedish CPI (consumer price index). The annual discount rate is 3% for both costs and health effects.

The risks
Disease risks All disease risks are annual age-and gender-specific excess incidence risk until the age of 95 years, see tables 2 to 5.
The COPD disease risk is taken from the Swedish population-based study Obstructive Lung Disease in Northern Sweden (OLIN), which was started in year 1985 . The risk is the reported average excess seven-year incidence among smokers in three age groups, of which the youngest was 45 years at baseline, see table 2. COPD was defined according to the spirometer GOLD definition.  The lung cancer disease risk is estimated from reports on lung cancer deaths until age 75 for smokers (15)(16)(17)(18)(19)(20)(21)(22)(23)(24) cigarettes/day) and non-smokers, see table 3. The annual excess death risk is estimated by a quadratic function of the accumulated risk until age 75 years. The lung cancer death risk is assumed 0 until the age of 40 years, and assumed constant between ages 75 and 95. The disease risk is obtained by adjusting the annual death risk by the annual crude survival rate of lung cancer in Sweden for a similar time period as the Peto data, from .  Source:    Source:  The CHD and stroke disease risk estimates are based on the Framingham CVD risk function, see table 4 and tables 5-6. As the Framingham CHD risk function only calculates CHD events, the division of these events into the particular diseases are based on recent Swedish register data, see table 7. To avoid over-estimation of risks, the risk factors for CHD and stroke are evaluated at minimal-risk levels; 120 mmHg for systolic blood pressure (SBP), HDL-cholesterol (HDL) at 1.5 and cholesterol (Chol) at 4. Diabetes is set at 0, while the variable smoking (smok) is set at 1 for the smokers.  Death risks All death risks are age-and gender disease-specific conditional risks; in some cases estimated as fractions of the general population age-and gender-specific mortality risk, see tables 2 to 4, and in some cases based on Swedish register data, see tables 8 to 11.
The COPD death risk is estimated from the study Obstructive Lung Disease in Northern Sweden (OLIN), which reported the 20-year mortality in three age groups. Comparison with the general age-specific mortality risks revealed no excess risk of death among those younger than 58 years and older than 70 years, but a considerable increased risk among those aged 58-70 years at follow-up. The excess risk was estimated at on average around 5 times the age-and gender-specific general population death risk, see table 2.
The lung cancer death risk is based on survival data from the Swedish National Cancer Registry, see table 8. The death risks for year 3 and 4 after diagnosis are estimated by linear interpolation between years 2 to 5. Lung cancer survivors at 5 years are assumed recovered, and returned to the health state healthy.
The death risks from CHD and stroke are taken from Swedish registers, see tables 9 to 11, or published scientific reports, see table 5. The death risks for AMI, stroke and IHD are divided into risks the first year after the first event and the second and following years after first event.   The model also incorporates the possibility of dying in unrelated diseases. The death risk in the health state Healthy is the average 5-year age group-and gender-specific risk adjusted for all model disease deaths, the last column in table 12. In disease health states, the risk of dying in unrelated disease is the average 5-year age group-and gender-specific   risk adjusted for the deaths in each respective disease. For ages below 39 years the risk in the age group 35-39 years is used, and for ages 80-84 years the risk >79 years. For ages above 84 years, the general population age-and gender specific death risk is used for the unrelated death risk. As the lung cancer death risks are so high, the unrelated death risks for lung cancer among individuals aged above 84 years had to be adjusted, by deducting 0.05. For those aged below 85 years, the age-and gender-specific general population risk of death is only used for calculating some disease-specific death risks, see tables 2 and 4. The risk is taken from the Swedish national mortality statistics for the year 2014 (Statistics Sweden, 2015).

Changes in risk after quitting smoking
The excess disease risks for smokers are not eliminated immediately after quitting smoking. This "lead time" is 36 years for lung cancer, 16 years for CHD, and 11 years for stroke, while for COPD some excess risk remain life-long, see heading effect of quitting in tables 2 to 4. The disease risks after quitting are constructed by adjusting the smokers' risks by the remaining risk. The remaining risk is modelled as fractions of risk, given the number of years since quitting. The annual remaining risks are estimated by linear interpolation. The effects on the risk for CHD and stroke are modelled on the dummy variable smoking, adjusting the value of 1 by the remaining risk fraction.

The societal costs
The model is reflecting the societal perspective, including disease-related costs for all sectors of the Swedish society. The costs included are medical treatment costs, costs for institutional care and technical aids, pharmaceutical costs, informal care and other patient and relatives' costs, and morbidity productivity costs.
Most of the data on societal costs are taken from Swedish studies published during the 2010s. Data reported as distributions, i.e. with the Gamma parameters for costs, or bootstrapped 95 percent confidence interval were preferred and used in the model to 13 enable stochastic estimation. If data was reported as mean and standard deviation, the Gamma distribution was simulated employing the Treeage function. In one case, data was reported as fraction of patients consuming a specific resource, which was used for sampling within the model. Otherwise the reported point estimate, usually the average cost across the patient group, was used. If no Swedish data on a cost item was found, the cost was taken from studies reporting data from settings assumed similar to the Swedish.
All costs are reported in SEK year 2014 (USD 1=SEK 6.86; Euro 1=SEK 9.10), adjusted when necessary with the Swedish CPI. To adjust reported Gamma distributed parameters to the price level, only the second parameter, i.e. the scale parameter, was adjusted.

Medical treatment costs
Recent Swedish estimates on medical treatment costs were possible to obtain for all model diseases, see table 13. The costs are paid by the regional healthcare authorities.

Pharmaceutical costs
Costs for pharmaceuticals in Sweden ought to be divided between the county councils and the patients, as patients pay a considerable share in co-payment. This is however not possible, given the data available. Table 15 therefore presents the drug costs to the regional healthcare authorities. The costs of pharmaceuticals dispensed during hospital stays are included in the medical treatment costs.

Informal care and other patient and relatives' costs
These costs include the value of care given to patients by relatives and other costs for patients or relatives, such as time, co-payments paid for health care and drugs as well as costs for transportation, modifications at home etc. Complete estimates could not be obtained for any disease, see table 16, except IHD which however might be outdated. Informal care in present-day Sweden probably constitute a sizeable part of total societal costs.

Productivity costs
The productivity costs only value the lost production because of morbidity before the age of 66 years, not mortality. The productivity costs for lung cancer is simulated within the model, via sampling from the fraction of patients on sick leave and combined with ageand gender-specific average monthly wages, including 40% employer taxes. Remaining data is taken from the literature, see table 17, and most estimates are recent. The costs are valued by the human capital method, and thus only include losses in salaried work before the official age of retirement.

Life years lost
The number of life years lost (YLS) are calculated until the age of 95 years, and only for individuals dead in the modelled diseases. Life years lost are presented both discounted 3% and undiscounted.

QALYs
The number of quality-adjusted life years (QALYs) are calculated during healthy years and years spent diseased, until death or the age of 95 years.
The QoL weights used during healthy years are mean age group-and gender-specific population weights, see table 18. The data is somewhat dated, but it is the only general population QoL weights available in Sweden. The QoL of the age group 20-29 years is used  also for younger ages, and the QoL of the age group 80-88 years is used for those aged 89-95 years. This last assumption is probably an overestimate.
The disease-specific QoL used in the health states are all, except one, modelled as decrements from the average population age-group and gender-specific QoL, see table 19. For lung cancer no data was available on the marginal effect of the disease on the population average QoL, so a fixed value over the ages and genders had to be used.

Sensitivity analyses
Several univariate and multivariate sensitivity analyses have been performed. Analyses on some methodological issues, as well as a probabilistic sensitivity analysis, have also been performed. The analyses are reported for men and women aged 50 years.
To give another measure of the uncertainty surrounding the cost-effectiveness ratio, the 95% confidence interval for the difference between smokers and quitters is reported, calculated from the standard deviation of outcomes.

Results
In this chapter, the model estimates of QALYs, YLS and societal costs are presented for men and women in some selected ages, mainly for validation purposes. More detailed simulation outcomes as well as the results of the sensitivity analyses are presented for men and women at age 50 years. Model estimates can be obtained for men and women for all ages between 15 and 95 years.

The model estimates
In table 20 the simulation results for QALYs (quality-adjusted life-years) experienced until the age of 95 years are presented, for the selected ages 15,30,50 and 70 years at the start of the simulations. As can be expected, the number of QALYs are highest in the younger age groups, and somewhat higher for women in most age groups. In the selected age groups, the differences between smokers and quitters are at a maximum at age 30; 0.68 for females and 0.81 for males. The confidence intervals, calculated via the mean and standard deviation (sd) from the 10 000 model runs, indicate that there are differences in QALYs between smokers and quitters.
The YLS (life-years saved) lost before the age of 95 years are presented in tables 21 and 22, discounted 3% and undiscounted. The differences in discounted YLS between smokers and quitters are somewhat higher than the differences in QALYs. The undiscounted YLS in table 22 show the number of years that smokers and quitters are expected to lose before the age of 95 years. For the ages 15, 30, and 50 the number of lost life-years is estimated at around 6 years for women smokers and 9 years for men, implying that the female smokers are estimated to live until age 89 and the male until age 86. In the oldest age group presented here, age 70, the number of lost life-years are only 1-2 years. The quitters are estimated to lose considerably fewer life-years; 1-4 years for the women and 3-5 years for the men. As expected, the difference between smokers and quitters diminish with age, with a maximum at around 5 years for the females and around 6 years for the males at age 15.
The societal costs estimated for the smokers and quitters for the selected age groups are presented in table 23. The highest costs are found for age 50; 200 000 SEK and 250 000 SEK for the smokers and 130 000 and 170 000 for the quitters, in both cases higher among the men. The highest difference between smokers and quitters is however found at age 30, with a difference of 100 000 among the females and 120 000 among the males. The difference among the eldest, at age 70, is around 20 000 SEK. These cost differences reflect the amount that tobacco cessation interventions could spend on achieving one quitter and still be cost-saving.

Selected model outcomes
The underlying estimated disease outcome is presented in figures 2 and 3, for the age 50 years. For both women and men, there is a marked decrease for quitters in the number of diseased and dead in the model diseases, which is somewhat offset by an increase in the number of deaths in unrelated diseases. The number of diseased and deaths are higher for    1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  men, mainly originating from CHD. The model disease with the highest smoking-related incidence is COPD, for both genders. The increase in unrelated deaths for the quitters is an example of competing risks, which decreases the difference in life-years and QALYs between smokers and quitters. Table 24 and 25 shows the full model simulation results of the societal cost savings because of tobacco quitting at age 50 years. For women, the highest estimated savings are found in lung cancer, COPD and stroke at around 15-20 000 SEK per quitter. For men the cost savings because of lung cancer are considerable higher, at around 35 000, due to the higher incidence among the men. The cost item with the largest cost savings are medical treatment costs for both genders, at around 30 000 SEK. Most of the difference in savings between men and women originate from the productivity costs, possibly reflecting disease onset at younger ages among men. Note that a cost saving of zero means that no cost is being modelled, as cost data was lacking.

Sensitivity analyses
The results of the sensitivity analyses are presented on women and men at starting age 50 years. Figure 4 shows the results for women and figure 5 for men.
All analyses show a similar pattern between men and women, and also similar ranges. The univariate sensitivity analyses on the model parameters, analyses A to E, result in small changes in costs and QALYs. Also the multivariate analyses F and G, which are constructed as scenarios that allow the risk parameters to vary consistently upwards or downwards, and along with the costs in analysis G, show moderate changes from the base case estimates. The methodological choices have a more pronounced effect, as the largest difference in QALYs is achieved by varying the discount rate (analysis H) between 0 and 5%, which also affects the costs substantially. The two analyses that reflect the choices of which costs to include in the estimates, analysis I that reflects the UK NICE health care and social services perspective and analysis J that only include Swedish data published since the year 2005, both decrease the cost differences between smokers and quitters.  The scatter plot of the bootstrap analysis based on the microsimulation results for women and men aged 50 are shown in figures 6 and 7. The uncertainty is higher for the men, as the plots are more scattered. All plots are however situated in the cost decrease and QALY increase quadrant, with costs below -20 000 SEK and QALYs over 0.2.

Discussion: Model validity
The discussion of the model validity is structured around four aspects as proposed by : the structure of the model, the inputs to the model, the results of the model and the value of the model to the decision-maker.
The structure of the model The structure of the model is a Markov model constructed for microsimulations, on the three most smoking-related disease groups; lung cancer, COPD, and CVD including stroke and CHD. The present updated version of the model includes one less CHD disease compared to the first version of the model, as unrecognized acute myocardial infarction now is included in the IHD disease, mainly because the disease definition is rarely used nowadays. Choosing only three disease groups is a clear simplification as smoking is known to cause hundreds of different diseases. The effects from smoking, and thus quitting, are furthermore confined to the individuals themselves; no side-effects on other individuals such as environmental tobacco smoke or smoking uptake are included. These two features leads to an underestimate of the true effects of tobacco quitting.
The same disease-specific approach has been taken by most other tobacco cessation models , even though some of them include more diseases, such as asthma. Another approach would be to use the overall differences in mortality between current, former, and never-smokers taken from large US studies, as some early tobacco cessation models did Tengs et al, 2001). In order not to overestimate the effects of quitting tobacco, we chose to model the smoking-related risk for certain diseases instead, as it is improbable that all differences in mortality and morbidity between smokers and former smokers are due to the smoking habit (Doll et al, 1994).
The model aims to reflect disease onset related to smoking tobacco. As disease in all the three disease groups included in model may be caused by other factors than smoking only the excess risks for smokers are modelled. For the diseases lung cancer and COPD this implies that the risk for smokers found in epidemiological studies is adjusted by the risk found for non-smokers. For the disease group CHD and stroke, where a large fraction of disease onset is caused by other factors than smoking, this adjustment for smokers´ excess risk was performed by setting the other risk factors in the risk function at minimal risk levels. This is an underestimate, as the risk factor levels among smokers can be expected to be at least as elevated as among the general population. The underestimate is aggravated by the fact that the functional form of the risk function results in a multiplier effect of the risk factors.
The present version of the model includes seven health states: lung cancer, COPD, stroke, and CHD divided into four diseases. This is a clear simplification, as the costs and QoL can be expected to vary considerable between patients with different severity levels within the diseases. This is particularly true for COPD which is a chronic progressive disease, i.e. the diseased get more severely ill over time. However, a model with 7 health states with accompanying disease-specific death risks, costs and QoL weights is fairly complex as well as data-demanding. For the purposes of this study´s model, the division of diseases into severity levels was not deemed necessary.
An obvious problem with the model, inherent in all Markov models, are the mutually exclusive health states; any individual can only contract one disease, and once diseased the individual never recovers (apart from the very rare 5 year survivors in lung cancer). This feature implies both an overestimate and an underestimate of the true effects. The underestimate stems from the fact that co-morbidity is very common, especially among the individuals with the chronic diseases COPD, CHD, and stroke. The overestimate of costs and effects arise as individuals stay in the health states until death. If the costs and outcomes associated with the health states are taken from severely ill individuals, then these become grossly overestimated. This overestimate is partly offset by the use of separate costs for the first and subsequent years, for all societal costs due to AMI and stroke. In order not to overestimate the numbers of years spent in disease states, the possibility of dying in unrelated diseases is present in all health states. This feature is also included in the CHD Policy Model (Weinstein et al, 1987).
Most tobacco cessation models are built for cohort estimation , but this model is constructed for individual-level estimation using the microsimulation methodology. As the data available admitted a microsimulation structure, e.g. the risk functions, the methodology was chosen as the advantages to model and to obtain a richer data set with results that reflect the heterogeneity of outcomes between individuals was deemed to offset the disadvantages of calculation burden. The use of the software Treeage also facilitates the use of microsimulation. Age-and gender-specific estimates can thus be obtained from the model, between ages 15 and 95 years.
The model stages are one-year long, which seems accurate given the risk estimates and the long time horizon of the model. The reason for the model maximum age of 95 years is the lack of risk estimates for older ages. Some extrapolations of risk estimates to the age of 95 years indeed resulted problematic, as some disease-specific death risks expressed as multipliers of the average age-specific death risk resulted in risks above 1. Further extrapolations beyond the age of 95 years were deemed unnecessary, as most of the relevant differences between smokers and quitters would have arisen by that age.

The inputs of the model
The second aspect of model validity is the inputs of the model. The model contains a large number of data taken from different sources. This is of course a threat to the internal validity of the model, shared with most models. However, the data have been chosen to reflect current Swedish circumstances. The current updated version of the model has exchanged almost all cost data, if more recent estimates were available, and all death risks to recent Swedish register data. As the number of studies on any particular data items are few, no meta-analysis or any other synthesis of data was carried out. The disease risks are of course are pivotal for the result. The lung cancer disease risks are probably the best that can be obtained, from a large epidemiological study . The risk equation used for CHD and stroke is taken from the Framingham studies, and even though there are more recent risk scores developed from the study (D'Agostini et al, 2008), the  risk functions are still frequently employed. The disease COPD has been the subject of a large long term epidemiological study in Sweden, The Obstructive Lung Disease in Northern Sweden (OLIN) , which is thus the most relevant data source for the model.
In the model, there is an increased risk for a smoking-related disease remaining for some years after the tobacco cessation, in accordance with epidemiological evidence (Surgeon General, 1990;. The feature is also considered a marker of high quality tobacco cessation models .
The majority of the cost data are taken from Swedish studies published during the 2010s. To take fully advantage of the microsimulation structure and to obtain stochastic estimates, the preferred data sources were the ones reported as distributions, i.e. as Gamma parameters or bootstrapped 95 percent confidence intervals. If no Swedish data was found, an international estimate was instead used in order to seek to represent the full societal costs. However, apart from certain cost items and for some of the diseases, the lack of data results in considerable underestimates of the true societal costs. This is particularly true in the cases of the costs for care, both institutional and informal. The institutional care could amount to considerable costs, exemplified by the costs for stroke and AMI patients, see table 14. In particular for lung cancer the lack of data results in considerable underestimates of the true disease-related costs. This is why the possible overestimate of the informal care for the disease, obtained from an Italian study, probably does not bias the overall result. To investigate the issue, one sensitivity analysis only included recent Swedish data. The analysis lead to decreases in cost savings for quitters aged 50 years of around 30%.
The QoL estimates are constructed as disease-specific decrements from the average ageand gender-specific QoL, except for lung cancer for which no QoL decrement could be found . The average population age-and gender-specific QoL weights, which are certainly not 1, are also used during healthy years for the base case estimates. This means that the model assumes that an individual that avoids the smoking-related diseases is not having perfect health, but the health of an average Swede at the same age, as recommended .
The stated purpose of the model is to reflect the societal perspective, which for Sweden includes the morbidity productivity costs, but not the productivity costs resulting from mortality. All the model data on productivity costs value them according to the human capital approach for individuals under the age of 65, the customary Swedish age of retirement.
A full societal perspective might also include other aspects, considering that this is a model on individuals that are participating in an intervention that aims to change their lifestyle. The previous version of the tobacco cessation model, version 1 , reported 28 sensitivity analyses that modelled some effects on the tobacco quitters, by including savings from cigarette purchases and a decreased QoL because of withdrawal effects during the first year. When that analysis was applied to an intervention, a decreased QoL during the first year was also deducted for the smokers that failed to quit, as the failure to achieve a personal goal might to lead to a decrease in QoL.
The results of the model The third aspect of model validity is the results of the model, e.g. a comparison with reality or with other study results. A direct comparison with reality is not possible, since the model covers the ages 15-95 years, with a follow-up time of 80 years for the youngest age group.
The model estimates that around 60% of the women and 70% of the men aged 50 at the start of the simulations will contract one of the modelled diseases, and that around 50% of those will die in the diseases before the age of 95 years. The disease risks for the quitters at age 50 are not eliminated; 30-40% of them will still contract the smoking-related diseases because of remaining disease risks after quitting. As expected, the unrelated deaths increase among the quitters, in sum leading to an increase in YLS (undiscounted) of 2-3 years for those quitting at age 50, compared with continuing smokers. The increases in QALYs (discounted 3%) are smaller because of less-than-perfect health among those aged 50 years and above; 0.61 for women and 0.77 for men. The disease outcomes are fairly similar to the estimates from the previous versions of the model, but because of decreased death risks, the outcomes in terms of YLS and QALYs are considerably higher. The 2004 version of the model estimated an increased YLS of 0.93 and of 1.66 for women and men aged 50-54 years, and QALY gains of 0.36 and 0.71, respectively. The differences are due to the longer time perspective of the present version, 95 years versus 85 years, and the somewhat decreased case-fatality risk (i.e. the mortality risk among those with disease) because of improvements in medical technologies during the past decade.
Apart from increases in health, the societal cost savings because of quitting smoking are considerable. For men, the cost savings amount to around 100 000 SEK for quitters aged between 15 and 50 years, and around 70-90 000 SEK for women. Even in the age group 70 years there are estimated cost savings of around 20 000 SEK per individual quitter. This implies that substantial funds could be invested in smoking cessation interventions, and the interventions would still be cost-effective, or even cost-saving. The cost savings in the present model are considerably higher than those of the previous model, in part due to changes in price year.
Comparisons of model estimates with other models' are difficult to perform, as the time horizon, costs included, jurisdiction, and the diseases included differ. Among the recently reported model estimates , there are two Australian models. The model developed within ACE  report estimates of life-years saved that are considerable higher than the present model's; 5.7 years for men and 6.6 years for women in age group 50-54 years. That model time horizon is however 100 year, but it is unlikely There have been two, to my knowledge, reports of tobacco cessation model estimates for Sweden, one using the Benesco model  and one using an extended version of the HECOS model . Comparison with those model estimates are unfortunately not possible, due to lack of reporting detail. However, estimates from the previous version of this model were fairly consistent with the HECOS model estimates  for Sweden, available at the time .

The value of the model to the decision-maker
The fourth aspect of validity is the value of the model to the decision-maker. There are several models on tobacco cessation that conforms to international recommendations on how to perform cost-effectiveness analyses . This model however reflect Swedish circumstances, with Swedish cost and QoL data, why the model might be useful for Swedish decision-makers.
We hope that the model will be used to perform economic evaluations of a range of tobacco cessation interventions. For tobacco prevention interventions, i.e. prevention of initiation of smoking, another model version, version 2, has been constructed and is available for analyses. The use of these models will in time enable incremental and marginal calculations of the cost-effectiveness of different tobacco interventions and their components and suitable target groups. The basis for decisions on which tobacco cessation and prevention interventions to implement will then be more comprehensive.
Another frequent use of models is to forecast future events. This model is not suitable for estimating what the costs of smoking will be in the future. The reason is that the model does not incorporate any adjustments of possible future developments. The risk of smoking is based on studies with follow-up periods of sometimes 30 years, which means that the risks are reflecting the smoking behaviour among smokers 30 years ago. The changes in cigarette content and in the frequency of smoking might lead to changes in disease risk in the future. Also the costs for the smoking-related diseases might change in the future, because of changes in health care technology. Another example would be the value of the morbidity productivity costs, as well as informal care, as wages and productivity often are expected to increase in the future.
Nevertheless, the model actually forecasts what the costs for smokers and quitters will be in 80 years' time, for the youngest age group. That implies that we know that the model forecasts will be wrong, but it is of minor significance as the model is constructed to be used for comparisons between two groups, smokers and quitters, thus eliminating some

The uncertainty
Another aspect of model validity is the uncertainty surrounding the model estimates.
The univariate sensitivity analyses on the model parameters (analyses A-F in figures 4 and 5 for men and women aged 50) show minor deviations from the base case result, while the multivariate analysis on costs and risks combined (analysis G) affects in particular the cost estimates. The methodological choices affect the results to a greater extent, with the discount rate (H) heavily influencing the QALYs and the more restricted perspective (I) decreasing the cost-savings. The multivariate analysis that only include higher-quality data (J) also imply decreases in the cost differences between smokers and quitters, but the difference remains substantial; around 50 000 SEK for females aged 50 years and 60 000 SEK for men, respectively. The overall conclusion from the parameter sensitivity analyses is that the QALY gains are at least 0.35 and 0.40 and the cost savings at least SEK 35 000, for female and male quitters aged 50, respectively.
The probabilistic analysis shows no uncertainty whether quitting tobacco leads to costsavings and increases in QALYs, as all bootstraps are placed in the southeast quadrant of the cost-effectiveness plane. The bootstrap results exhibit a mixture of first and second order uncertainty, as it reflects both the probabilistic structure of the Markov model and the simulation of some parameter values .
Another measure of uncertainty is the confidence intervals around the estimated mean differences, reported in tables 20-23. However, that measure is not fully appropriate as the large sample sizes of the Monte Carlo simulation (10 000 runs) diminishes the standard error of the mean .
The structural uncertainty of the model, i.e. whether the results would be different if the model would have been constructed in another way, have not been studied. Alternatives to the chosen model structure could have been deterministic or discreet event simulations, more or less health states, other functional forms of risk functions, and other subgroups than men and women and five-year age-groups model results. The flaw is however shared with most tobacco quitting models .

Checking for technical errors
The model contains a large number of trackers, i.e. variables that count events, to enable checking for technical errors. Tentative runs were executed after the introduction of every new variable, with cost items undiscounted, and the simulation results examined manually. Thus, the model has been thoroughly checked for technical errors.

Conclusions
The aim of this study is to develop a model predicting health and economic consequences of smoking cessation, to be used for cost-effectiveness analyses of smoking cessation interventions. The updated model strives to incorporate data that is recent, accurate and appropriate for Sweden in year 2015. The model also adhere to Swedish recommendations on how to perform cost-effectiveness analyses within the health care sector. Data is however lacking to completely fulfil these requirements. Many model parameters are based on very few studies. Some information just does not exist, at least not accessible to us.
These are issues shared with most model, however. The purpose of modelling is to assemble the most accurate information at a point of time, to enable decision-making at that particular point of time. This is in accordance with one of the fundamentals of economics: decision-making under uncertainty, which implies that decisions have to be made even if there is no full information. We hope that the model will be applied to a range of different tobacco cessation interventions, which in time will enable a more comprehensive basis for decision-making. Identify the study as an economic evaluation or use more specific terms such as "cost-effectiveness analysis", and describe the interventions compared. For each intervention, report mean values for the main categories of estimated costs and outcomes of interest, as well as mean differences between the comparator groups. If applicable, report incremental cost-effectiveness ratios.
Page 15, Table 3 Page 17, Table 4 20a Single study-based economic evaluation: Describe the effects of sampling uncertainty for the estimated incremental cost and incremental effectiveness parameters, together with the impact of methodological assumptions (such as discount rate, study perspective).

Not applicable Characterising uncertainty 20b
Model-based economic evaluation: Describe the effects on the results of uncertainty for all input parameters, and uncertainty related to the structure of the model and assumptions.

21
If applicable, report differences in costs, outcomes, or cost-effectiveness that can be explained by variations between subgroups of patients with different baseline characteristics or other observed variability in effects that are not reducible by more information.

Discussion
Study findings, limitations, generalisability, and current knowledge 22 Summarise key study findings and describe how they support the conclusions reached. Discuss limitations and the generalisability of the findings and how the findings fit with current knowledge.

Pages 18-20
Other Source of funding 23 Describe how the study was funded and the role of the funder in the identification, design, conduct, and reporting of the analysis. Describe other non-monetary sources of support.

Conflicts of interest 24
Describe any potential for conflict of interest of study contributors in accordance with journal policy. In the absence of a journal policy, we recommend authors comply with International Committee of Medical Journal Editors recommendations.