The modelled cost-effectiveness of nicotine preloading

Introduction

The objective of the current analysis was to compare the long-term cost-effectiveness of pharmacotherapy for 4 weeks prior to setting a smoking quit date with no pharmacotherapy before the quit date through extrapolation of the results of the clinical trial. The analysis was conducted using a modified version of the European-study on Quantifying Utility of Investment in Protection from Tobacco (EQUIPT) return on investment (ROI) tool.66 At the core of the EQUIPT ROI tool is a validated economic model, which forecasts the short- and long-term outcomes for smokers who cease smoking within the next calendar year or who continue smoking.67 The model allows the estimation of the impact of smoking cessation interventions based on these outcomes. As detailed below, the economic model has been modified slightly to allow the estimation of the comparative costs and effectiveness of the interventions within the current clinical trial, which compared the use of nicotine patches for 4 weeks prior to an attempt to quit smoking with no use of pharmacotherapy prior to the quit attempt.

Methods

Clinical trial

The clinical trial compared treatment with nicotine patches for 4 weeks prior to a planned quit date (intervention) with no pre-treatment prior to a planned quit date (control). Both the intervention and control groups also received standard behavioural counselling prior to the quit date. Biochemically validated 12-month abstinence was achieved by 14.0% of the intervention group and 11.3% of the control group. In the primary analysis, this gave an unadjusted OR of 1.24 (95% CI 0.97 to 1.58; p = 0.086) and a RD of 2.71 (95% CI –0.37 to 5.78; p = 0.085). As expected, there was an imbalance between arms in the frequency of use of varenicline for post-cessation medication; adjustment for this gave an OR of 1.30 (95% CI 1.02 to 1.66; p = 0.036) and a RD of 3.31 (95% CI 0.22 to 6.39; p = 0.036) between the intervention and control groups.

Economic model

A Markov model (EQUIPTMOD), developed as part of EQUIPT, was used to estimate the long-term costs and effects of using NRT pre-treatment compared with no pre-treatment in a hypothetical cohort of smokers with the same age and sex distribution as in the trial, who make a one-time attempt at quitting smoking. EQUIPTMOD allows the determination of the impact of either continuing or stopping smoking on both morbidity and mortality outcomes and costs. In this chapter, we estimate the lifetime costs, QALYs and life-years associated with NRT pre-treatment compared with no pre-treatment.

Methods

Design of the model

There are three primary states within the model: (1) current smoker, (2) former smoker and (3) dead (Figure 8).

FIGURE 8 Model schematic. (Figure not reproduced here; see the original article.)

With each 1-year cycle, the cohort is subjected to a set of transition probabilities that allow them to either stay within their current state or move to one of the other two states. Death is an absorbing state, meaning that those who enter this state remain within the state. We ran the model until participants reached the age of 100 years or died.

The model is replicated for different population cohorts based on the age and sex distribution of patients within the clinical trial. To obtain trial population-level estimates, these cohort-level estimates are weighted by the percentage of the trial population falling into each age and sex cohort.

To estimate the impact of the intervention on costs and smoking-related outcomes, two separate models were created: one that simulates a cohort of smokers who quit in the first year and one that simulates a cohort of smokers who do not quit in the first year. The results of the two models are combined by weighting the outputs by the trial population and the effectiveness of the intervention and the control.

Transition probabilities

Transition probabilities incorporated within the model include:

• the probability of quitting smoking in the first cycle (year) based on the effectiveness of the intervention and the control within the trial
• the age- and sex-specific annual probability of mortality associated with current smoking and being a former smoker
• the probability of smoking cessation and relapse over the long term in current and former smokers.

The probabilities of smokers successfully quitting in the first year after the intervention were based on the biochemically validated 12-month abstinence rates from the clinical trial. Based on this measure, 14.0% of those in the intervention group and 11.3% of those in the control group were abstinent at 12 months. In the base-case analysis, the 12-month abstinence rate was assumed to be predictive of long-term cessation. This was tested within sensitivity analyses through applying a rate of relapse of 30% in the second year after the intervention.68 Smokers who did not quit smoking, or who relapsed, were subjected to the same background quit rate in all subsequent years, which is reflective of the balance of quitting and relapsing over a lifetime.69

The probability of death by age, sex and smoking status was calculated using the attributable risk of death based on smoking status. The following was required for this calculation: age- and sex-specific probabilities of death in the general population; the age- and sex-specific prevalences of smoking status (current, former or never smokers); and the RR of death by smoking status. Mortality rates were based on actuarial life tables and the age- and sex-specific prevalences of smoking were derived from the survey data.70,71 The RRs of death for former and current smokers compared with never smokers were derived from published data19 (see Appendix 1, Table 29).

Long-term cessation and relapse were modelled through the application of an underlying quit rate, which applies to all age- and sex-specific cohorts after the first year. This was estimated at a value of 2% per annum and represents the balance between those quitting smoking each year and those who start smoking or relapse.69

Prevalence of smoking-related diseases

The smoking-related population attributable fraction relating to current and former smokers for smoking-related diseases was estimated based on epidemiological data. The model estimates, for each cycle, the prevalence of smoking-related diseases based on the proportion of the cohort who are either current smokers or former smokers.

Four diseases, lung cancer, coronary heart disease (CHD), chronic obstructive pulmonary disease (COPD) and stroke, were found to be responsible for the majority of the smoking-related disease burden and were, therefore, incorporated within the model.72,73 The number of cases each year of these diseases was derived through calculation of the attributable fraction based on the prevalence of smoking-related diseases in England, the prevalence of smoking status in England (current, former or never smokers) and the relative effect of smoking on the prevalence of each disease49,74–78 (see Appendix 1, Tables 30–36).

Costs

Within the model, costs comprised both the smoking cessation costs of the intervention and control arms of the clinical trial and the health-care costs associated with smoking-related diseases.

Prior to the planned quit date, both the control group and the intervention group received standard behavioural counselling. In addition, those in the intervention group also received 4 weeks of NRT. Based on the within-trial analysis, this resulted in a mean cost per person of £58 [standard error (SE) £0] in the intervention group and £11 (SE £0) in the control group. Furthermore, prior to the quit date, both treatment arms received NHS Stop Smoking Services, with a mean cost per person of £124 (SE £1) in the intervention group and £122 (SE £1) in the control group. The cost of the NHS Stop Smoking Services accessed over the 12 months after the quit date was also included in the analysis: £133 (SE £6) for the intervention group and £103 (SE £5) for the control group, based on the within-trial analysis. This resulted in total costs overall for smoking cessation of £295 (SE £6) for the intervention group and £236 (SE £6) for the control group.

The annual costs for an individual with each of the diseases were estimated based on a systematic literature review. Costs relate to prevalent cases of disease. The annual cost for COPD is £1042.57, for lung cancer is £7002.65, for CHD is £1247.79 and for stroke is £5886.6679–82 (see Appendix 1, Table 36).

All costs were inflated to the year 2016 using the Hospital and Community Health Services Index83 (see Appendix 1, Table 37).

Utility values

Utility decrements associated with smoking-related diseases were derived from the US Medical Expenditure Panel Survey.84 Utility values based on current smoking status (current or former smoker) were derived from the Health Survey for England adjusted for relevant covariates including disease prevalence.85 The EuroQol-5 Dimensions, a validated instrument for measuring health-related quality of life, was used for both sets of utility values and UK tariffs were used for calculating utility scores86 (see Appendix 1, Table 38).

Outcomes

The costs, life-years and QALYs associated with both the intervention and the control groups were estimated over the lifetime of the cohort. In the base case, both costs and effects were discounted at a rate of 3.5% per annum as per NICE guidance.87 A 1.5% per annum discount rate for costs and outcomes was tested in the sensitivity analyses. The analysis was conducted from the perspective of the NHS.

The incremental cost–utility ratio, expressed as the incremental cost per QALY gained, was reported to assess the cost-effectiveness of the intervention group compared with the control group.

Key assumptions

As is the case with all models, a number of assumptions were required to estimate the economic impact of tobacco-control interventions; these have been detailed in a previous publication67 and are summarised in Box 1.

The population-based mortality rates are adjusted using the RRs of death in smokers and former smokers, which are derived from the literature.19 Although the reference is dated and absolute mortality may have changed, the assumption that the relative effect was likely to be maintained, and the choice of study, was justified based on the prospective nature of the study, the sample size (n = 34,439) and the years of follow-up (40 years).

The current model does not explicitly adjust for the time since quitting because of the absence of distributional data regarding time since quitting; duration of smoking; and the risk of smoking-related disease and mortality. Rather, an average risk of smoking-related disease and mortality is applied to former smokers. As this is a cohort rather than an individual patient simulation model, the impact of this assumption may be limited.

Given the lack of data to support an alternative assumption, the prevalence of each disease is assumed to be independent of the prevalence of other diseases. Also, the model assumes that, in the case of multiple diseases, the disutility associated with the disease with the greatest disutility is applied. This is a conservative approach in that it provides a lesser estimate of the QALY gains from smoking cessation than either a multiplicative or an additive approach.

For all diseases, in people aged < 35 years, the risk of smoking-related disease was assumed to be equal across smoking groups. This was deemed to be the most appropriate assumption given that data regarding the differential rate of all diseases by smoking status were not available for this age group. This assumption is both conservative and, given the very low prevalence of disease in this group, unlikely to have a significant impact on the results.

The underlying quit rate, which applies to all cohorts after the first year, represents the balance between those who quit smoking each year and those who start smoking or relapse. This assumption is supported by a meta-analysis, which showed that there was no difference in relapse rates after 12 months, regardless of whether the patients used or did not use an intervention to quit smoking.68

Handling uncertainty

The effect of parameter uncertainty on the calculated outcomes was assessed through probabilistic sensitivity analyses involving Monte Carlo simulation.88 Standard probability distributions were included for the natural history parameters, RRs and ORs, costs and utilities. Beta distributions were used for transition probabilities and utility values by smoking status, normal distributions were used for utility decrements due to disease, log-normal distributions were used for RRs and ORs and gamma distributions were used for costs. Population-level data were assumed to be fixed. Distribution parameters can be found in Appendix 1, Tables 35–38. The background quit rate was parameterised using a beta distribution. As distributional data were not available we assumed a conservative sample size of 50 for the parameterisation.

A set of 5000 outcome estimates were obtained, with the results displayed using a scatterplot of incremental costs versus QALYs and a CEAC that graphically presents the probability that the intervention is cost-effective as a function of the willingness of decision-makers to pay for an outcome (QALY).88

Results

Base-case analysis

The total lifetime costs within the intervention group were £13,111 and within the control group were £13,177 (a difference of £66). The costs associated with smoking cessation, which included both behavioural and pharmacological interventions, were higher in the intervention group (£295) than in the control group (£236; a difference of £59). Health-care costs associated with smoking-related diseases over a lifetime, however, were lower in the intervention group (£12,816) than in the control group (£12,941; a difference of £125) (Table 26).

A greater number of individuals quit smoking in the intervention group than in the control group (139 per 1000 smokers vs. 112 per 1000 smokers, a difference of 27 individuals). The greater number of individuals quitting smoking in the intervention group resulted in more discounted life-years lived in the intervention group than in the control group (17.977 life-years vs. 17.957 life-years) and more QALYs (14.300 vs. 14.267). The intervention is dominant over the control as the intervention was both more effective, producing both more QALYs and more life-years, and less costly.

Additional analysis

An analysis was conducted adjusting for an imbalance in the use of varenicline post quit date within the two groups (Table 27). The adjusted RD for cessation at 12 months was 3.31 (95% CI 0.22 to 6.39), compared with a RD of 2.71 (95% CI –0.37 to 5.78) for the base case. The adjusted difference in smoking cessation costs was £63 (95% CI £52 to £78), compared with £59 (95% CI £44 to £74) in the base case.

After adjustment for varenicline imbalance, there were 33 more people who quit in the intervention group than in the control group; this is six more people who quit than in the unadjusted analysis. The difference in total costs was also greater, with the intervention costs being £90 lower than the control costs. The intervention also resulted in more QALYs (a difference of 0.040) and life-years (a difference of 0.025) than the control. As in the base-case analysis, the intervention is dominant over the control as it is both less costly and more effective.

Deterministic sensitivity analysis

The cost-effectiveness results were robust to an alternative discount rate of 1.5% per annum for costs and outcomes, with the intervention remaining cost saving (£19,273 for the intervention group vs. £19,384 for the control group) and resulting in more QALYs (19.091 for the intervention group vs. 19.043 for the control group) (Table 28).

With a relapse rate of 30% after the first year, the cost difference between the intervention group and the control group was reduced, as were the QALY and life-year differences. However, the intervention remained less costly and more effective than the control and is, therefore, still dominant over the control (Table 28).

Probabilistic sensitivity analysis

As can be seen from the scatterplot of incremental costs compared with incremental QALYs (Figure 9), the majority of the estimates (80%) are within the lower right quadrant, indicating that the intervention is dominant over the control. As is evident from the CEAC (Figure 10), at a willingness-to-pay (WTP) threshold of £20,000 per QALY there is a 94% probability that the intervention is cost-effective. There is a 95% probability that the intervention is cost-effective at a WTP threshold of £30,000 per QALY.

FIGURE 9 Scatterplot of incremental costs vs. incremental QALYs for the intervention group vs. the control group. (Figure not reproduced here; see the original article.)

FIGURE 10 Cost-effectiveness acceptability curve for the probability that the intervention is cost-effective with a relapse rate of 0% after year 1. (Figure not reproduced here; see the original article.)

If 30% of smokers who quit in the first year are assumed to relapse, there is a 93% probability that the intervention is cost-effective at a WTP threshold of £20,000 per QALY and a 94% probability that the intervention is cost-effective at a WTP threshold of £30,000 per QALY (Figure 11).

FIGURE 11 Cost-effectiveness acceptability curve for the probability that the intervention is cost-effective with a relapse rate of 30% after year 1. (Figure not reproduced here; see the original article.)