This analysis applied a cohort simulation model to determine the cost-effectiveness of text-based support in addition to current practice versus current practice alone. Current practice is defined as the mix of interventions currently available in the UK to help people stop smoking (as represented in the txt2stop trial). One in seven of the trial participants were using additional smoking cessation support at randomisation (82 % used nicotine replacement therapy, buproprion or varenicline; 4 % a telephone helpline; 3 % group or individual counselling; and 12 % used other support).
Model
This study uses a model adopted in previous economic evaluations of smoking cessation interventions conducted in UK from a health service perspective [5, 6]. The Markov state transition model (Fig. 1) used in the study by Flack et al. [5] is populated using the most recent UK data. At the start of the analysis, the simulated population consists entirely of smokers. A 6-month cycle is adopted, with transitions between smoking status occurring every 6 months according to the probability of remaining, in or moving to, one of three mutually exclusive states: smoker, former smoker, and dead. In each cycle it is assumed that both former and current smokers have a chance of developing the five main health consequences of smoking: lung cancer, stroke, myocardial infarction (MI), chronic obstructive pulmonary disease (COPD), and coronary heart disease (CHD) [7]. A lifetime horizon is chosen in order to calculate the incremental cost of text-based support, and the life years (LY) and quality adjusted life years (QALY) gained. This time period is necessary to allow for the inclusion of all costs and effects of the intervention. All costs are expressed in UK pounds (£) in terms of financial year 2009–2010. Costs are estimated from an NHS perspective and the discount rate used is 3.5 % for cost and outcomes as per NICE guidance [8]. In order to allow comparison with previous economic evaluations of smoking cessation interventions in UK we have used the same data sources as Flack et al. [5]; more recent data being used where available.
Study population
The number of LYs and QALYs gained by text-based support, and thus its cost-effectiveness, will depend on the initial age of the smokers as smoking mainly causes health conditions that need a long time period to develop after exposure to smoking. To allow for this, the analysis is conducted separately for three age groups: <30 years (30 % of the trial population), 30–40 years (31 %), and >40 years (39 %). The mean age for each age group was 24, 35 and 48 years, respectively, based on the txt2stop trial data. These values are adopted as the starting age in the cohort models. The number of LYs and QALYs gained also depends on the gender of the smokers. It is well established that women live longer than men but with a higher burden of disabilities [9]. In order to account for this, the analysis is conducted for each age group separately for men and for women. The overall results (number of life years, QALY and disease costs) are calculated as a weighted average of the results for the three age groups using the gender proportions observed in the trial (male 55 %, female 45 %).
Probabilities
The relative risk of quitting at 6 months for text-based support observed in the trial, 2.20 (95 % CI 1.80–2.68, P < 0.0001), is applied to the quit rate at 6 months in the control group (4.9 %) [1]. A 21 % relapse rate is assumed between 6 and 12 months for both the control and the treatment group [10]. A recent meta-analysis of 12 trials estimates that there is no difference between control and treatment groups in the relapse to smoking after 1 year of cessation (OR 1.11, 95 % CI 0.78–1.59). Based on this latter study, a lifetime relapse rate of 30 % among those who have quit for 12 months is assumed in the model [11].
Failure to take account of the background quit rate would lead to an overestimate of the effectiveness of text-based support, since a number of the additional quitters would have quit anyway in the future in the absence of test-based support. The background quit rate is likely to vary across age groups. The markedly higher proportion of ex-smokers among those aged 55 and over might indicate a higher background quit rate in this age group [12]. Also the background quit rate may be increasing as more support for smoking cessation becomes available. However, in the absence of good data, and following previous studies, a background quit rate of 2 % per year is assumed for all smokers (with or without text-based support) for all 6-month simulation cycles independently of the age and gender of the smokers [13].
Mortality rate in the general population by age and gender were retrieved from the Health Survey for England [12]. The prevalence of smokers, never smokers and quitters by age and gender in the UK population was obtained from the 2009 Office of National Statistics household survey [14]. The relative risk of dying of smokers versus never smokers and quitters by age was retrieved from a study conducted by Doll et al. [15]. These data were combined to calculate the probability of dying for a single individual in the cohort changes within each cycle according to the individual age, gender and smoking status (former smokers, smokers). As for previous studies the formula used was the following:
$$ \begin{aligned} {\text{Mortality rate}}_{ag} = & \, \left( {{\text{Mortality smoker}}_{a} *{\text{Prevalence of smoker}}_{ag} } \right) \, \\ & + \, \left( {{\text{Mortality former smoker}}_{a} *{\text{Prevalence of former smoker}}_{ag} } \right) \, \\ & + \, \left( {{\text{Mortality of never smoker}}_{a} *{\text{Prevalence of never smoker}}_{ag} } \right) \\ \end{aligned} $$
Where a is the age group and g is gender. The estimated mortality rates used to populate the model are reported in Table 3 of the “Appendix”.
Similarly, the probability of experiencing smoking-related diseases is estimated for each gender and age separately using the formula reported below (See Table 4 “Appendix”) [5]:
$$ \begin{aligned} {\text{Disease prevalence}}_{ag} &= \, \left( {{\text{Disease prevalence smoker}}_{ag} *{\text{ Prevalence of smoker}}_{ag} } \right) \, \\ & + \, \left( {{\text{Disease prevalence of former smoker}}_{ag} *{\text{prevalence of former smoker}}_{ag} } \right) \, \\ & + \, \left( {{\text{Disease prevalence of never smoker}}_{ag} *{\text{Prevalence of never smoker}}_{ag} } \right) \\ \end{aligned} $$
As with previous studies we include overall mortality by smoking status and did not consider disease-specific mortality in order to avoid double counting. Diseases within each cycle were assumed to be mutually exclusive (within each 6 months individuals can experience only one of the five diseases, survive with no disease or die). This assumption is consistent with previous studies.
As in Flack et al. [5] and Raikou and McGuire [6], the prevalence rates for lung cancer and COPD are taken from Forman et al. [16] and Britton [17], respectively (See “Appendix”). Prevalence of CHD, MI and stroke are taken from the study by Allender et al. [18] (See “Appendix”). The probability of developing lung cancer by smoking status and gender comes from Peto et al. [19]. while the relative risks of the other smoking-related diseases (CHD, MI, COPD and stroke) are from a study on the health consequences of smoking conducted by the Department of Health and Human Services [7] (See “Appendix”).
Health state values
The health state values assigned to smoking-related diseases and, in absence of these diseases, to smoking status follow Flack et al. [5]. Diseases such as lung cancer, COPD and stroke present several phases of disease progression. For example, Tengs and Wallace [20] identify four health state values according to the type of stroke: minor stroke, moderate stroke, acute stroke requiring hospitalization and major stroke. Similarly, health state values associated with lung cancer are affected by the type of treatment undertaken and the stage of the disease. However, to assign different values according to the severity level of the disease requires knowledge of the proportion of smokers and previous smokers in each of these states. Lacking these data, simple averages of the available values for each of the diseases are used as in previous evaluations [5]. The values used for each disease are: 0.58 for lung cancer, 0.48 for stroke, 0.80 for CHD and MI (the estimate for MI is an average of the values reported by Tengs and Wallace [20] for MI of different disease severities), and 0.73 for COPD (an average of the different values for COPD severity estimated by Rutten-van Molken et al. [21]). Finally, different values are assigned to smokers (0.75) and former smokers (0.78) as reported in the UK study conducted by Tillman and Silcock [22].
Costs
The cost of text-based support per smoker is the sum of three elements: the cost of enrolling smokers (including the cost of collecting information about age, gender education etc.), the cost of text messages (including the cost of setting a short code), and any royalty paid for use of the intervention.
Smokers wishing to use text-based support can register directly online or by SMS. The cost of web site maintenance is assumed to be zero in this analysis because the same site is used for other types of smoking cessation services. The cost of text messages per smoker, £16.12, includes the cost of setting up a short code (£0.06/participant), and the cost of sending the messages (£14.51). The lack of data on the proportion of smokers and former smokers at each disease stage does not allow consideration of how costs vary according to the severity of these diseases. Average cost estimates were used in the absence of these data. For example, in the case of stroke the estimated annual total cost of stroke in UK was divided by the number of people who experienced the disease [5]. The annual costs assigned to each of the smoking-related diseases are lung cancer (£5,921), stroke (£2,218), MI (£2,341), COPD (£997), and CHD (£1,144) [23–27]. All the costs are inflated to 2009–2010 prices using the hospital and community health services price index.
Sensitivity analysis
Deterministic sensitivity analyses and a probabilistic sensitivity analysis (PSA) were performed to assess parameter uncertainty. The impact of variations in the effectiveness of text-based support on cost-effectiveness was investigated by assuming that the relative risk ranged between 1.80 and 2.68 (the 95 % confidence interval around the effect observed in the txt2stop trial). Further analyses were performed to estimate the cost-effectiveness of text-based support for different lifetime relapse rates, 21 %, used as the lower value (reported by McGhan and Smith [28]) and 50 % (the highest value reported in the literature for the relapse rate between 6 and 12 months [29]).
The lower value for the background smoking cessation rate in the one-way sensitivity analysis (1.2 %) is the historic rate over the past 40 years in England, while the upper value (2.8 %) is the highest background cessation rate suggested by West [13]. In the base case analysis, advertising cost is assumed to be zero and it is assumed that 100 % of smokers will register online. It is not known whether the NHS would advertise the intervention using pre-existing channels at relatively low marginal cost or whether advertising on the radio/internet/TV will be utilised. In order to account for this element of uncertainty, the incremental cost-effectiveness of text-based support is estimated assuming an intervention cost ranging from £15 per smoker (assuming that all the smokers register on-line, no crave messages and 10,000 users per short code) to £60.22 (assuming an additional advertising cost of £44, as observed in the txt2stop trial). These figures are used for illustrative purposes. Given large-scale implementation of text-based support, advertising costs are likely to be lower than those incurred when advertising the opportunity to participate in a smoking cessation trial. There are some additional costs which could arise in practice, such as royalty payments for the use of the IT program for the text-based support intervention, and management costs to co-ordinate the provision of the service. Both of these would be influenced strongly by the scale of text-based support were it to be implemented, the larger the scale the lower the cost per smoker. To investigate the impact of these potential additional costs, the analysis was re-run with additional costs of £1 and £35 per smoker enrolled.
A second order Monte Carlo simulation with 1,000 iterations was used to assess the influence of parameter uncertainty on the study results. Parameters were considered independent. Following suggested practice, a lognormal distribution was assigned to relative risks, and beta distributions were assigned to the lifetime relapse rate, baseline quit rate and health state values [30] (See “Appendix”). A gamma distribution was adopted for unit costs. Each variable estimate was derived from its probability distribution (See “Appendix”).
Cost-effectiveness acceptability curves (CEACs) were constructed to represent uncertainty regarding the parameters of the model. The net monetary benefit from text-based support was estimated for each simulation using the following equation:
$$ {\text{Net monetary benefit}} = \lambda *(E_{T2S} - E_{CP} ) - ({\text{COST}}_{T2S} - {\text{COST}}_{CP} ) $$
where: λ represents the “willingness to pay” per QALY gained, (E
T2S − E
CP) is the incremental effectiveness (number of QALY gained) of text-based support, and (COSTT2S − COSTCP) is the incremental cost of text-based support. CEACs plot the proportion of simulations for which text-based support is cost-effective for a willingness to pay per QALY from £0 to £4,000.