A within-trial cost-effectiveness analysis of nicotine preloading

Introduction

The Preloading trial tested the effectiveness of a smoking cessation intervention. Such interventions typically have a very low cost per QALY and are among the most cost-effective in health care.48,49 This is because continued smoking is dangerous but cessation is surprisingly effective at reducing the effects of many previous years of smoking.19,50 As a consequence, one analysis suggests that, at any reasonable cost, improving the 6-month prolonged abstinence percentage by 1% should be cost-effective.20 In this chapter, we examine the cost-effectiveness of preloading relative to the control condition.

Methods

Cost-effectiveness analysis was conducted alongside the Preloading trial. Following NICE guidance, the analysis was performed from the NHS and Personal Social Services perspective, to reflect the NHS England decision-making framework, and on a ITT basis.51 All costs were presented in 2014/15 Great British pounds (£).

Costs

Total costs comprised the intervention costs, NHS Stop Smoking Services costs, AE costs, health resource use costs and medication costs.

Intervention costs

The intervention consisted of 4 weeks of NRT preloading (i.e. the use of NRT before quit day) and two sessions of behavioural support for the preloading group. For the control group, only two sessions of similar-intensity behavioural support were provided. For both arms, the participants were also given a four-page booklet containing information on smoking cessation pertinent to each group.

The NRT used in the intervention was a 21-mg patch every 24 hours provided by GlaxoSmithKline plc (GSK House, Middlesex, UK) at no cost, but, for the purposes of this analysis, we costed this as if it had been purchased by the NHS. The number of patches dispensed was extracted from the trial dispensing form. The cost of NRT for the intervention was calculated by multiplying the amount dispensed to the participants by the weighted average cost per patch. We assumed that the few participants with no record of dispensing had not been given any patches.

The behavioural support sessions were provided as part of the intervention at baseline and at –3 weeks. We included costs only for sessions that were attended so that not everyone incurred the cost of attending a second appointment. The behavioural support sessions were costed by adapting the existing cost template of NHS Stop Smoking Services according to the length of the sessions delivered in the trial.52

NHS Stop Smoking Services costs

After the intervention period, the participants were referred to NHS Stop Smoking Services for standard behavioural support and pharmacotherapies. NHS Stop Smoking Services were unable to supply participants’ attendance data at the NHS Stop Smoking Services behavioural support sessions, so several assumptions were made to estimate the NHS Stop Smoking Services behavioural support cost. We assumed that people had received support if they had an appointment date and a service provider noted on the CRF. People who had no appointment date were assumed not to have attended. If provider data were missing or an appointment was missing, participants’ attendance at NHS Stop Smoking Services was considered missing. Based on the NHS Stop Smoking Services costing report,52 the average number of sessions was multiplied by the inflated cost per session (£16) to calculate the cost of NHS Stop Smoking Services behavioural support sessions for those who attended.

Data on the use of post-quit date pharmacotherapy were collected through self-report at +1 week (1 week after quit date), with respondents reporting the type of medication or could record a response of ‘none’. If no response was recorded, these data on pharmacotherapy were considered missing. The pharmacotherapies were costed using the weighted average cost per prescribed item for the form specified.53 Table 14 lists the weighted average cost per prescription item for each type of pharmacotherapy used in the trial. Because this information was collected only at one time point and no other details were recorded, only one prescription item in the analysis was taken into account.

Adverse event costs

Adverse events resulting in hospitalisation were recorded according to the protocol.26 For the purpose of costing, only the AEs that were determined to be at least possibly related to the intervention were included. The cost of an AE was estimated using the weighted average cost of all Healthcare Resource Groups (HRGs) related to the cause of the hospitalisation.

Health resource use costs

Information on health resource use was collected through a self-reported questionnaire at baseline and 6 and 12 months’ follow-up, covering the 6-month period prior to each time point. The health resources reported included help for smoking cessation outside of the study, hospital visits and use of primary and community services. If an entire section was left blank, the data were coded as missing. If some of the questions were answered and others were left blank, these blanks were assumed to represent nil use of those services. A set of national average costs were applied to the quantities reported by the participants (Table 15). For home visits from all professionals, travel time and expenses were not available. We assumed that the consultation time was the same as that in a general practice; 12 minutes of travel time was added in addition to account for the time spent by the professional travelling to the appointment. However, travel expenses or any other reimbursement were not included. Overheads were included when applicable.

Medication costs

Medication data, excluding smoking cessation medication, were collected alongside the trial by self-report, including medication name, start date and end date. If participants were still taking the medication at the 12-month follow-up, an option of ‘ongoing’ was provided. Owing to the extreme complexity of the records, several simplifying assumptions were made. All medication usage was assigned to the corresponding period based on the following assumptions.

• All medicines recorded, unless they were specified as not prescribed, were assumed to be on prescription.
• Because the number of prescriptions and the amount taken were not recorded, it was assumed that one prescription would last a maximum of 1 month.
• If the start date was missing, there was no reasonable way to make assumptions about usage. The cost of the medicine was, therefore, considered missing.
• If the start date of a medicine was earlier than the quit date, we assumed that the patient had had one prescription of the medicine during the intervention period.
• If a person was recorded as using a medicine at the 12-month follow-up, the cost of this was calculated for the period starting from the recorded start date to the 12-month follow-up date. Depending on the start date, it was then allocated to the intervention period, 6 months post quit date and 12 months post quit date. Otherwise:
  • If the end date was missing, the cost of the medicine was assumed to be missing at 6 months post quit date and 12 months post quit date, because there was no reasonable way to make assumptions about usage.
  • If the end date was not missing, the cost was calculated according to the start and end dates and then allocated to the corresponding period.

As smoking cessation medications specifically were recorded, we excluded these from this medication list. It was also assumed that the participants with no medication recorded, but who had been followed up, had not taken any medication during the trial period.

The costs of medications were estimated by matching their generic name to that in the Prescription Cost Analysis, England 201553 and extracting a weighted average cost per item for that generic medicine. The form and dosage were used when available, but most records did not contain this information and so average costs were used.

Outcome

The outcome used in the cost-effectiveness analysis was 6-month prolonged Russell Standard abstinence, which was the primary outcome of the trial. The 4-week quit rate and 12-month prolonged abstinence, measured according to the Russell Standard, were also used for the secondary analysis to provide comparison with national statistics. The methods of calculation were the same as for the primary statistical analysis.

Analysis

Handling missing data

Following the Russell Standard, it was assumed that people who were missing from follow-up were smoking. Missing data for health resource use costs were handled by multiple imputation using chained equations as the primary method. A set of imputation models was specified, one for each variable with missing data. Each variable was then regressed on to all other variables, including completely recorded variables, and stratified by randomised group. Imputations were performed using predictive mean matching with 10 nearest neighbours for all cost variables on an aggregated level instead of individual usage. For continuous variables, imputations were performed using linear regression. Logistic regression was used for binary variables. Ordered logistic regression was used for categorical variables when the order was meaningful. The analysis models were fitted to each imputed data set separately and the results were estimated using Rubin’s rules.57

Base-case analysis

Combining the costs and outcomes, a cost-effectiveness analysis was performed at 6 months post quit date, producing an incremental cost-effectiveness ratio (ICER) in the form of the incremental cost per additional 6-month abstinence, comparing the intervention group with the control group.

Similar analyses were performed at the 4-week and 12-month follow-ups. The ICER measuring cost per participant who quit for 4 weeks could be used to compare the results of the intervention with national statistics on NHS Stop Smoking Services.

Sensitivity analyses

A non-parametric bootstrap resampling technique was adopted to assess the uncertainty around the primary analysis at 6 months post quit date. Bootstrapping has been proposed as an efficient approach for calculating the confidence limits for the ICER as its validity does not depend on any specific form of underlying distribution.58–61 A cost-effectiveness acceptability curve (CEAC) was plotted based on the outcomes of the 5000 bootstrap iterations.62

A complete-case analysis was performed on the participants who had complete data on costs at both baseline and 6 months’ follow-up, as well as on other baseline characteristics (age, gender, research centre, medical history). This was to examine the impact of multiple imputation.

Results

Costs

Intervention costs

The weighted average cost per patch for all 21-mg nicotine patches was £1.40, extracted from Prescription Cost Analysis, England 2015.53 During the trial, some participants received lower-dosage (14-mg) patches at a later stage of the intervention period. The extracted weighted average cost per unit for all 14-mg nicotine patches was the same as for 21-mg patches. Hence, all nicotine patches dispensed were costed at the same price (£1.40), regardless of the dosage.

Behavioural support sessions, provided by the trial team at baseline and at –3 weeks, were estimated to last 10 minutes per session for both arms. Based on the NHS Stop Smoking Services’ costing report,52 a 30-minute individual session, delivered by a band 5 or 6 smoking cessation adviser, was estimated to cost £14.82. By inflating this cost to 2014/15 prices54 and allocating the cost to a 10-minute session, we estimated the cost per session in the trial at £5.42.

Booklets used in the trial were printed in house and no precise costs were recorded. For the NHS, printing services are either included in overheads or outsourced to professional printing companies. The cost per booklet was estimated to be £0.15, based on an academic institution’s print services price list.63 However, the marginal cost of printing is likely to reduce as the amount of printing increases, so this estimate is probably higher than the cost that would be incurred by the NHS.

In total, the intervention cost amounted to £59 per participant in the intervention group and £11 per participant in the control group (Table 16).

NHS Stop Smoking Services costs

Table 17 shows the utilisation and mean cost of standard NHS Stop Smoking Services behavioural support and pharmacotherapies before the quit date. Of the 813 responding participants in the intervention group, 98.4% indicated that they had received standard NHS Stop Smoking Services behavioural support. In the control group, 97.6% of 798 responding participants indicated that they had received standard NHS Stop Smoking Services behavioural support. A slightly higher proportion of responding participants in the intervention group than in the control group had taken some form of pharmacotherapy as a smoking cessation aid (83.8% of 717 vs. 80.3% of 722). Costs were similar between arms.

Adverse event costs

During the trial, there was one SAE that was identified as possibly being related to the intervention, which occurred in the intervention group. The SAE was a heart attack that required hospitalisation, with the number of bed-days not recorded. The weighted average cost of all HRGs related to myocardial infarction (all complication and comorbidity scores) was included in the analysis to account for the adverse effect of the intervention. This cost was estimated to be £1469 per episode.56

Health resource use costs

Health resource use was recorded in three sections of the CRF: (1) help for smoking cessation, (2) hospital visits and (3) primary and community services. The completeness of these three sections did not vary much at the same time point within arms, indicating that the missing values were probably the result of loss to follow-up. Tables 18–20 summarise the usage of each service and the mean cost of each cost item. The low mean usage and large SDs suggested the uneven use of services among participants in both arms. Although a small group of participants used these services frequently, the majority used them only a little. For instance, at baseline, the maximum number of visits to a GP in the previous 6 months was 26 in both arms. However, over half of the participants visited a GP only once or not at all (54% in the intervention group and 51% in the control group).

The unadjusted mean cost of services among the available cases did not appear to be different between arms at the same time point, except for hospital-related services. The consistently higher mean cost of hospital-related services in the control group was likely to be driven by the longer hospital stays.

Medication costs

Based on the assumptions described earlier in Methods, over half of the participants were prescribed medication during the intervention period. Among the participants providing data, the majority were on medication at each time point (Table 21). The mean cost of medication did not differ much between arms.

Analysis

Multiple imputation

Table 22 shows the missing data percentages for the variables included in the imputation model. Although the primary outcome time point was 6 months, in order to provide as much information as possible some baseline covariates and the 12-month variables are also included. As the highest percentage of missing data was 35.32%, the number of imputations was set to be 35.64

Base-case analysis

At 6 months post quit date, the intervention group had a higher proportion of participants remaining abstinent than the control group (17.5% vs. 14.4%). The difference between arms was 3.1 percentage points (95% CI –0.3 to 6.4 percentage points). The total costs at 6 months post quit date included the intervention costs, NHS smoking cessation support during the intervention period, health resource use costs and medication costs in the 6 months after the quit date. The breakdown of the results is shown in Table 23. Although the unadjusted mean total costs in the intervention group were lower than in the control group, after adjusting for health resource use costs at baseline, age, gender, research centre and presence of medical conditions at baseline, the mean total costs were £4 higher in the intervention group than in the control group (95% CI –£188 to £222). Combining the adjusted difference in mean total costs and difference in abstinence rate at 6 months post quit date, the ICER was calculated as £129 per additional 6-month abstinence. The bootstrapped CI for both the adjusted difference in mean total costs and the ICER showed a wide uncertainty.

In the national statistics for the NHS Stop Smoking Services,65 only the smoking cessation costs per person who quit were reported, including for behavioural support and pharmacotherapy if applicable. For the purpose of comparison, a smoking cessation cost was also calculated by adding the cost of the nicotine patches and/or the behavioural support cost to the NHS Stop Smoking Services cost incurred before the quit date. As expected, owing to the additional cost of the intervention, the intervention group had a higher mean cost than the control group (£182 vs. £133). The cost per participant who was proven to have quit at 4 weeks was £501 in the intervention group and £417 in the control group. The adjusted difference was almost the same as the unadjusted difference, with the intervention group costing £49 more than the control group. The incremental cost per additional participant who was proven to have quit at 4 weeks was, therefore, £1114.

At the 12-month follow-up, the difference between total costs was greater than at the 6-month follow-up, with the costs in the intervention group being much lower than those in the control group (£1557 vs. £2082). The intervention group had lower total costs, by £233, even after adjusting for baseline health resource use costs and other covariates. As abstinence was higher in the intervention group than in the control group (14.0% vs. 11.3%), the results demonstrated a dominance condition, in which the intervention group was less costly and more effective than the control group. Similar to the results at the 6-month follow-up, the bootstrapped CIs showed a wide range of uncertainty.

Sensitivity analyses

To explore the uncertainty surrounding the results, 5000 bootstrap iterations were used to plot the cost-effectiveness plane and CEAC. Figure 6 presents the cost-effectiveness plane, which shows a cluster of ICERs scattering across all four quadrants. Although the overall majority fell to the right side of the zero incremental abstinence rate line, indicating a more effective intervention, over half of those concentrated in the north-east quadrant, indicating a more expensive intervention.

FIGURE 6 Cost-effectiveness plane of incremental costs and incremental abstinence rate at 6 months post quit date, comparing the preloading group with the control group. (Figure not reproduced here; see the original article.)

Figure 7 presents the CEAC to illustrate the bootstrapping iterations. The CEAC shows that the probability of preloading being cost-effective, compared with the control group, increased with the increase in costs (i.e. uncertainty of the results reduced). When the costs reached £3600, the probability of preloading being cost-effective was approximately 80%. For the probability to reach 90%, another £3300 would be required. Soon after this point, the curve became flat, stabilising at about 96%.

FIGURE 7 Cost-effectiveness acceptability curve of the probability of preloading being cost-effective compared with the control group. (Figure not reproduced here; see the original article.)

Complete-case analysis

To assess the impact of multiple imputation on the results, an analysis of the outcomes at the primary outcome time point (6 months post quit date) was conducted on complete cases only. In this analysis, only the participants who had complete data on costs both at baseline and at 6 months’ follow-up, as well as other baseline covariates, were included. Therefore, 462 participants in the intervention group and 426 participants in the control group were not included in the complete-case analysis.

Among the complete cases, the 6-month prolonged abstinence rate was 25.3% in the intervention group and 23.0% in the control group. The difference between arms was slightly smaller than in the imputed data set (2.3 vs. 3.1 percentage points). Although the health resource use cost in the 6 months prior to baseline was similar to that in the imputed data set, the health resource use cost in the 6 months post quit date was much lower than in the imputed data set. In the intervention group, it was £507 among the complete cases, compared with £640 in the imputed data set. The difference was more prominent in the control group, with a cost of £458 among the complete cases and £770 in the imputed data set. This led to a reverse of the relative position. In the imputed data set, the health resource use cost in the 6 months post quit date was higher in the control group, whereas among the complete cases it was higher in the intervention group. The only SAE was excluded in the complete case analysis because of missing data following that event. There was, therefore, no cost for the SAE incurred among complete cases. As a result, the adjusted difference in total costs at 6 months’ follow-up was estimated as £138, which is larger than in the imputed data set. The ICER was calculated as £6000 per additional 6-month prolonged abstinence, with a much wider CI (Table 24).

As the smoking cessation costs before quit date did not differ much between the base case and complete cases, a comparison was made of health resource use costs. In addition, the health resource use cost among the available cases (i.e. the cost variable at the single time point is not missing) was also provided for reference (Table 25). It is apparent that the mean cost drawn from imputed data is closer to that of the available cases. This suggests that the complete-case analysis was likely to have underestimated the mean health resource use cost because of the requirement of completeness on multiple variables, especially for the control group at 6 months’ follow-up.