Motivational support intervention to reduce smoking and increase physical activity in smokers not ready to quit: the TARS RCT

Article history

The research reported in this issue of the journal was funded by the HTA programme as project number 15/111/01. The contractual start date was in May 2017. The draft report began editorial review in June 2021 and was accepted for publication in September 2021. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The HTA editors and publisher have tried to ensure the accuracy of the authors’ report and would like to thank the reviewers for their constructive comments on the draft document. However, they do not accept liability for damages or losses arising from material published in this report.

Copyright statement

Copyright © 2023 Taylor et al. This work was produced by Taylor et al. under the terms of a commissioning contract issued by the Secretary of State for Health and Social Care. This is an Open Access publication distributed under the terms of the Creative Commons Attribution CC BY 4.0 licence, which permits unrestricted use, distribution, reproduction and adaption in any medium and for any purpose provided that it is properly attributed. See: https://creativecommons.org/licenses/by/4.0/. For attribution the title, original author(s), the publication source – NIHR Journals Library, and the DOI of the publication must be cited.

2023 Taylor et al.

Scientific background

Smoking is the main cause of premature death and preventable morbidity in high-income nations. 1 In England, the broader annual cost to society of smoking is about £11B, including £2.5B to the NHS in England. 2 The smoking prevalence rate among the UK population has reduced to 14.7%,3 but proportions vary widely by socioeconomic and mental health status, which contributes to growing health inequalities.

According to the UK’s National Institute for Health and Care Excellence (NICE) Public Health No. 10 guidelines for smoking cessation,4 smokers are recommended to focus on identifying a quit date and abrupt cessation, using pharmacological and motivational support as appropriate. For those not immediately intending to quit or not wanting to quit, the limited evidence suggests that smoking reduction may lead to a greater likelihood of quitting and subsequent successful abstinence. 5–8 A wide range of approaches to reduction, such as using pharmacological and behavioural support and self-initiated approaches, have been suggested. Motivational support appears to have the potential to reduce smoking, and the more that is provided, the greater the likelihood of successful quitting. 9

Smoking reduction studies fall into two types: (1) those involving smokers who want to quit but are willing to reduce first rather than quit abruptly (see Lindson et al. 6 for a review) and (2) those involving smokers who do not want to quit (immediately) but are interested in smoking reduction or harm reduction (see Lindson-Hawley et al. 7 for a review). None of the studies included in these reviews explicitly involved an intervention to promote physical activity (PA) as part of the behavioural support. Most of the ‘motivational-phase intervention’ studies included in the latter review, which involved the administration of nicotine replacement therapy (NRT) in one form or another, and, to a lesser extent, vaping or a pharmacological aid, provided imprecise evidence of effects on smoking cessation or reduction because of limitations in trial methods. Only two studies (deemed to be of low quality) investigated the effects of behavioural support or advice; these also provided imprecise evidence of effects on smoking cessation or reduction. 10,11 Since then, the findings have been published from two US trials involving smokers not motivated to quit. In one, with 560 participants, there was some evidence that three sessions of two different forms of behavioural support (delivered over 4 weeks), compared with usual care, increased quit attempts and 12-month non-carbon monoxide (CO)-verified abstinence. 12 There was some evidence that changes in targeted constructs mediated the intervention effects on specific outcomes. 13 In the other study, with 255 participants, there was some evidence that four sessions of health education [and, less strongly, motivational interviewing (MI)] support increased smoking abstinence at 6 months. 14 Neither of these studies involved the promotion of PA as a way to manage smoking behaviour.

Data from the English Smoking Toolkit Study (2011–14) suggest that 50% of smokers are interested in smoking reduction and approximately 30% of UK smokers reported using e-cigarettes to achieve this. 15 With concerns about the safety of e-cigarettes and the unwillingness of some smokers to use them, there is a need for other evidence-based approaches to self-regulate smoking. 16

Physical activity as an aid for smoking cessation and reduction

There is considerable interest among smokers in using PA to manage smoking acutely and chronically,17,18 with acute moderate-intensity exercise being as efficacious as vigorous exercise. 19–21 However, a systematic review revealed that only one of 24 randomised controlled trials (RCTs) involving smokers attempting to quit showed that an exercise programme can increase abstinence for at least 6 months, relative to a passive control condition. 22 However, most studies were of low quality and focused on efficacy, rather than being pragmatic and offering an acceptable and feasible intervention for smokers wanting to quit more generally.

The logic behind PA being helpful for managing smoking involves both implicit and explicit processes. 17 For example, a smoker could focus on being physically active with associated emotional benefits, which may reduce cognitive and emotional triggers for smoking implicitly. Explicitly, exercise can acutely manage cigarette cravings and withdrawal symptoms,19,20 and chronically manage weight gain following changes in smoking or nudge smokers towards a healthier identity. 23

Prospective population surveys and trials show that weight gain and fear of weight gain can create a reluctance to quit smoking and remain abstinent; this may be especially true among women and initially heavier smokers. 24–26 A meta-analysis study reported an average gain of 4.67 kg [95% confidence interval (CI) 3.96 to 5.38 kg] after 12 months of abstinence. 27 Among a range of options for preventing long-term weight gain after smoking cessation, PA appears to be one of the most promising,28 by increasing energy expenditure and metabolic rate, and also by self-regulation of energy intake, particularly emotional snacking. 29

The potential for the TARS intervention

In a unique randomised pilot trial, Exercise Assisted Reduction then Stop (EARS), there was encouraging support for the short-term effects of a behavioural intervention for increasing PA and smoking reduction on number of cigarettes smoked and abstinence. 23,30 Previous studies13,14,31 have reported that health education, motivational support targeting perceived confidence and self-regulation to manage smoking can facilitate long-term smoking abstinence, and a mix of these approaches was embedded in the pilot intervention. In the EARS pilot trial,23,30 intervention participants had an average of 4.2 sessions, by telephone or face to face, with a health trainer (HT), with a range of 0 to 8 sessions. Compared with the control group, intervention participants were significantly more likely to reduce smoking by at least 50% (39% vs. 20%), to attempt to quit (22% vs. 6%), and to have CO-verified abstinence at 4 weeks (and up to 8 weeks) after their quit day (14% vs. 4%) and at 16 weeks (10% vs. 4%). More participants in the intervention group reported using PA for controlling smoking: 55% versus 22%, and 37% versus 16%, at 8 and 16 weeks, respectively. 23 A health economic analysis estimated that the intervention cost was approximately £192 per participant and preliminary cost-effectiveness modelling indicated that the intervention is cost-effective. 23

Following the encouraging EARS pilot trial,23 the Trial of physical Activity-assisted Reduction of Smoking (TARS) sought to establish the effectiveness and cost-effectiveness of behavioural support for increasing PA and reducing smoking on longer-term abstinence among smokers not immediately ready to quit.

Developing the TARS intervention

Following the work conducted in the EARS pilot trial, we refined the intervention to ensure that it addressed several issues raised in the process evaluation and that it was fully manualised for transparency and to be operational to guide HT training, supervision and intervention delivery across multiple sites. The intervention components and content and the delivery details have been described elsewhere. 32 Adaptations from the EARS pilot trial intervention included the following: (1) accommodating participants who were using or wanted to use vaping to manage smoking and align with guidance and practice in usual care; (2) improving the focus in the training and supervision of HTs on encouraging participants to manage social influence to increase PA and reduce smoking, which appeared to be the least evident HT competency;33 and (3) improving the focus in the training and supervision of HTs on promoting PA as well as smoking reduction (and quitting), which was also identified in an analysis of HT competencies. Additional patient and public involvement (PPI) work was conducted to assess smokers’ views of the intervention and opportunities for further refinement prior to finalising the intervention manual.

The intervention training manual was used to guide training and supervision of eight HTs across four sites. A group training session was held over 3 days for all HTs and their line managers at the University of Plymouth, the University of Oxford, the University of Nottingham and St George’s, University of London. The process evaluation involved a qualitative analysis of interviews with each HT after training to identify gaps in understanding of the intervention and scope for ongoing supervision. All HTs remained in post throughout the intervention delivery phase of the trial. Further interviews took place with all the HTs after they had had an opportunity to rehearse and deliver the early intervention sessions. Supervision sessions led by Tom P Thompson occurred less frequently as the HTs became more familiar with their role and operating procedures.

The trial was managed by the Peninsula Clinical Trials Unit (PenCTU), but a secure parallel bespoke online programme was also developed to manage the flow of intervention participants and their engagement with the HTs. This allowed Tom P Thompson to manage the HT resource (i.e. assign intervention participants to particular HTs at each site, subject to availability) and also view a live record of the number of sessions with intervention participants and HT notes of sessions to facilitate remote supervision.

Aims and objectives

The overarching aim of the TARS was to establish if an individually tailored behavioural intervention for smokers wanting to reduce but not immediately quit provided an effective and cost-effective approach to supporting increases in PA, smoking reduction, number of quit attempts and subsequent prolonged smoking abstinence.

The specific aims of the trial were to establish if the intervention, compared with support as usual (SAU), would:

• increase the proportion of participants achieving CO-verified 6-month floating prolonged abstinence at 9 months post baseline

• increase the proportion of participants reporting a ≥ 50% reduction in the number of cigarettes smoked (between baseline and 3 months, and baseline and 9 months)

• increase the proportion of participants achieving CO-verified 12-month floating prolonged abstinence at 15 months post baseline

• increase self-reported PA at 3 and 9 months post baseline, and accelerometer-assessed PA at 3 months post baseline

• improve body mass index (BMI), quality of life, sleep, cigarette cravings and other beliefs about smoking and PA at 3 and 9 months post baseline.

Further aims were as follows:

• to estimate the intervention, health-care and social care costs, compared with SAU, at 9 months post baseline and determine the cost-effectiveness of the intervention compared with SAU (1) at 9 months and (2) over a longer-term/lifetime horizon

• to conduct an embedded mixed-methods process evaluation to explore the mechanisms of action of the intervention and acceptability of trial processes.

Trial design

The TARS was a multicentred, pragmatic, two-group, parallel, randomised controlled superiority clinical trial. It compared the effects of (1) tailored support to reduce smoking and increase PA with (2) brief advice to reduce or quit smoking on CO-verified 6-month floating prolonged abstinence and other smoking and PA outcomes. The trial included a mixed-methods embedded process evaluation and economic analysis. The trial design is summarised in Figure 1, from Taylor et al. 32

FIGURE 1.

Participant pathway. Reproduced from Taylor et al. 32 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: https://creativecommons.org/licenses/by/4.0/.

Recruitment to the trial took place over 16 months (January 2018 to May 2019). Follow-up measures were conducted at 3, 9 and 15 months post baseline. The 15-month assessment was conducted only with those participants who were CO-verified quitters at 9 months.

Ethics approval and research governance

The trial was approved by the South West – Central Bristol Research Ethics Committee (reference number 17/SW/0223) and the Health Research Authority. The trial was registered with the International Standard Randomised Controlled Trial Number register (reference number ISRCTN47776579) prior to commencement.

Patient and public involvement

A group of current and former smokers was convened at an early stage to act as PPI representatives for the trial. The group reviewed the proposed methods, the intervention and any implications for this trial of the use of e-cigarettes and NRT. Views differed on the merits of e-cigarettes and NRT to reduce smoking, and how various forms of PA may help; this was explored further in the set-up phase of the trial. A PPI representative contributed to the trial via their membership of the TARS Project Management Group and Trial Steering Committee (TSC). Prior to and during intervention development and implementation, the trial team also engaged with key stakeholders involved in commissioning and delivering research-type community interventions outside stop smoking services, to assess where the proposed intervention would best fit and its perceived value.

Patient and public involvement representatives provided valuable input into the training of HTs across each site and built on substantial experience gained during the EARS pilot trial. Notably, HTs rehearsed with PPI smokers how best to approach some of the intervention components that had been less well delivered in the EARS pilot trial, such as self-regulation of PA alongside smoking, integrating the concepts of PA and smoking, and working on social influence on the two behaviours.

Participants and settings

Adult smokers wanting to reduce the amount they smoke, but with no immediate plans to quit smoking, were recruited from general practices, secondary care and community settings around four collaborating university sites in the UK: Nottingham, Oxford, Plymouth, and St George’s, University of London (South London).

Inclusion criteria

• Adult smokers wanting to reduce but not quit smoking in the next month.

• Aged ≥ 18 years.

• Smoke ≥ 10 cigarettes per day for at least 1 year. This was irrespective of use of other nicotine-containing products, for example e-cigarettes and/or NRT products.

• Able to give informed consent.

Exclusion criteria

• Unable to engage in at least 15 minutes of moderate-intensity PA (as judged by the potential participant).

• Any illness or injury that might be exacerbated by exercise.

• Unable to engage in the trial and/or the intervention because of a language barrier or for other reasons (e.g. if the person presents an unacceptable level of risk to the HT or research team members).

Participant identification and approach

The participant pathway is summarised in Figure 1.

A range of methods were employed to identify and approach potential participants, to ensure that the trial was as inclusive as possible:

• An electronic record search using either EMIS Health (EMIS Health, Leeds, UK) or SystmOne (The Phoenix Partnership Ltd, Leeds, UK) at participating NHS general practices, and, in Plymouth, an electronic record search of stop smoking services records followed by an invitation sent by the general practice.

• Opportunistically by NHS primary care practitioners, NHS secondary health-care practitioners, NHS Stop Smoking Services, community-based routes (including pharmacies, dental practices, local businesses, adverts in the local media and via social media). Potential participants were then invited to contact the local researcher for further details about the trial.

The strategy used to approach potential participants via general practices was informed by the findings of a study within a trial (SWAT) conducted in the early phase of recruitment to the TARS;34,35 see Report Supplementary Material 1 for details. In brief, the efficiency and value for money of three different invitation methods were compared. These methods were as follows: (1) a full postal information pack sent to potential participants via Docmail^® (CFH Docmail Ltd, Radstock, UK) (the commercial communication service embedded in the general practice), (2) a single-page postal invitation via Docmail and (3) a text message sent from the general practice. Despite being the most expensive invitation method, the full postal information pack was the most efficient recruitment method; hence, this was predominantly used by general practices thereafter.

Non-responders to the invitation sent by the general practice were initially contacted by the local researcher, either by post (postcard or letter) or via e-mail, telephone call or text message, and (when possible) subsequently telephoned by a member of the general practice staff, so as not to disadvantage those with low literacy levels.

Screening and informed consent

On receipt of an expression of interest, received by post, e-mail, text message or telephone, the local researcher contacted the potential participant by telephone to discuss the trial and assess eligibility. At this point, the trial-specific participant information sheet was provided (by post or e-mail) to those who had not already received this (e.g. those self-referring via the community-based routes). Subsequently, the local researcher confirmed eligibility and sought informed consent from eligible and willing individuals, verbally (in person or by telephone), and a copy of the completed consent form was then posted to the participant and saved in the PenCTU’s participant records. All ineligible participants were advised to seek smoking reduction advice in line with local usual practice.

Intervention

The intervention was initially designed and evaluated for the EARS pilot trial. 23 There was acceptable training and delivery fidelity for the three HTs involved, and acceptable intervention engagement. There were some signals of effectiveness. The need to make some small adaptations was identified and incorporated into the HT manual,36 and into the training and supervision of eight HTs who delivered the intervention across four sites.

The intervention is described in detail elsewhere. 32 Briefly, participants in the intervention arm were offered individually tailored behavioural support from a HT in up to eight weekly sessions, the goal being to build a participant’s motivation and confidence to reduce smoking and increase PA, and possibly to make a quit attempt. An additional six weekly sessions were offered to those participants who wanted support after quitting smoking. The intervention was based on self-determination theory37 and the use of evidence-based health behaviour change techniques. Self-determination theory helps to link the basic human need to feel competent, autonomous and in control with motives that drive human behaviour. The TARS intervention aimed to enhance participants’ sense of competence, control and connectedness related to reducing smoking and increasing PA. The content had some overlap with interventions included in similar studies with a focus on smoking reduction for those smokers not wanting to immediately quit. 9–11,14 The components of the intervention are summarised in Appendix 1 and were used to assess delivery fidelity in the EARS pilot trial. 33

Support as usual

Following randomisation, all participants (intervention and SAU arms) received standardised written guidance on smoking reduction and cessation from the local researcher, with signposting to the support offered at local level (see Appendix 2). In the absence of formal programmes for use of e-cigarettes or licensed nicotine-containing products (LNCPs) to support reduction, for participants not wanting to immediately quit, participants in both arms purchased their own NRT or e-cigarette product.

Randomisation, allocation concealment and blinding

Following completion of baseline measures, participants were randomised by a member of the PenCTU by means of a web-based system created by the PenCTU in conjunction with a statistician independent of the trial team (ensuring allocation concealment from the research staff).

Participants were individually randomised to either the intervention or the control group (1 : 1 ratio) using random permuted blocks, with stratification by recruitment site and the two-item Heaviness of Smoking Index (HSI). 38 For the purposes of stratification, HSI total scores were categorised as 0–4 (low) or 5 and 6 (high).

It was not possible to blind participants to their allocated group. Every effort was made to ensure that the local researchers conducting follow-up assessments remained blind to a participant’s allocation. Participant self-report questionnaire booklets and accelerometers were posted out from, and returned to, the PenCTU without knowledge of the trial arm allocation. However, it was possible that participants disclosed the nature of any support received to reduce smoking during contact with the researcher. In accordance with the statistical analysis plan (SAP), primary analysis of the primary and secondary outcomes was undertaken by the trial statisticians blinded to allocation group, after the 9-month follow-up. Analyses that necessitated unblinding (i.e. repeats of the primary analysis allowing for partial clustering, sensitivity analyses and analysis of 15-month data) were conducted unblinded.

Data collection and management

Data were collected and maintained at the PenCTU in accordance with the current legal and regulatory requirements [the Data Protection Act 1998,39 the General Data Protection Regulation (GDPR) 201640 and later the Data Protection Act 201841].

Participant numbering

Following receipt of an expression of interest, each individual was allocated a unique number and was subsequently identified in all trial-related documentation by their identification number and initials only. A record of names, addresses, telephone numbers and e-mail addresses linked to participants’ identification numbers was stored securely on the trial database at the PenCTU for the purposes of the trial only.

Data processing

All data recorded in case report forms and questionnaire booklets were double-entered by PenCTU staff into a bespoke password-protected Microsoft SQL Server database (Microsoft Corporation, Redmond, WA, USA) and encrypted using Secure Sockets Layer version 3 (QuoVadis Online Ltd, London, UK). Access to identifiable information was restricted and permission based.

A parallel, linked, secure, bespoke, online data system was used to manage intervention engagement post randomisation. This GDPR-compliant system captured all HT attempted and actual contact with intervention participants in real time to produce summary data and to aid supervision sessions and intervention management (e.g. for use should a HT be unavailable).

Raw data from the accelerometer were downloaded using the GENEActiv personal computer software (version 3.0_09.02.2015) (Activinsights Ltd, Kimbolton, UK) and analysed in R (The R Foundation for Statistical Computing, Vienna, Austria) using package GGIR version 1.2-8 (https://cran.r-project.org/web/packages/GGIR/index.html). GGIR performs autocalibration with the reference of local gravity. 42,43 Raw acceleration data are used to compute Euclidean norm minus one [measured in milligravity units (mg)44]. Data were analysed from the first to the final midnight using 5-second epochs.

Non-wear was detected if the standard deviation (SD) of two axes was < 13 mg with a range of < 50 mg in windows of 60 minutes. Time spent in activity intensities was established using published thresholds. 45

Computed variables included average daily moderate to vigorous physical activity (MVPA) accumulated in any 5-second epochs, and diurnal inactivity.

Data cleaning

Local researchers aided the completion of the baseline questionnaire over the telephone or face to face, but at follow-up the questionnaire was posted to participants. Possible confusion over the reporting of grams or ounces of loose tobacco led to some extreme and implausible values in self-reported smoking for some participants. Hence, loose tobacco amounts of ≥ 2 ounces were reassigned as grams (of relevance to 9 and 16 participants at 3 and 9 months, respectively). However, even after this change, at 3 and 9 months, 28 and 23 participants, respectively, smoked the equivalent of > 100 cigarettes, on average, per day over the preceding week. Therefore, data for the total number of daily cigarettes smoked (derived from self-reported number of cigarettes, cigars and amount of loose tobacco) were recalculated using a truncated upper limit of 100 cigarettes per day.

Some self-reported total weekly minutes of MVPA and sleep also included implausible values. In line with other surveys (e.g. International Physical Activity Questionnaire46), we opted to cap all values of MVPA to 1260 minutes per week (equivalent to 3 hours per day). At 3 months and 9 months, data from 43 and 38 participants, respectively, were truncated. Similarly, the sleep data were truncated to no more than a daily average of 12 hours in the previous week: at 3 months and 9 months, 7 and 6 responses, respectively, were truncated as a result. Values of sleep of < 3 hours were set as missing: at 3 months and 9 months, 16 and 19 responses, respectively, were classed as missing.

Baseline assessment

The baseline assessment booklet was administered to consented participants by the local researcher, over the telephone or in person, depending on local practice and a participant’s preference. Baseline measures are summarised in Table 1.

Follow-up assessments

Follow-up assessments, scheduled at 3, 9 and 15 months post baseline, are summarised in Table 1.

At 3 and 9 months, all participants were posted a self-report questionnaire booklet to complete and return to the PenCTU using prepaid, pre-addressed envelopes. See the questionnaire on the project web page (www.journalslibrary.nihr.ac.uk/programmes/hta/1511101/#/documentation). At 15 months, all participants with CO-verified abstinence at 9 months were posted a self-report questionnaire booklet to complete and return to the PenCTU. Participants received a £20 shopping voucher on receipt of a questionnaire booklet at the PenCTU. To increase response rates, motivational postcards (in sealed envelopes) were posted to participants before the questionnaires were sent out. Standard reminder letters were sent and telephone calls made to remind non-responders to return the questionnaire booklet. To maximise follow-up data on key outcomes, non-responders were then given the option to complete only the key questions about smoking behaviour and to submit these responses by e-mail, telephone or text, if preferred.

Participants who self-reported that they had quit smoking were invited to attend a face-to-face visit with the local researcher to verify abstinence. In making arrangements for the visit, the researcher checked whether or not a participant had restarted smoking in the interim; the visit went ahead only with participants who confirmed that they had not smoked a puff in the 7 days preceding the day of the visit. At the visit, abstinence was verified by expired CO, measured using a CareFusion MicroCO Meter (Williams Medical Supplies Ltd, Rhymney, UK). Expired CO of < 10 parts per million (p.p.m. ) indicated abstinence.

At 3 months, an objective measure of total weekly minutes of MVPA was collected in a sample of participants. Participants were asked to wear a waterproof GENEActiv accelerometer51 on the wrist of the non-dominant hand constantly for 1 week and then return the device to the PenCTU. The finite supply of accelerometers determined which participants were posted an accelerometer (i.e. if an accelerometer was available at the PenCTU to be posted, this was posted to the next participant due a 3-month questionnaire).

Process evaluation

A mixed-methods process evaluation focused on trial methods (in an internal pilot study), and if and how the intervention was working, using data from the trial database; audio-recordings of intervention sessions; and audio-recorded and transcribed interviews with participants, HTs, research assistants and general practitioners (GPs)/practice managers (PMs). The methods, procedures, data analysis and findings are reported in more detail in Chapter 4.

Measures

Sample size

The sample size was calculated based on a two-sided Fisher’s exact test. For an increase of 6% in the CO-verified abstinence rate in the intervention group, compared with the SAU group, between 3 and 9 months post baseline [i.e. from 5% in the SAU group to 11% in the intervention group, corresponding to an odds ratio (OR) of 2.35 or a relative risk of 2.2; number needed to treat 17], 450 participants were required per group, and therefore 900 participants were required in total, to detect this difference at the 5% significance level with 90% power (Table 2). The abstinence rates used to calculate the sample size were conservative estimates, consistent with those from the preceding pilot trial23 and those reported from a systematic review of pharmacological interventions. 55,56

As the primary analyses followed the Russell Standard on handling missing cessation outcome data due to loss to follow-up,57 with only participants who were unavoidably lost to follow-up not included (expected to be < 5% of recruited participants), the sample size was not inflated to allow for loss to follow-up.

Table 2 illustrates the range of likely effect sizes detectable, based on recruiting 450 smokers to each group, across plausible values for the control abstinence rate. Under these various scenarios, our sample size would allow us to detect a relative risk ranging, at best, from 1.66 to 2.70.

Statistical methods

The reporting and presentation of this trial is in accordance with Consolidated Standards of Reporting Trials (CONSORT) guidelines. 58,59 The prespecified analyses were detailed in the SAP, approved by the chief investigator and independent statisticians on the TARS oversight committees; see the SAP on the project web page (www.journalslibrary.nihr.ac.uk/programmes/hta/1511101/#/documentation). The primary analyses of the primary and secondary outcomes were performed following the intention-to-treat principle. All analyses were undertaken using Stata^® version 14 (StataCorp LP, College Station, TX, USA), with statistical programming for the primary analyses independently double-coded in R and analyses cross-checked.

Changes to the project protocol

Participant recruitment and flow through the trial

A total of 915 participants were recruited to the trial and randomised between January 2018 and May 2019 (16 months).

Figure 2 shows the CONSORT flow diagram for recruitment to the trial. Of the initial 1441 smokers expressing an interest in participating, 915 (63.5%) individuals were contactable and eligible, and eventually randomised. The main reasons for ineligibility were smoking < 10 cigarettes per day (n = 92, 6.4% of those showing an interest), being unable to engage in at least 15 minutes of moderate-intensity PA (n = 16, 1.1%) and wanting to quit immediately (n = 7, < 1%). We were unable to contact 385 (26.7%) smokers who showed an initial interest in the trial.

FIGURE 2.

The TARS CONSORT flow diagram. Follow-up at 15 months was contingent on CO-verified abstinence at 9 months. Reasons for ineligibility, losses to follow-up and not completing verification of self-reported abstinence are given in the following tables in Appendix 4. a, Table 30; b, Table 31; c, Table 32; d, Table 33; e, Table 34; f, Table 35; g, Table 36; h, Table 37; i, Table 38; j, Table 39. QB, questionnaire booklet.

Appendix 3 shows the number and percentage of participants who entered the trial via the different recruitment methods, by allocated group and overall. The findings from a SWAT to examine the effectiveness and associated costs of different approaches to primary care recruitment (i.e. full Docmail invitation, single letter via Docmail, and text message) have been reported elsewhere34,35 (see also Report Supplementary Material 1). The majority of participants (n = 659, 72.0%) were recruited via a general practice, via invitation letter (which may or may not have contained full details of the trial; some participants received full details and some received a brief version and an invitation to read more on the website or by contacting a researcher), text message or opportunistically.

Of the 915 randomised participants, 624 (68.2%) provided follow-up data at 3 months, and 583 (63.7%) provided follow-up data at 9 months (see Figure 2). There was little difference in the proportions followed up between allocated groups.

Baseline participant characteristics and allocated group comparability

Appendix 5 shows the success of the stratified randomisation algorithm in achieving balance between the allocated groups in terms of the stratification factors (HSI and recruiting site). Overall, the percentages of participants recruited in London, Plymouth, Oxford and Nottingham were 31.1%, 27.0%, 23.7% and 18.1%, respectively. Trial outcomes at baseline and follow-up are reported below.

Table 3 shows the baseline sample demographics and smoking indicators by allocated group and for the whole sample. There was good balance between allocated groups at baseline. The mean age of participants was 49.8 (SD 13.9) years, with a wide range (18–82 years). Overall, 55.4% of participants identified as female and 84.9% identified as being white. The majority of the participants (41.9%) were either married or had a partner. The proportion of participants with an advanced level of education (first or higher degree) was 28.6%, and 45.7% of participants were in paid employment. Just under one-third of participants (32.6%) reported having their first cigarette within 5 minutes of waking. The majority of the participants (59.3%) reported smoking 11–20 cigarettes per day at baseline, with a derived 18.5% having a high baseline HSI. The overall mean number of daily cigarettes (derived from self-reported bought cigarettes, loose tobacco and cigars) smoked at baseline was 18.0 (SD 13.4) (Table 4).

Summary statistics for smoking outcomes, physical activity, sleep and weight/body mass index at baseline and at 3 and 9 months, by allocated group

Table 4 shows descriptive data by allocated group for self-reported smoking level, BMI, weight, and PA and sleep at baseline and at 3 and 9 months, together with accelerometer-recorded PA at 3 months. We do not report the number of cigarettes smoked at 15 months because there are too few participants for meaningful interpretation of these data.

The descriptive data for the total number of cigarettes smoked underwent data cleaning, described in Chapter 2. The summary statistics for the derived daily equivalent total number of cigarettes smoked, prior to truncation, are shown in Report Supplementary Material 2. At baseline, in both the cleaned/truncated and non-truncated data, there was good balance between allocated groups in the total number of cigarettes smoked daily, with a broad range. Among participants providing data at 3 and 9 months, the median number of total cigarettes smoked per day remained fairly stable for participants in the SAU group whereas, on average, intervention participants reduced their consumption compared with baseline. The summary statistics in Table 4 are produced from PA and sleep data that have also undergone data cleaning, described in Chapter 2. Descriptive data before cleaning/truncation for reported PA and sleep are reported in Report Supplementary Material 2.

The descriptive data captured from the separate self-reporting of cigarettes, cigars and loose tobacco consumed are shown in Appendix 6. The percentage of the overall sample who at baseline reported smoking even a puff of bought cigarettes, loose tobacco or cigars was 54.2%, 49.7% and 1.0%, respectively (with some participants smoking more than one tobacco product). Appendix 7 shows the descriptive data when participants with missing data were removed from the calculation of the summary statistics: the percentage of the overall sample who at baseline reported smoking even a puff of bought cigarettes, loose tobacco or cigars was then 64.5%, 62.0% and 1.5%, respectively. The number and percentage of participants who reported using any LNCP at each time point, by allocated group, are also shown in Appendix 7. Additional information about the reported weekly and daily use of LNCPs is shown in Report Supplementary Material 3. At baseline, overall, only 14.0% of participants used any LNCP, increasing to 40.8% and 39.0% overall at 3 months and 9 months, respectively, among participants who were followed up (see Appendix 7) (or 26.0% and 22.8% if we assume that those who were missing at follow-up were not using LNCPs).

Table 5 shows the self-reported, CO-verified point prevalence and 6-month floating prolonged abstinence outcomes across time points by allocated group. At 3 months, 5.5% of participants in the intervention group self-reported being abstinent, compared with 2.8% in SAU group, which dropped to 3.7% and 1.7%, respectively, based on CO-verified abstinence. At 9 months, 8.3% self-reported being abstinent in the intervention group, compared with 7.9% in SAU group, which dropped to 5.5% and 4.6%, respectively, based on CO-verified abstinence. Descriptive data for CO-verified 6-month floating prolonged abstinence (between different time points) are also shown in Table 5. The number and percentage of participants who reported a quit attempt of at least 24 hours, by allocated group, are shown in Appendix 8. At 3 months, 19.2% of responding participants in the intervention group reported a quit attempt, compared with 13.5% in the SAU group. At 9 months, 27.6% of participants in the intervention group who responded reported one or more quit attempts in the previous 6 months, compared with 25.3% in the SAU group.

Analyses of primary outcome

Under the Russell Standard criteria,61 assuming all non-responding participants were still smoking except those irretrievably lost to follow-up (shown in Table 5), the allocated groups contributing to the primary analysis differed in size by only one participant (Table 6). As Table 6 shows, 2.0% (n = 9) of intervention participants had CO-verified 6-month floating prolonged abstinence between 3 and 9 months, compared with 0.9% (n = 4) of SAU participants. From the prespecified primary analysis, the adjusted OR is 2.30 (95% CI 0.70 to 7.56). In other words, the odds of achieving the primary outcome in the intervention group were, on average, more than double the odds in the SAU group. However, the difference was not statistically significant. Table 6 also presents the adjusted absolute difference and relative risk. The unadjusted analyses gave very similar results to the adjusted analyses, across the different methods of estimating the effect of the intervention (see Table 6).

Applying the prespecified best-case scenario to the primary outcome, the direction of effect switched to favouring the SAU group, with 27.5% of participants (n = 124) categorised as being CO-verified 6-month floating prolonged abstainers in the SAU group, compared with 24.2% (n = 109) in the intervention group, with an adjusted OR of 0.84 (95% CI 0.62 to 1.14). This was reflected in all estimates of the intervention effect, both adjusted and unadjusted, and none of these results was statistically significant.

Owing to the sparse primary outcome data, the planned complier-average causal effect analysis, to estimate the intervention effect among participants attending at least two HT sessions, was not performed. Instead, descriptive statistics of the primary outcome by the prespecified categories of engagement are presented in Appendix 10. All nine participants in the intervention group who achieved the primary outcome attended at least two HT sessions.

The SAP prespecified a sensitivity analysis including additional adjustment for selected baseline characteristics if any were imbalanced between allocated groups. As there was no evidence of imbalance, this sensitivity analysis was not required.

The SAP also detailed planned sensitivity analyses to explore the impact of the required changes in verifying self-reported abstinence as a result of COVID-19; however, again because of the sparse primary outcome data, only descriptive statistics are reported. At the 9-month follow-up, 46 CO tests and two saliva tests were carried out among participants who had self-reported being abstinent from smoking. There were 24 out of 25 (96.0%) subsequent CO-verified abstinences in the intervention group and 20 out of 21 (95.2%) in the SAU group. Both saliva cotinine tests (one in each allocated group) verified the self-reported abstinence at 9 months. At the 15-month follow-up, 12 CO tests and nine saliva cotinine tests were carried out in participants who self-reported being abstinent from smoking. All 12 CO tests confirmed self-reported abstinences (eight in the intervention group, four in the SAU group). Of the nine saliva cotinine tests, three out of five (60.0%) confirmed the self-reported abstinences in the intervention group and three out of four (75.0%) confirmed the self-reported abstinences in the SAU group.

Secondary analyses of the primary outcome

The planned formal interaction analyses of the primary outcome with respect to prespecified potential moderating variables were not undertaken as there were too few instances of CO-verified 6-month floating prolonged abstinence between 3 and 9 months. However, the distributions of the 13 participants who were CO-verified abstinent across the prespecified moderating variables are shown in Appendices 11 and 12. The sparseness of the primary outcome data also meant that there was inadequate information to support the planned mixed-effects modelling with partial clustering by HT and the exploration of whether or not the number of HT sessions attended was associated with the primary outcome in the intervention group.

Primary analyses of the secondary outcomes

We described how self-reported smoking data were cleaned/truncated in Chapter 2; descriptive data are presented for cleaned/truncated data in this chapter and for non-truncated data in Appendices 13, 14 and 17, and Report Supplementary Material 2. Sensitivity analyses revealed no effects on the analyses as a result of these data-cleaning procedures.

Table 7 shows the estimated intervention effects on the 3-, 9- and 15-month point prevalence and 6-month floating prolonged abstinence smoking outcomes. Only self-reported point prevalence abstinence at 3 months was (marginally) statistically significant, in favour of the intervention group (5.5% vs. 2.9% for the intervention and SAU groups, respectively). The adjusted OR was 1.99 (95% CI 1.00 to 3.94). This effect was weaker and marginal, again in favour of the intervention group, for CO-verified point prevalence abstinence at 3 months (adjusted OR 2.19, 95% CI 0.93 to 5.14). There was no evidence of an intervention effect on point prevalence abstinence (from self-reported or CO-verified measures) at 9 or 15 months. When expressed as a risk difference, the estimated intervention effect on CO-verified 12-month floating prolonged abstinence between 3 and 15 months was marginal (adjusted analysis: risk difference 0.02, 95% CI 0.00 to 0.04), whereas, despite a point estimate adjusted OR of 6.33, the OR was not statistically significant (95% CI 0.76 to 53.10).

Additional information on the estimated intervention effects on the proportion of participants self-reporting at least one quit attempt of at least 24 hours’ duration, reported at 3 and 9 months, is provided in Report Supplementary Material 4.

Table 8 shows that there was a statistically significant intervention effect in terms of the proportion of participants reducing their smoking level by at least 50%, relative to baseline, at both 3 months (18.9% in the intervention group vs. 10.5% in the SAU group, adjusted OR 1.98, 95% CI 1.35 to 2.90) and 9 months (14.4% in the intervention group vs. 10.0% in the SAU group, adjusted OR 1.52, 95% CI 1.01 to 2.29).

Table 9 shows that there was a statistically significant intervention effect on the total number of cigarettes smoked per day at 3 months (derived from self-reported cigarettes, cigars and loose tobacco smoked), with the intervention group reporting smoking, on average, almost six fewer cigarettes per day than the SAU group, relative to baseline (adjusted mean difference –5.62, 95% CI –9.80 to –1.44). There was no evidence of a difference at 9 months. These conclusions were consistent when the analyses were re-run on the pre-truncated data (see Appendix 13).

Table 10 shows that there was a significant intervention effect on self-reported weekly MVPA at 3 months, but not at 9 months. At 3 months, the intervention group reported doing an average of 82 minutes more MVPA per week (95% CI 29 to 134 minutes) than the SAU group, having adjusted for baseline MVPA and stratification factors, based on the analysis of the truncated data. These conclusions were consistent when the analyses were re-run on the pre-truncated data (see Appendix 14). There was no evidence of a significant between-group difference in accelerometer-recorded MVPA among the subsample of those wearing an accelerometer at 3 months (regardless of whether or not this analysis was adjusted for baseline self-reported MVPA). The correlation between self-reported and accelerometer-measured MVPA (with at least 1 day of 10 hours of wear-time) at 3 months among the subcohort receiving the accelerometer devices was 0.204 (95% CI –0.002 to 0.409). This was too low, according to the prespecified cut-off point (of > 0.3) stipulated in the SAP, to justify further analysis comparing agreement between the two measures of MVPA at 3 months.

As the results in Appendices 15–17 show, there was no evidence of a statistically significant intervention effect on any of the remaining secondary outcomes (LNCP use, BMI or sleep) at 3 or 9 months.

Secondary analyses (potential modifiers) of secondary outcomes

Similarly to the primary outcome, the number of 6-month floating prolonged abstinences between 9 and 15 months was too small to enable us to undertake the planned analysis of interaction effects between the allocated group and the prespecified potential modifiers (e.g. IMD). Interaction analyses were performed for the reduction in smoking outcomes as planned. The potential interaction effects were also explored for any CO-verified 6-month floating prolonged abstinence (i.e. between 3 and 9 months or 9 and 15 months) and point prevalence abstinence secondary outcome measures, as well as self-reported amount smoked daily in equivalent number of cigarettes, LNCP use, MVPA and BMI. The interaction models could not be fitted at all to the prolonged 12-month abstinence outcome owing to the sparseness of the data. Appendix 18 shows the p-values for the interaction effects when the model ran successfully.

None of the interactions between the allocated group and each potential modifier was found to be statistically significant for any of the reduction in smoking outcomes, any of the point prevalence abstinence outcomes or other prolonged abstinence secondary outcomes.

The IMD was found to have a statistically significant modifying effect on the intervention effect for the number of cigarettes smoked daily at both 3 months (p = 0.004) and 9 months (p = 0.001), but in differing ways. To illustrate this modifying effect, the mean numbers of cigarettes smoked daily for those in the 10th and 90th centiles of IMD rank are displayed in Figures 3 (at 3 months) and 4 (at 9 months), with the intervention group in light blue and the SAU group in dark blue. At 3 months, it appears that the intervention was most effective, compared with SAU, for participants in areas with lower IMD values (more deprivation) and became increasingly less effective up the social gradient to a point where it was not more effective than SAU among participants in areas with the least deprivation. At 9 months, the intervention appeared to be more effective than SAU in the most deprived areas, but SAU was more effective than the intervention in the least deprived areas.

FIGURE 3.

Interaction plot of allocated group and IMD rank for number of equivalent cigarettes smoked daily at 3 months. Predictive margins with 95% CIs.

FIGURE 4.

Interaction plot of allocated group and IMD rank for number of equivalent cigarettes smoked daily at 9 months. Predictive margins with 95% CIs.

Introduction

The TARS intervention was designed to promote a decrease in smoking behaviour (cessation and reduction) and an increase in PA via changes in targeted processes outlined in our intervention process model (Figure 5). It was intended that changes in the two behaviours (smoking and PA) would be explicitly and/or implicitly integrated.

FIGURE 5.

Intervention process model.

The process evaluation aimed to:

• assess the extent to which participants engaged with the intervention

• assess whether or not key intervention processes were delivered, received and enacted as intended

• illustrate how (and if) the intervention processes worked

• understand how people engaged with the intended intervention processes.

This mixed-methods process evaluation was designed with multiple complementary workstreams embedded within the delivery of the TARS. The intervention was designed around a process model that embedded a set of intervention components and core competencies (CCs) developed through the pilot trial (see Appendix 1). The model and CCs also underpinned the development of the HT manual;36 the HT training; and the delivery of the intervention, including ongoing supervision.

Methods

The embedded mixed-methods process evaluation was split into two phases: (1) an initial evaluation linked to the internal pilot phase (months 1–4) and (2) the subsequent main trial phase consisting of four workstreams to support the aims of the process evaluation.

The workstreams for the main trial consisted of (1) data related to levels of intervention engagement (aim 1); (2) assessment of delivery, receipt and enactment fidelity (aim 2); (3) mediation analyses of changes in PA and process measures on outcomes (aim 3); and (4) an embedded qualitative study (aims 3 and 4).

Structure of this chapter

A summary of the internal pilot phase is presented, followed by sections for each workstream. Each subsequent section has a brief description of the methods for that section/workstream, followed by the results. The overall discussion brings together the results to describe a narrative of the processes triangulated from multiple data sources.

Internal pilot phase

Intervention engagement

Intervention fidelity

A comprehensive assessment of fidelity, beyond the scope of this report, is being completed by a member of the research team (Jane Horrell) as part of their ongoing Doctor of Philosophy (PhD) course of study. The work will provide a detailed assessment of the five Behavior Change Consortium domains of fidelity (design, training, delivery, receipt and enactment)62 while also developing and testing novel methods for assessing fidelity. Outputs from the PhD will be published elsewhere.

This report provides an assessment of how well the intervention was delivered in line with intended processes (delivery fidelity), how well the participants understood the processes (receipt) and whether or not the participants showed any evidence of enacting the processes and behaviours after receiving the intervention (enactment).

Results

Mean (SD) scores for the 11 CCs drawn from coding all 72 recorded sessions are shown in Appendix 24 for each coder and overall.

Inter-rater reliability across competencies was assessed across all items using a two-way mixed, absolute agreement, intraclass correlation coefficient (ICC). The resulting ICC was in the excellent range (ICC = 0.95), suggesting that coders had a very high level of agreement.

Delivery fidelity scores for the 11 CCs are shown in Figure 6. The mean score for overall delivery was 3.2 (SD 1.4), suggesting overall competent delivery. Active participant involvement (CC 1) was the highest scoring and was delivered at a proficient standard [mean 4.1 (SD 1.2)]. All other competencies were delivered at an advanced beginner/competent level except managing social influence on PA, which was delivered at a novice level [mean 1.7 (SD 1.6)]. The smoking-related competencies (i.e. CCs 2, 4, 6 and 10) tended to be delivered more proficiently than their PA equivalents (i.e. CCs 3, 5, 7 and 11), as shown in Figure 6.

FIGURE 6.

Box plot of fidelity scores by CC. CC 1, active participant involvement; CC 2, build motivation for cutting down; CC 3, build motivation to increase PA; CC 4, self-monitor and set goals to reduce smoking; CC 5, self-monitor and set goals to increase PA; CC 6, review efforts/problem-solving for reducing smoking; CC 7, review efforts/problem-solving for increasing PA; CC 8, integrate idea of changing smoking and PA; CC 9, reinforce health identity shift; CC 10, manage social influences on smoking; and CC 11, manage social influences on PA. Outlier circles are labelled with the corresponding participant identification number.

The mediating role of process measures on physical activity and smoking outcomes, and of physical activity on smoking outcomes

The aims were to examine if changes in PA mediate intervention effects on smoking outcome effects, and if changes in smoking beliefs and PA beliefs mediate intervention effects on smoking and PA behaviour outcomes, respectively.

Methods

To limit the number of mediation analyses, we examined mediation effects only when there were significant (p ≤ 0.05) intervention effects on the outcome. Three exploratory mediation models were considered: (1) the mediating effects of changes from baseline to 3 months in self-reported MVPA on intervention effects on smoking outcomes; (2) the mediating effect of changes in smoking-related process measures on intervention effects on smoking outcomes; and (3) the mediating effect of changes in PA-related process measures on intervention effects on MVPA. The a priori models are represented diagrammatically in Figure 7.

FIGURE 7.

Mediation analyses diagrams.

The candidate mediators from the smoking process measures comprised the single items:

• importance of reducing

• importance of quitting

• confidence in reducing

• confidence in quitting

• support

• quitting planning

• frequency of urges to smoke

• strength of urges to smoke.

They also comprised additional multi-item scales:

• action planning (two items) for changes in smoking

• coping planning (two items) for changes in smoking

• self-monitoring (two items) of smoking.

The candidate mediators from the PA process measures comprised the single items:

• importance of PA

• confidence

• support

• PA for controlling smoking.

They also comprised additional factors constructed from single items:

• action planning (derived from three items) for changes in PA

• coping planning (derived from two items) for changes in PA

• self-monitoring (derived from two items) of PA.

The mediation analysis was carried out using structural equation models:66 this simultaneously fitted two regression equations. In one, the mediator (change in process measure from baseline to 3 months) was regressed on the intervention, and adjusted for baseline values of the process measure and stratification variables, with the intervention effect on the mediator providing the coefficient for the ‘A path’. In the other, the outcome was regressed on the mediator and on the intervention, and adjusted for baseline values of the outcome, when appropriate, and stratification variables. This part of the structural equation model provided an estimate of the natural direct effect of the intervention adjusted for the effect of the mediator. The latter path may be considered the ‘B path’, and the product of this and the ‘A path’ provided an estimate of the natural indirect effect, the so-called mediated effect, of the intervention on the outcome. The remaining direct influence of the intervention on the outcome is the ‘C path’. Because the sampling distribution of the product of the two coefficients can be assumed to be normally distributed only in certain cases, the calculation of the standard error may lead to lower power in the test of the product. 67,68 Therefore, the CIs for the mediated path were estimated through the bootstrap resampling method, with 1000 replications.

Results from the mediation analysis

There was no evidence from the mediation analysis that self-reported changes in PA mediated changes in smoking-related outcomes at 3 or 9 months, as shown in the bottom rows of Tables 16–18.

TABLE 16 - Mediation analysis of smoking process measures and self-reported MVPA as mediators of self-reported number of cigarettes smoked daily at 3 months

SE, standard error.

Note

Bold indicates effects significant at the 5% level.

Mediator	n	A path, coefficient (SE)	B path, coefficient (SE)	Mediated effect		C path, coefficient (SE)
				Coefficient (SE)	95% CI
Important I reduce my smoking	584	0.201 (0.060)	–3.526 (1.328)	–0.709 (0.407)	–1.690 to –0.101	–5.081 (2.145)
Confident I can reduce my smoking	582	0.802 (0.083)	–1.922 (0.912)	–1.541 (0.694)	–3.071 to –0.243	–4.184 (2.262)
Confident I can quit smoking	585	0.489 (0.080)	–0.132 (0.980)	–0.065 (0.473)	–0.930 to 0.871	–5.488 (2.188)
People close to me support me reducing smoking	581	0.420 (0.084)	–1.636 (0.870)	–0.687 (0.412)	–1.595 to 0.077	–4.960 (2.183)
Action planning for changes in smoking	573	1.554 (0.165)	–1.220 (0.421)	–1.896 (0.711)	–3.471 to –0.689	–3.754 (2.251)
Coping planning for changes in smoking	568	1.343 (0.167)	–0.905 (0.447)	–1.215 (0.575)	–2.398 to –0.142	–4.965 (2.232)
Self-monitoring of smoking	568	0.981 (0.135)	–1.586 (0.505)	–1.556 (0.519)	–2.736 to –0.660	–3.715 (2.257)
Plans for quitting smoking	575	0.483 (0.099)	–0.941 (0.761)	–0.455 (0.348)	–1.207 to 0.157	–5.292 (2.192)
Urge to smoke in previous week	583	–0.553 (0.084)	0.890 (0.876)	–0.492 (0.540)	–1.706 to 0.422	–5.145 (2.176)
Strength of urges to smoke in previous week	585	–0.466 (0.081)	0.959 (0.925)	–0.447 (0.437)	–1.358 to 0.430	–5.164 (2.172)
Standardised difference in self-reported MVPA (truncated), baseline to 3 months	608	0.195 (0.064)	0.841 (1.102)	0.164 (0.265)	–0.228 to 0.943	–5.470 (2.162)

TABLE 17 - Mediation analysis of smoking process measures and self-reported MVPA as mediators of ≥ 50% reduction in self-reported smoking at 3 months

SE, standard error.

Note

Bold indicates effects significant at the 5% level.

Mediator	n	A path, coefficient (SE)	B path, coefficient (SE)	Mediated effect		C path, coefficient (SE)
				Coefficient (SE)	95% CI
Important I reduce my smoking	584	0.201 (0.060)	0.091 (0.126)	0.018 (0.026)	–0.020 to 0.086	0.725 (0.208)
Confident I can reduce my smoking	582	0.802 (0.083)	0.272 (0.087)	0.218 (0.072)	0.083 to 0.362	0.498 (0.217)
Confident I can quit smoking	585	0.489 (0.080)	0.229 (0.091)	0.112 (0.055)	0.015 to 0.234	0.591 (0.210)
People close to me support me reducing smoking	581	0.420 (0.084)	0.011 (0.081)	0.005 (0.034)	–0.062 to 0.072	0.685 (0.210)
Action planning for changes in smoking	573	1.554 (0.165)	0.150 (0.042)	0.233 (0.070)	0.110 to 0.380	0.498 (0.218)
Coping planning for changes in smoking	568	1.343 (0.167)	0.154 (0.044)	0.207 (0.068)	0.085 to 0.353	0.529 (0.217)
Self-monitoring of smoking	568	0.981 (0.135)	0.156 (0.049)	0.153 (0.057)	0.059 to 0.290	0.480 (0.216)
Plans for quitting smoking	575	0.483 (0.099)	0.188 (0.073)	0.091 (0.039)	0.026 to 0.177	0.641 (0.212)
Urge to smoke in previous week	583	–0.553 (0.084)	–0.087 (0.079)	0.048 (0.046)	–0.037 to 0.146	0.637 (0.209)
Strength of urges to smoke in previous week	585	–0.466 (0.081)	–0.076 (0.082)	0.035 (0.041)	–0.039 to 0.121	0.645 (0.208)
Standardised difference in self-reported MVPA (truncated), baseline to 3 months	608	0.195 (0.064)	0.044 (0.102)	0.009 (0.021)	–0.032 to 0.053	0.648 (0.205)

TABLE 18 - Mediation analysis of smoking process measures and self-reported MVPA as mediators of ≥ 50% reduction in self-reported smoking at 9 months

SE, standard error.

Note

Bold indicates effects significant at the 5% level.

Mediator	n	A path, coefficient (SE)	B path, coefficient (SE)	Mediated effect		C path, coefficient (SE)
				Coefficient (SE)	95% CI
Important I reduce my smoking	584	0.201 (0.060)	0.179 (0.151)	0.036 (0.032)	–0.012 to 0.110	0.380 (0.237)
Confident I can reduce my smoking	582	0.802 (0.083)	0.184 (0.099)	0.148 (0.079)	0.011 to 0.330	0.257 (0.246)
Confident I can quit smoking	585	0.489 (0.080)	0.129 (0.104)	0.063 (0.066)	–0.057 to 0.198	0.268 (0.237)
People close to me support me reducing smoking	581	0.420 (0.084)	0.065 (0.092)	0.027 (0.043)	–0.061 to 0.110	0.287 (0.236)
Action planning for changes in smoking	573	1.554 (0.165)	0.033 (0.045)	0.051 (0.075)	–0.090 to 0.197	0.295 (0.244)
Coping planning for changes in smoking	568	1.343 (0.167)	0.040 (0.048)	0.054 (0.069)	–0.081 to 0.199	0.273 (0.244)
Self-monitoring of smoking	568	0.981 (0.135)	0.096 (0.056)	0.094 (0.053)	0.002 to 0.205	0.170 (0.246)
Plans for quitting smoking	575	0.483 (0.099)	0.032 (0.081)	0.015 (0.039)	–0.060 to 0.098	0.301 (0.236)
Urge to smoke in previous week	583	–0.553 (0.084)	–0.160 (0.090)	0.088 (0.051)	–0.002 to 0.199	0.254 (0.237)
Strength of urges to smoke in previous week	585	–0.466 (0.081)	–0.117 (0.093)	0.055 (0.048)	–0.033 to 0.151	0.280 (0.236)
Standardised difference in self-reported MVPA (truncated), baseline to 3 months	608	0.195 (0.064)	–0.056 (0.116)	–0.011 (0.024)	–0.068 to 0.033	0.251 (0.232)

Changes between baseline and 3 months in 5 of the 10 smoking process measures mediated intervention effects on the self-reported number of cigarettes smoked daily at 3 months, and in 6 of the 10 measures mediated intervention effects on whether or not participants reduced their smoking by ≥ 50% from baseline to 3 months. Increasing beliefs in being able to reduce smoking, action planning, coping with setbacks and self-monitoring smoking all mediated intervention effects on both the number of cigarettes smoked daily and ≥ 50% reduction up to 3 months. At 9 months, there were no mediating effects on smoking reduction other than confidence in reducing and self-monitoring, which had weakened and become marginally significant since the 3-month visit.

Changes in both confidence in being physically active and self-monitoring mediated intervention effects on self-reported MVPA at 3 months, as shown in Table 19.

TABLE 19 - Mediation analysis of PA process measures on self-reported MVPA at 3 months

SE, standard error.

Note

Bold indicates effects significant at the 5% level.

Mediator	n	A path, coefficient (SE)	B path, coefficient (SE)	Mediated effect		C path, coefficient (SE)
				Coefficient (SE)	95% CI
Important I am physically active	613	0.080 (0.056)	16.302 (15.180)	1.304 (1.651)	–0.389 to 7.581	78.157 (26.947)
Confident I can be physically active	617	0.208 (0.068)	34.939 (13.544)	7.267 (3.432)	1.963 to 16.121	72.658 (26.905)
People close support me being physically active	612	0.128 (0.074)	22.219 (11.394)	2.844 (2.189)	–0.243 to 9.234	76.806 (26.805)
PA is a way of controlling smoking	616	0.476 (0.085)	17.749 (10.045)	8.449 (5.448)	–0.794 to 20.625	70.974 (27.402)
Action planning for changes in PA	599	1.270 (0.231)	6.828 (3.598)	8.672 (5.057)	–0.648 to 19.398	72.325 (27.962)
Coping planning for changes in PA	603	0.784 (0.155)	5.543 (5.497)	4.346 (4.583)	–5.664 to 13.551	78.971 (27.682)
Self-monitoring of PA	598	0.854 (0.152)	12.957 (5.817)	11.065 (5.421)	1.491 to 23.460	68.012 (27.957)

Qualitative study

The embedded qualitative study aimed to promote understanding from the perspectives of both the participants and the HTs as to how the key intervention processes portrayed in the intervention process model (see Figure 5) were being used, and which offered the most benefit, and to identify any possible changes or areas where the intervention processes were not working and could be improved.

Discussion

The internal pilot phase of the process evaluation demonstrated acceptability of the intervention and trial methods, which was supported by the trial subsequently exceeding its recruitment target and demonstrating good levels of intervention engagement. Overall, participants in the intervention attended an average of 4.8 sessions across all sites, with 76% of participants receiving the a priori ‘dose’ of the intervention (i.e. two or more sessions), which exceeded the levels observed in the pilot trial. 30 The high levels of engagement support the intervention design and delivery for engaging with people who smoke but do not want to immediately quit. However, as shown by the primary statistical analysis and supported by the qualitative work of this process evaluation, the intervention had less of an effect than expected on smoking cessation induction. Although the intervention successfully supported a reduction of smoking (evidenced from all data sources), it failed to translate this into increased smoking cessation, and the qualitative work demonstrated barriers to this in terms of the participants (e.g. low and unchanged importance of quitting smoking, motivation linked solely to reduction), the HTs (not feeling confident to push the message of cessation and concerns about undermining achievements related to reduction) and the intervention design (e.g. described as not long enough to affect changes related to cessation). The EARS pilot trial excluded people wanting to use LNCPs, but with the increased prevalence of LNCP use over time, participants were not excluded for wanting to use LNCPs in the TARS. The HTs were trained to provide advice on and referrals to other services should an individual be interested. However, other services were not a formal part of the intervention. Limited qualitative data showed that this may have been a valued addition to the intervention. There was an overall increase in the proportion of participants who reported using LNCPs at baseline from 14.0% to 40.8% and 39.0% at 3 and 9 months, respectively (based on a denominator of number of participants followed up) (or 26.0% and 22.8%, assuming that those who were missing at follow-up were not using LNCPs, at least among those followed up). However, there were no significant differences in LNCP use between the groups at the follow-up assessments and LNCP use did not modify the intervention effects for any of the smoking outcomes, albeit the data were too sparse to fully investigate this.

There appeared to be some variation between sites in terms of number and duration of sessions provided, but the main statistical analysis showed no effect of site, or whether or not at least two sessions took place with a HT, on intervention effects, suggesting that variation in intervention dose did not have an impact on its effects. Further work is needed to understand if delivery fidelity (i.e. the standard and quality of delivery), and not just the number and duration of sessions, influenced smoking outcomes; this will be completed as part of Jane Horrell’s PhD.

Fidelity assessment demonstrated that the core components of the intervention were delivered to a predefined level of acceptability overall, but smoking-related HT competencies were delivered to a higher standard than the related competencies for PA. Results from enactment fidelity assessment indicate that behaviour change relating to PA was also lower than that relating to smoking reduction. These findings were also reflected by the results from the delivery and receipt fidelity checks. The PA element may have been delivered to a lesser extent because a proportion of participants were less motivated and/or amenable to increasing their PA, and therefore influenced the HT to be more smoking focused. This difference was also supported in the qualitative work, which highlighted some of the different challenges and barriers faced in addressing the different behaviours. A similar difference was found in the fidelity assessment of the pilot trial,33 and although the standard of delivery did improve comparatively (following learning from the pilot trial, including adaptations in training and delivery) in the current trial, and although the difference in standard of delivery between smoking and PA components is not as pronounced, more investigation of the barriers to addressing PA alongside smoking behaviour is needed.

Changing the two behaviours simultaneously did not appear to be a barrier for those participants who were able to increase their PA levels and open to increasing them, and there is evidence that this operated on both practical (distraction from cravings, removal from the opportunity to smoke) and psychological (feeling healthier) levels. As shown in the fidelity and qualitative work, motivation, past experience of PA and perceived fitness influenced participants’ willingness to incorporate PA into their behaviour change goals. Related to this, there was some evidence from the qualitative data that experiencing the incompatible nature of the two behaviours (for those who engaged in PA) did drive motivation for changing smoking behaviour and could lead to multiple behaviour change. Data from the process measures collected at baseline and 3 months also supported an increase in participants’ beliefs about using PA as a way to control smoking behaviour, and qualitative data demonstrated the perceived utility of using PA to manage smoking behaviour; however, changes in PA were shown not to mediate smoking outcomes, further demonstrating the complex nature of the two behaviours.

Active participant involvement, and the non-judgemental and empowering client-led nature of the intervention (drawing on MI techniques) were shown to be delivered to a high standard and to be fundamentally important parts of the intervention for both initiating and maintaining engagement with participants and effecting change. Linked to this, the focus on reduction rather than cessation promoted a sense of confidence among participants who may have previously had negative experiences with traditional cessation services, including feeling judged should they fail to quit.

The data showed that the intervention successfully increased participants’ self-reported levels of confidence for both changing PA and reducing smoking, as well as showing that it increased perceived importance of reducing smoking. However, it did not support significant changes in the perceived importance of quitting smoking or in the importance of PA. These data support both the fidelity work and the qualitative work that highlighted difficulties with progressing to cessation, as well as some resistance and difficulty in delivering and supporting changes in PA. This could be a function of the intervention (which was not designed around a ‘fear’ message for either behaviour) of the focus of the intervention on reduction rather than cessation and of PA being seen as secondary target for change after smoking reduction.

The process data demonstrated significant changes at 3 months in all but one process measure related to smoking, suggesting that the intervention successfully supported change in the targeted processes. For PA, all but two process measures showed significant improvement. Those processes that failed to improve were again linked to perceived importance, and to that of social support for PA. Managing social influence for PA was the least well-delivered component of the intervention, which may explain the failure to observe improvements in the process data.

Self-monitoring of behaviour was identified as an important process for changing both behaviours, and qualitative data supported it as important from the perspectives of both the participants and the HTs. Three-month process data showed that intervention participants successfully increased the amount of self-monitoring for both behaviours, and self-monitoring was shown to mediate the effects on smoking reduction and PA at 3 months, highlighting it as a fundamental process in the intervention.

It was repeatedly observed in the data that the duration of the programme was not considered long enough by either the participants or the HTs. Although the mediation analyses demonstrated some success in changes in the processes mediating the behaviours at 3 months, this was not observed at 9 months. It is therefore reasonable to suggest that a longer intervention may produce longer-lasting effects. Linked to this, the qualitative data showed that some participants and HTs did not think 8 weeks was long enough to get somebody who does not want to quit initially to a point where they are motivated to quit, and booster sessions would have been valued to increase rates of cessation.

Overall, there was evidence that the components of the intervention were delivered to an acceptable level and that the intervention successfully resulted in positive change in most of the key psychological and behavioural processes, although this did not translate into increased rates of quitting. There was difficulty in progressing participants to cessation, and PA was often shown to be a ‘secondary’ target after managing smoking consumption, as demonstrated by relatively poorer delivery, fewer successful changes in the process data and support for smoking reduction being the primary target in the qualitative data. Further work is needed to understand if those participants who reduced their smoking up to 3 months were also more likely to report a quit attempt and point prevalence abstinence (at 3 months) from TARS data, in line with other studies. 5,9

Introduction

In this chapter, we present an estimation of the direct cost associated with the delivery of the TARS intervention, the incremental impacts of the intervention on costs and quality of life as observed in the TARS, and the translation of the primary clinical effectiveness outcome (smoking cessation) on lifetime costs and quality-adjusted life-years (QALYs). Two cost-effectiveness analyses are presented, jointly answering the following research questions:

• What is the estimated resource use and cost associated with the TARS intervention?

• Is the TARS intervention a cost-effective use of NHS resources, versus brief advice (control), based on economic outcomes (health and social care resource use and quality of life), as measured in the TARS?

• Is the TARS intervention a cost-effective use of NHS resources, versus brief advice (control), over the long term, based on the effectiveness of the TARS on smoking cessation when modelling the effectiveness, lifetime costs and QALYs?

In all analyses, costs are given in 2018/19 Great British pounds. An NHS and Personal Social Services (PSS) perspective is used.

Estimating the cost of delivering the TARS intervention

The TARS intervention has been described in detail in Chapter 2. In the TARS RCT, data have been collected to measure the resource use associated with delivering the intervention, compared with brief advice (control).

The main components of resource use for the TARS intervention, informed by the pilot trial,23 are as follows:

• HT time, which comprises contact time with clients as well as non-contact time directly attributable to the TARS intervention

• training and supervision of HTs to support the delivery of the TARS intervention

• venues/locations for face-to-face contacts with clients and HT non-contact time

• pedometers and notebooks given to all clients.

Trial-based cost-effectiveness analysis of the TARS intervention

Results

Costs

Descriptive statistics for resource use and costs across the 9-month follow-up period are given in Tables 23 and 24. These statistics are for all available cases, that is participants for whom resource use data at 3 and 9 months were available. Descriptive statistics for resource use and costs at each individual time point (baseline, 3 months and 9 months) are given in Appendix 30, NHS and Personal Social Services resource use.

TABLE 23 - Health and social care resource use: number of contacts over the 9-month follow-up

A&E, accident and emergency.

Note

Includes resource use measured at 3 and 9 months post randomisation.

Item	SAU group		Intervention group
	Participants (n)	Mean (SD) [range]	Participants (n)	Mean (SD) [range]
Primary care and community services
GP at surgery/health centre	241	4.87 (6.70) [0–47]	248	3.80 (3.82) [0–24]
GP via telephone	241	1.68 (3.53) [0–30]	248	1.56 (2.66) [0–18]
GP at home	241	0.07 (0.64) [0–7]	248	0.00 (0.00) [0–0]
Nurse at surgery/health centre	241	1.59 (2.68) [0–20]	248	1.56 (3.04) [0–26]
Nurse via telephone	241	0.27 (1.47) [0–15]	248	0.13 (0.58) [0–6]
Nurse at home	241	0.01 (0.09) [0–1]	248	0.01 (0.14) [0–2]
Physiotherapist at surgery/health centre	241	0.81 (2.91) [0–29]	248	0.62 (1.80) [0–12]
Physiotherapist at home	241	0.01 (0.09) [0–1]	248	0.01 (0.11) [0–1]
Occupational therapist at surgery/health centre	241	0.22 (1.12) [0–12]	248	0.15 (0.79) [0–8]
Occupational therapist at home	241	0.06 (0.45) [0–5]	248	0.15 (1.63) [0–24]
Social worker	241	0.35 (2.42) [0–27]	248	0.34 (2.73) [0–40]
Care worker	241	0.66 (8.04) [0–124]	248	3.46 (26.00) [0–238]
NHS Stop Smoking Service	241	0.67 (1.92) [0–13]	248	0.81 (2.30) [0–15]
Walk-in centre	241	0.19 (0.96) [0–9]	248	0.28 (0.99) [0–9]
Secondary care services
Outpatient attendances	241	1.74 (3.23) [0–26]	247	2.13 (5.78) [0–60]
Overnight stays	241	0.17 (0.70) [0–7]	248	0.14 (1.06) [0–16]
Inpatient days/nights	241	0.78 (5.29) [0–66]	248	0.17 (0.86) [0–9]
A&E attendances	241	0.31 (1.03) [0–9]	248	0.29 (0.90) [0–8]
Day-case procedures	241	0.57 (1.54) [0–15]	247	0.59 (1.70) [0–18]

TABLE 24 - Costs (£, 2018/19) of health and social care resource use, over the 9-month follow-up

A&E, accident and emergency.

Note

Includes resource use measured at 3 and 9 months post randomisation.

Item	SAU group		Intervention group
	Participants (n)	Mean (SD) [range]	Participants (n)	Mean (SD) [range]
Primary care and community services
GP at surgery/health centre	241	190.91 (262.44) [0–1842]	248	149.02 (149.60) [0–941]
GP via telephone	241	26.08 (54.82) [0–466]	248	24.16 (41.28) [0–279]
GP at home	241	5.24 (50.67) [0–552]	248	0.00 (0.00) [0–0]
Nurse at surgery/health centre	241	19.75 (33.30) [0–249]	248	19.40 (37.82) [0–323]
Nurse via telephone	241	2.07 (11.49) [0–117]	248	1.04 (4.51) [0–47]
Nurse at home	241	0.33 (3.61) [0–40]	248	0.48 (5.63) [0–79]
Physiotherapist at surgery/health centre	241	50.89 (182.99) [0–1824]	248	38.81 (113.38) [0–755]
Physiotherapist at home	241	0.52 (5.72) [0–63]	248	0.76 (6.89) [0–63]
Occupational therapist at surgery/health centre	241	17.95 (93.11) [0–998]	248	12.07 (65.41) [0–665]
Occupational therapist at home	241	5.18 (37.22) [0–416]	248	12.74 (135.97) [0–1996]
Social worker	241	22.98 (159.53) [0–1780]	248	22.33 (180.15) [0–2638]
Care worker	241	7.12 (86.72) [0–1338]	248	37.37 (280.51) [0–2568]
NHS Stop Smoking Service	241	24.50 (52.69) [0–137]	248	28.24 (55.62) [0–137]
Walk-in centre	241	4.01 (20.25) [0–189]	248	5.93 (20.80) [0–189]
Secondary care services
Outpatient attendances	241	221.07 (409.25) [0–3298]	247	269.62 (733.58) [0–7611]
Overnight stays	241	65.53 (267.80) [0–2696]	248	54.36 (408.66) [0–6163]
Inpatient days/nights	241	393.13 (2663.80) [0–33,261]	248	83.32 (432.86) [0–4536]
A&E attendances	241	56.89 (187.94) [0–1645]	248	53.07 (164.13) [0–1462]
Day-case procedures	241	427.43 (1157.86) [0–11,279]	247	444.44 (1275.31) [0–13,534]

The most apparent differences in resource use [mean number of contacts (mean cost)] between the groups are as follows:

• GP at surgery/health centre [3.80 (£149) in TARS intervention group vs. 4.87 (£191) in control group]

• care worker [3.46 (£37) in TARS intervention group vs. 0.66 (£7) in control group]

• outpatient attendances [2.13 (£270) in TARS intervention group vs. 1.74 (£221) in control group]

• inpatient days/nights [0.17 (£83) in TARS intervention group vs. 0.78 (£393) in control group].

We have not examined the statistical significance of any of these comparisons because of concerns of multiple testing.

Outcomes

Sustained quits to 9 months (primary trial outcome)

As shown in Chapter 3, there was a relative risk of sustained abstinence at 9 months of 2.267 (95% CI 0.705 to 7.288), and an absolute risk difference of 0.011 (95% CI −0.004 to 0.027).

EuroQol-5 Dimensions, five-level version

The EQ-5D-5L measurements are summarised in Table 25 for all available cases. There is a general pattern that the proportion of respondents reporting no problems on the different domains decreases from baseline. This may be due to reversion to the mean following recruitment to the trial or could be due to differential non-response at subsequent time points. Self-assessed health by the visual analogue scale suggests that there may have been some improvement in health-related quality of life in the TARS intervention group.

TABLE 25 - Descriptive data for the EQ-5D-5L dimensions at baseline and at 3 and 9 months

EQ VAS, EuroQol Visual Analogue Scale.

Health-related quality of life	SAU group			Intervention group
	Baseline	3 months	9 months	Baseline	3 months	9 months
EQ-5D-5L dimensions, n (%)
Mobility (N)	453	300	271	453	310	281
1, n (%)	344 (76)	188 (63)	167 (62)	322 (71)	201 (65)	165 (59)
2, n (%)	58 (13)	57 (19)	48 (18)	66 (15)	52 (17)	50 (18)
3, n (%)	37 (8)	34 (11)	41 (15)	54 (12)	37 (12)	42 (15)
4, n (%)	13 (3)	21 (7)	14 (5)	8 (2)	18 (6)	22 (8)
5, n (%)	1 (0)	0 (0)	1 (0)	3 (1)	2 (1)	2 (1)
Self-care (N)	453	299	269	453	310	279
1, n (%)	409 (90)	242 (81)	208 (77)	411 (91)	271 (87)	232 (83)
2, n (%)	23 (5)	25 (8)	35 (13)	29 (6)	22 (7)	24 (9)
3, n (%)	12 (3)	18 (6)	18 (7)	9 (2)	15 (5)	18 (6)
4, n (%)	7 (2)	13 (4)	7 (3)	3 (1)	2 (1)	4 (1)
5, n (%)	2 (0)	1 (0)	1 (0)	1 (0)	0 (0)	1 (0)
Usual activities (N)	453	299	272	453	310	281
1, n (%)	349 (77)	163 (55)	141 (52)	346 (76)	187 (60)	152 (54)
2, n (%)	49 (11)	71 (24)	66 (24)	58 (13)	66 (21)	68 (24)
3, n (%)	41 (9)	38 (13)	37 (14)	39 (9)	43 (14)	43 (15)
4, n (%)	9 (2)	25 (8)	25 (9)	8 (2)	12 (4)	13 (5)
5, n (%)	5 (1)	2 (1)	3 (1)	2 (0)	2 (1)	5 (2)
Pain/discomfort (N)	453	301	274	453	314	281
1, n (%)	217 (48)	97 (32)	96 (35)	221 (49)	112 (36)	92 (33)
2, n (%)	124 (27)	104 (35)	94 (34)	120 (26)	103 (33)	97 (35)
3, n (%)	78 (17)	52 (17)	47 (17)	86 (19)	68 (22)	57 (20)
4, n (%)	26 (6)	30 (10)	27 (10)	16 (4)	24 (8)	22 (8)
5, n (%)	8 (2)	18 (6)	10 (4)	10 (2)	7 (2)	13 (5)
Anxiety/depression (N)	455	301	273	456	314	281
1, n (%)	215 (47)	101 (34)	84 (31)	210 (46)	120 (38)	112 (40)
2, n (%)	98 (22)	90 (30)	89 (33)	122 (27)	101 (32)	81 (29)
3, n (%)	83 (18)	58 (19)	52 (19)	81 (18)	59 (19)	52 (19)
4, n (%)	32 (7)	22 (7)	31 (11)	24 (5)	21 (7)	23 (8)
5, n (%)	27 (6)	30 (10)	17 (6)	19 (4)	13 (4)	13 (5)
EQ VAS
n	457	298	268	457	306	275
Mean	65.3	64.3	65.5	64.3	67.8	67.0
SD	20.0	21.2	21.8	19.9	20.3	21.5
Range	0–100	5–100	10–100	0–100	0–100	0–100

EuroQol-5 Dimensions utility values and quality-adjusted life-years

Table 26 shows the EuroQol-5 Dimensions (EQ-5D) utilities (measured using EQ-5D-5L, valued using crosswalk to EQ-5D-3L) among available cases at each time point, and the mean QALYs over 9 months among available cases. These results suggest that a difference in utility values at 3 months is largely responsible for an overall difference in raw mean QALYs of 0.017 in favour of the TARS intervention group.

TABLE 26 - Preference-based health-related quality of life and QALYs, by group

Measure: time point	SAU group		Intervention group
	Participants (n)	Mean (SD) [range]	Participants (n)	Mean (SD) [range]
EQ-5D: baseline	451	0.76 (0.25) [–0.25 to 1.00]	452	0.77 (0.24) [–0.25 to 1.00]
EQ-5D: month 3	297	0.66 (0.31) [–0.25 to 1.00]	306	0.72 (0.25) [–0.25 to 1.00]
EQ-5D: month 9	267	0.67 (0.29) [–0.25 to 1.00]	279	0.68 (0.27) [–0.40 to 1.00]
EQ-5D QALYs	263	0.51 (0.20) [–0.12 to 0.75]	276	0.53 (0.17) [–0.23 to 0.75]

Data completeness

There were a considerable number of participants for whom it was not possible to estimate costs and/or QALYs, resulting in a total of 470 out of 915 (51.4%) being included in the base-case analysis (complete-case analysis). See Appendix 30, Table 50, for further details.

Restricting to complete cases had the effect of decreasing the costs and QALYs in the control group but increasing the costs and QALYs in the intervention group. See Appendix 30, Table 51, for further details.

The assumption that data were missing completely at random was examined using Little’s test84 (with the Stata command mcartest85). Specifically, we applied Little’s test to the log-transformed costs at 3 and 9 months and the EQ-5D utility values at 3 and 9 months. This resulted in p = 0.15, suggesting insufficient evidence to reject the null hypothesis of missing completely at random. We further examined whether missingness could be predicted from the covariates to be used in regressions (covariate-dependent missingness) and obtained p = 0.99. This suggests that the complete-case analysis including the specified covariates may be unbiased.

Cost-effectiveness analysis

The primary cost-effectiveness analysis results are shown in Table 27. The TARS intervention is estimated to lead to a net cost increase of £173.50 (95% CI £354 saving to £514 increase) and to a QALY loss of 0.006 (95% CI 0.033 loss to 0.021 gain). The TARS intervention is therefore estimated to be dominated by (i.e. be less effective and more costly than) the control. The INMB for the TARS intervention is −£362 (95% CI −£1288 to £694), confirming that the TARS intervention is not expected to be cost-effective at a threshold of £30,000 per QALY, but that there is uncertainty around this, because the 95% CI for the INMB crosses zero. The cost per additional quit (sustained abstinence from 3 months to 9 months) was £15,500.

TABLE 27 - Within-trial cost-effectiveness analysis results

CI, confidence interval.

95% CIs for costs, outcomes and net monetary benefit are bias-corrected bootstrap CIs; 95% CIs for ICERs are produced by the bootstrap acceptability method.

There is no value of willingness to pay for 1 additional QALY at which we are confident which of the interventions has superior value.

Economic end points	Participants (n)	SAU group		Intervention group		Difference (Intervention − SAU)
		Mean	95% CIa	Mean	95% CIa	Mean	95% CIa
Costs (£)	470	1385	1077 to 1795	1559	1277 to 1874	+ 173.50	−354 to 514
QALYs	470	0.508	0.486 to 0.531	0.502	0.476 to 0.525	−0.006	−0.033 to 0.021
ICER (cost per additional QALY)						Dominated	b
Net monetary benefit (1 QALY valued at £30,000) (£000s)		13.87	13.11 to 14.72	13.51	12.68 to 14.33	−0.362	−1.288 to 0.694

The uncertainty around cost-effectiveness is demonstrated in the cost-effectiveness acceptability curve in Figure 8.

FIGURE 8.

Cost-effectiveness acceptability curve according to willingness to pay for 1 additional QALY.

Sensitivity analyses

Unit costs

When the lowest TARS intervention delivery cost from a sensitivity analysis (£203.99) was used, incremental costs reduced to £126.13. The TARS intervention was still dominated: as expected, incremental QALYs were negative, but the INMB improved from −£362 to −£315. When the highest cost from a sensitivity analysis (£291.96) was used, the incremental cost of the TARS intervention increased to £243.25 and the INMB worsened to −£432.

One-way sensitivity analyses were conducted for the unit costs of all NHS and PSS resources and are presented in Figure 9. Unit costs were increased or decreased by 20%. No single unit cost was sufficiently important for a variation of 20% to change the sign of the INMB (i.e. to make the TARS intervention cost-effective).

FIGURE 9.

Tornado diagram for one-way sensitivity analyses of NHS/PSS resource unit costs.

Incremental costs were most sensitive to the inpatient length of stay. In the base-case analysis, the cost per inpatient day was £503.96, and the mean resource use was 0.17 days in the TARS intervention group, versus 0.78 days in the control group (even after the exclusion of an outlier). This is partially caused by hospital stays of > 1 week in the control arm (four in the control group vs. zero in the TARS intervention group in the complete-case analysis), although there would still be a mean of 0.36 inpatient days in the control group if length of stay was Winsorised to 7 days.

Regression model specification

We explored the impact of the specification of the regression models on costs and QALYs by using alternative families and link functions in the GLM, and by excluding baseline costs and QALYs from the predictors. For costs and QALYs, we explored combinations of families (gamma, Gaussian and inverse-Gaussian), link functions (log, identity, inverse, inverse-squared) and transformation of baseline costs/QALYs (log, identity, inverse). We calculated the Akaike information criterion (AIC) for each model.

For costs, the lowest AIC (7681.0) was obtained using a gamma family with the log-link and log-transformed baseline costs (i.e. the base case). The second lowest AIC (7682.5) was obtained using a gamma family with identity link and untransformed baseline costs. The AIC for a linear model (Gaussian family with identity link and untransformed baseline costs) was 8758.8. Further details are given in Appendix 30, Regression model specifications for costs and quality-adjusted life-years.

For QALYs, the lowest AIC (−694.4) was obtained using a linear model. The second lowest AIC (−625.5) was obtained using a gamma family with identity link and log-transformed baseline EQ-5D-5L scores. The base case (gamma family with log-link and untransformed baseline EQ-5D-5L scores) had an AIC of −599.5. Further details are given in Appendix 30, Regression model specifications for costs and quality-adjusted life-years.

The choice of GLM family, link function and transformation of baseline costs/quality of life had a moderate effect on cost-effectiveness (see Appendix 30, Table 56). In all analyses, costs were estimated to be higher in the TARS intervention group, but incremental costs varied from £51 to £227 (base case £174). In the base-case analysis, incremental QALYs were small and negative (−0.006), whereas, in sensitivity analyses, incremental QALYs were small and positive (0.002 to 0.007).

The inclusion or exclusion of baseline EQ-5D-5L scores had a significant impact on the estimation of incremental QALYs: when these were included (base case), we estimated small and negative incremental QALYs (−0.006), but, when these were excluded, we estimated modest and positive incremental QALYs (0.026). This corresponds to losing 11% fewer QALYs from the maximum possible 0.75 QALYs than in the control arm. It is generally considered important to include baseline health state values in regression models for QALYs,86 which is why they are included in the base-case analysis. Fitting the models with and without baseline health state values helps us to see how much of the apparent difference in QALYs can be explained by differences in baseline health between the groups.

The mean baseline EQ-5D-5L scores were 0.744 in the control group and 0.780 in the TARS intervention group when restricting to complete cases. This difference was not primarily caused by a failure of randomisation: the baseline EQ-5D-5L scores were 0.756 in the control group and 0.766 in the TARS intervention group when all 904 participants providing baseline EQ-5D-5L data are included. The restriction to complete cases led to a disproportionate elimination of participants in the TARS intervention group with poor baseline EQ-5D-5L data. Overall, 13% (115/904) of participants had a baseline EQ-5D-5L score of < 0.5, and, when restricting to complete cases, again 13% (60/470) of participants had a baseline EQ-5D-5L score of < 0.5. Among participants in the TARS intervention group, 11% (50/452) had a baseline EQ-5D-5L score of < 0.5, but this reduced to 8% (20/239) when restricting to complete cases.

Handling of missing data

As noted previously, there is some reason to suspect that missingness is related to baseline quality of life and treatment allocation. Combined with the moderately high rate of missing data (only 51.4% of participants were included in the complete-case analysis), this gives some concern that missing data may bias the analysis results.

We conducted a multiple imputation analysis using multiple imputation by chained equations. The conditional models used predictive mean matching and included all analysis variables plus relationship status, employment status, EuroQol Visual Analogue Scale scores and SF-12v2 scores. We generated 50 imputation sets.

The results of the analysis (Table 28) show that there is still no statistically significant difference in either the costs or the QALYs between the control and intervention groups, but that the expected incremental costs are now negative (small saving of £38) and incremental QALYs are now positive (small gain of 0.009). Therefore, based on expected values, the TARS intervention is now dominant, rather than dominated by the control, and the INMB is now positive (although the 95% CI for the INMB still crosses zero).

TABLE 28 - Cost-effectiveness analysis using multiple imputation

CI, confidence/credible interval.

95% CIs for costs, outcomes and net monetary benefit are based on Rubin’s rule; 95% CIs for ICERs are produced by the acceptability method.

There is no value of willingness to pay for 1 additional QALY at which we are confident which of the interventions has superior value.

Economic end points	Participants (n)	SAU group		Intervention group		Difference (Intervention − SAU)
		Mean	95% CIa	Mean	95% CIa	Mean	95% CIa
Costs (£)	914	1538	1235 to 1842	1500	1269 to 1731	−38	−413 to 336
QALYs	914	0.499	0.481 to 0.517	0.508	0.491 to 0.526	0.009	−0.014 to 0.033
ICER (cost per additional QALY)						Dominant	b
Net monetary benefit (1 QALY valued at £30,000) (£000s)						0.317	−0.535 to 1.169

The cost-effectiveness acceptability curve for the multiple imputation analysis (Figure 10) shows that there is still no value of willingness to pay for 1 QALY at which we are confident (at the 0.05 significance level) that the TARS intervention or control is economically superior, but suggests a greater probability that the TARS intervention is cost-effective than was suggested by the complete-case analysis.

FIGURE 10.

Cost-effectiveness acceptability curve with multiple imputation.

Subgroup analyses

The results of subgroup analyses are shown in Appendix 30, Table 56. Although many of these comparisons are underpowered, we did find statistical significance for incremental costs (null hypothesis being zero) in the two younger age subgroups. Those with high HSI scores are expected to result in cost savings, but also QALY losses, for an ICER of £20,200 per QALY. As this is in the south-west quadrant, the ICER must be greater than the cost-effectiveness threshold for the TARS intervention to be cost-effective. Generally, there are no strong signals for participant groups in which the TARS intervention would be expected to be more or less cost-effective.

Review of model-based approaches to estimating cost-effectiveness of smoking cessation interventions

We conducted a review of the published cost-effectiveness literature for smoking cessation to support the development of a modelling framework to evaluate the TARS intervention.

Full details of the review are provided in Appendix 31, with the methods and results summarised in this section.

Model-based cost-effectiveness analysis of the TARS intervention

The TARS has not demonstrated superiority of the TARS intervention versus control in the primary outcome to the prespecified significance level of the study (0.05), as explained previously. Although it has been argued that demonstrating statistical significance in a clinical effectiveness outcome should not be a prerequisite for deciding to introduce a new treatment (e.g. Claxton91), and it is not uncommon for economic evaluations to find in favour of a new treatment even if neither clinical outcomes nor costs have been shown to be different,92 it is nevertheless important to ensure that the fundamental message of the trial – a negative result – is not obscured by an economic evaluation, especially given all the possible sensitivity analyses that can be included in a model-based analysis.

The model-based analysis presented in this subsection explores the potential cost-effectiveness of the TARS intervention, and incorporates the uncertainty surrounding treatment effectiveness using standard decision-analytic modelling methods. The decision to conduct a model-based economic evaluation was prespecified and is not an indication that the effectiveness of the TARS intervention is considered to be proven.

Methods

Statement of the problem/objective of the research

The benefits of smoking cessation are not expected to be realised immediately and cannot be fully captured by RCTs, whose follow-up periods are generally long enough to establish whether or not a quit attempt has been successful, but not long enough to see benefits in terms of reducing the incidence of smoking-related disease and smoking-related mortality. A modelling framework is needed to translate the effectiveness data from the TARS into lifetime estimates of QALYs and costs that fall on the NHS and PSS.

Perspective

The cost-effectiveness analysis adopts the perspective of the NHS and PSS, in line with the NICE reference case,93 which is typically followed in health economic evaluations in the UK.

Interventions

The interventions compared in the cost-effectiveness analysis are the TARS intervention and brief advice (i.e. a standard care control).

Model type and rationale for structure

A proportional multistate life table (PMSLT) cohort simulation model has been developed,94 using a cycle length of 3 months. This is similar to a Markov cohort simulation, but it makes a number of assumptions to allow the modelling of multiple (co)morbidities without leading to a combinatorial explosion of health states. This model approach has been used in the World Health Organization Global Burden of Disease studies and is well suited to applications where a risk factor for multiple diseases is modified. 95 Higashi and Barendregt96 have previously used the PMSLT methodology to evaluate the cost-effectiveness of personal smoking cessation support.

A key assumption of the PMSLT approach is that, after the population has been separated according to the risk factor(s) of interest, the diseases being modelled are independent of each other (apart from competing mortality). Specifically:

• Having one morbidity does not put an individual at greater or lesser risk of incidence of another morbidity.

• The mortality rates of diseases are additive (e.g. if disease A alone contributes an additional 100 deaths per 100,000 person-years and disease B alone contributes an additional 200 deaths per 100,000 person-years, then the two diseases together will contribute an additional 300 deaths per 100,000 person-years).

The model follows two identical cohorts, one of which receives the TARS intervention and the other receives a control (brief advice). The relative effectiveness of the TARS intervention is introduced in the first cycle of the model, and thereafter the cohorts are modelled identically.

Our modelling approach incorporates smoking as a risk factor for morbidity and mortality and separates the population into three groups according to whether they are currently smoking, or, if they have quit, whether they quit after receiving the intervention (see Short-term quitting). The model allows for individuals to relapse after quitting, and to spontaneously quit.

The model predicts the risk of incidence and mortality of four smoking-related diseases: COPD, coronary heart disease, lung cancer and stroke. These were selected by consideration of the role these diseases play in the burden of smoking (health and economic),97,98 and the systematic review of existing economic models of smoking cessation (see Review of model-based approaches to estimating cost-effectiveness of smoking cessation interventions).

The model also predicts excess mortality due to smoking that is not already realised through the four smoking-related diseases.

Figure 11 gives an overview schematic of the model.

FIGURE 11.

Overview model schematic.

The model uses the life-table method (an alternative to the half-cycle correction) to account for transitions occurring during cycles. 99

Base-case analysis cohort characteristics

In the base-case analysis, we consider men who are aged 50 years and are current smokers. In heterogeneity analyses, we consider men and women (current smokers) aged 25–85 years. The mean age in the TARS was 49.8 years, and, although women made up the majority of participants in the TARS (55%), men are more likely to be current smokers in the general population.

Modelling smoking status over time

In the model, the population is divided into current smokers and former smokers. At the start of the model, the whole cohort is a current smoker. Over time, current smokers may spontaneously quit, and former smokers may relapse. We separate former smokers into those who quit in the first cycle (at the time the TARS intervention is delivered) and remain abstinent, and those who quit at a subsequent cycle. The states are labelled ‘current smoker’, ‘abstinent smoker’ and ‘other former smoker’. As it is only possible to enter the ‘abstinent smoker’ state in the first cycle, the time since quitting is known (subject to rounding to the cycle length). In the ‘other former smoker’ state, it is expected that there will be a distribution of times since quitting, but we track the mean time since quitting. 100

Short-term quitting

The TARS intervention is expected to be delivered in eight sessions across 8 weeks, with the intention that any quit attempts that are made as a result of the intervention are made by 3 months after initiation. We assume that the effect of the TARS intervention is restricted to the first cycle (which ends 3 months after the model starts).

In the first cycle, there is a baseline probability of smoking cessation of 3.9% (for a cohort aged 50 years; for a cohort aged 30 years the probability is 5.6% and for a cohort aged 70 years the probability is 3.7%). This probability is derived from the rate of spontaneous quitting described in Long-term relapse and spontaneous quits. Those quitting smoking in the first cycle transition to the ‘abstinent smoker’ health state. In the control arm, the baseline rate is applied, whereas, in the TARS intervention arm, the relative effectiveness of the TARS intervention is introduced as a relative risk. A relative risk was selected because, in the absence of differential mortality by smoking status (which is not expected to occur immediately), the relative risk of being in the ‘abstinent smoker’ health state is preserved in the presence of relapse (it applies equally in both arms) and subsequent quits (they enter the ‘other former smoker’ state rather than the ‘abstinent smoker’ state).

Those who remain in the ‘abstinent smoker’ state to the end of the third cycle (to 9 months from the start of the model) will have achieved the ‘primary outcome’ of TARS (prolonged abstinence from 3 to 9 months post randomisation). At the end of the third cycle, the relative risk of being in the ‘abstinent smoker’ state will be the same as was introduced in the first cycle.

The relative risk itself was calculated to be 1.34 on the basis of the absolute risk difference from TARS (1.12%). We chose to use the absolute risk difference from TARS as opposed to the relative risk (2.27) because it is the absolute risk difference that will drive incremental costs and QALYs and because the uncertainty distribution for the relative risk is asymmetric and extends to implausibly high effectiveness values.

Because TARS has not demonstrated statistical significance in the primary outcome, the model does not assume that the risk difference of 1.12% is known precisely. Instead, as with virtually all model inputs, it is associated with uncertainty: in this case, the uncertainty distribution is informed by TARS. In sensitivity analyses, the risk difference is varied, and, in some instances, the risk difference is zero or even negative. The input distribution for the risk difference is normally distributed, with a mean of 0.0112 (SD 0.0079), such that 95% of the distribution is between −0.0040 and 0.0267 (the 95% CI for the risk difference).

Long-term relapse and spontaneous quits

Hawkins et al. 101 used data from multiple waves of the British Household Panel Survey to identify longitudinal data on smokers who had attempted to quit smoking, and to assess the risk of relapse. We used Bayesian methods102 to fit a Gompertz model to the risk of relapse according to time since quitting, accounting for left truncation, interval censoring and right censoring present in the data reported. The Gompertz model suggests that just over 60% of quit attempts ultimately end with relapse. Full details are provided in Appendix 32, Smoking behaviour.

The annual probability of spontaneously quitting smoking was estimated from the Understanding Society study, waves 6–9. 103 Flexible regression splines were used to account for the relationship between age and spontaneous quitting. As it was only possible from the data to identify individuals who were non-smokers 1 year after previously describing themselves as current smokers, it was necessary to account for those smokers who attempted to quit but relapsed before being surveyed the following year. We estimate relatively high rates of quit attempts (15–25 quit attempts per 100 person-years), compared with the existing literature104–108 (annual quit probabilities of around 4%). Full details are provided in Appendix 32, Smoking behaviour.

Modelling smoking-related morbidities

We modelled each of the four smoking-related morbidities in a similar way. For each, we had a health state for being alive and unaffected by the disease, a health state for being alive and affected by the disease, and a health state for having died as a result of the disease. Each of these health states was further composed of abstinent smokers, current smokers and other former smokers (Figure 12). Owing to the structure of the PMSLT model, each of these models is not concerned with tracking how many have died from other causes.

FIGURE 12.

Example of morbidity model (for COPD).

For COPD and lung cancer, we assumed that it was not possible to go directly from being alive and unaffected by the disease to having died as a result of the disease, and that individuals would need to first visit the alive and affected health state prior to dying from the disease. Therefore, there were three parameters of interest (for each combination of smoking status, age and gender): the prevalence at the start of the model (when the whole cohort is current smokers), the incidence of the disease and the case fatality rate for the disease.

For coronary heart disease and stroke, we assumed that it would be possible to go directly from being alive and unaffected by the disease (or at least to have not had the disease diagnosed) to having died as a result of the disease. We therefore had an additional parameter for these diseases, which was the probability that disease incidence would be fatal.

Chronic obstructive pulmonary disease

Effect of smoking status

Rodriguez et al. 109 estimated that the OR for COPD incidence would be 6.15 for current smokers versus never smokers, and 3.45 for former smokers versus never smokers. They also estimated an OR of 0.61 for former smokers versus current smokers. From the CIs presented, we calculated that, on a logarithmic scale, the ORs for current and former smokers versus never smokers would have a correlation of 0.4. Rodriguez et al. 109 did not separate former smokers by time since quitting, so we incorporated the effect of time since quitting using estimates from Hoogenveen et al. ,110 which led to an estimate that 78% of the excess risk due to smoking would be eliminated 10 years after quitting.

Prevalence at the start of the model

We estimated the prevalence of COPD among smokers from the 2018 Health Survey for England. 111 Further details are provided in Appendix 32, Prevalence of morbidities at start of model.

Incidence

Table S5 of McLean et al. 112 reports data from the Clinical Practice Research Datalink 2011. The number of events and the exposure (person-years) were inferred from the rates and CIs. A restricted cubic spline regression with interaction between age and gender was used to predict population-level incidence. Estimates for the OR for developing COPD within 1 year according to smoking status from Rodriguez et al. 109 were used to adjust these to incidence rates for current smokers.

Case fatality rate

We determined the case fatality rate by using data presented by Shavelle et al. 113 For each combination of smoking status, gender and age group, we calculated a weighted average of the excess mortality over the distribution of different COPD stages.

Lung cancer

Effect of smoking status

Remen et al. 114 estimated the association between smoking (including time since cessation for former smokers) and lung cancer incidence. Compared with never smokers, the OR was 17.6 for current smokers and 4.47 for former smokers. We used the estimates from model 7 of table 4 in Remen et al. 114 to incorporate the effect of cessation, including the log-transformed time since cessation, to adjust the OR for lung cancer incidence. As incidence is generally low, we assumed that the ratio of incidence rates would be well approximated by an OR.

Prevalence at the start of the model

The prevalence at the start of the model (when the whole cohort is made up of current smokers) varies from < 1% for those aged < 60 years to around 4% for those aged ≥ 80 years. For further details, see Appendix 32, Prevalence of morbidities at start of model.

Incidence

Population lung cancer incidence estimates were taken from the Office for National Statistics’ cancer registration statistics115 and were then adjusted according to smoking status using the estimates from Remen et al. 114

Case fatality rate

A Weibull survival model was fitted to UK lung cancer survival estimates116 (5-year survival by gender and age group, and 1-, 5- and 10-year survival by gender across age groups). The shape parameter for this model was 0.44, which indicates a decreasing hazard of mortality over time. To estimate a suitable average hazard of mortality (case fatality rate), we calculated the discounted life expectancy for each combination of age group and gender and used this to identify the exponential survival model with the same discounted life expectancy. As it is not possible within the model to know at what age an individual developed lung cancer (and as survival is generally poor), it was assumed that the average case fatality rate could be determined from the current age in the model, despite the survival estimates coming from the age at diagnosis.

Coronary heart disease

Effect of smoking status

Shields and Wilkins117 estimated the effect of smoking on heart disease, including estimating the effect of time since quitting among former smokers. The relative risk for current smokers versus never smokers was 1.6 for men and 1.7 for women.

Prevalence and incidence

Prevalence, incidence and mortality data for ischaemic heart disease were sourced from the 2019 Global Burden of Disease study data. 118

Case fatality rate

To estimate the probability of incident disease being fatal, we relied on incidence and case fatality data from Smolina et al. ,119 who studied hospital admissions and deaths due to acute myocardial infarction. They identify which of these are ‘first’ acute myocardial infarctions (no record of acute myocardial infarction in prior 12 years), but this does not rule out that these individuals would have already been diagnosed with angina. We assumed that 50% of these would be in individuals with no prior diagnosis of angina and estimated that 14–48% would be fatal (depending on age). 119 We then estimated the case fatality rate among the prevalent population by attributing the remaining mortality not explained by fatal incident myocardial infarction to mortality in the prevalent population, as shown in Equation 2, where β is the probability that an incident case is fatal.

⁽²⁾ Fatality = Mortality − Incidence β ( 1 − Prevalence ) Prevalence

Stroke

Effect of smoking status

Myint et al. 120 estimated the relative risk for stroke according to smoking status. For current smokers versus non-current smokers (i.e. including never smokers and former smokers), the relative risk was 2.25 for those aged < 65 years and 1.35 for those aged ≥ 65 years. They also found that previous smoking versus never smoking was not associated with stroke risk. We instead incorporated the delayed effects of smoking cessation using the model estimated by Hoogenveen et al. ,110 which predicts that 93% of the excess risk due to smoking is eliminated by 10 years after cessation.

Prevalence and incidence

Prevalence, incidence and mortality data for stroke were sourced from the 2019 Global Burden of Disease study data. 118

Case fatality rate

For stroke, we incorporated both a probability that an incident stroke would be fatal and the excess rate of death among prevalent stroke cases. The probability that an incident stroke would be fatal was estimated according to age from a study by Seminog et al. ,121 and the fatality rate among prevalent cases was estimated using Equation 2.

Smoking-related excess mortality

Smoking is a risk factor for many conditions beyond COPD, lung cancer, coronary heart disease and stroke; therefore, our model incorporates smoking-related excess mortality not already captured by the specific modelling of morbidities. The model was structured so that, in the control arm, the mortality due to COPD, lung cancer, coronary heart disease and stroke was subtracted from excess mortality, to avoid double-counting. In the TARS intervention arm, the excess mortality was taken from the control arm, meaning that overall mortality would be lowered if mortality due to the modelled morbidities was decreased. All calculations were stratified by smoking status (current smoker, abstinent smoker, other former smoker).

As part of our review of existing models (see Appendix 31), we considered a number of alternative data sources for the excess mortality due to smoking, and, in particular, how smoking cessation can reduce the rate of excess mortality. We concluded that the study by Jha et al. 122 was the most appropriate study, as it was based on comparatively recent data from a (US) nationally representative longitudinal survey, and it distinguishes between short-term and long-term quitters by using the age at quitting as a predictor for excess mortality. To improve the precision of the estimate for the risks of current smokers versus those of never smokers, we also incorporated data from Prescott et al. 123 Full details of the estimation are given in Appendix 32, Smoking-related excess mortality.

The resulting excess mortality used in the model is shown in Figure 13, demonstrating that smokers quitting in their 30s and 40s could expect to avoid most excess mortality, but that there would be less benefit obtained by smokers quitting at older ages (although at any age it was still expected to be beneficial to quit smoking). Specifically, by age 61 years, it is estimated that half of the excess mortality (in terms of the hazard ratio vs. never smokers) is ‘baked in’ and cannot be reversed by quitting.

FIGURE 13.

Excess mortality due to smoking.

We have described how we estimated the relative risks of mortality for individuals according to their smoking status, but further steps were necessary to estimate the absolute mortality rates. Mortality rates are published for the population as a whole (combining current smokers, former smokers and never smokers); it is erroneous to assume that these represent the rates for never smokers. To estimate the mortality rate among never smokers, we used the proportion of the population with each smoking status (current, former and never),124 and assumed that former smokers quit at the mid-point between age 16 years and their current age (e.g. we assumed that former smokers aged 56 years would have quit at age 36 years). We then calculated the appropriate hazard ratios using the sigmoid function described previously and used these with gender-specific period life tables125 to infer the mortality rate among never smokers.

Health state utility values

To estimate the effects of smoking status and morbidities on preference-based health-related quality of life, we used estimates from Vogl et al. ,126 who used data from the 2006 round of the Health Survey for England to estimate the effects of smoking status, age, the number of limiting conditions and a number of other covariates on EQ-5D-3L utility values. Ordinary least squares regression was conducted and reported in supplementary materials. From this, we assigned utility values as shown in Equation 3.

⁽³⁾ U = 0.9609 − 0.0014 × Age − d − m ,

⁽⁴⁾ d = { 0.0169 , Former smoker , 0.0327 , Current smoker

⁽⁵⁾ m = { 0 , No morbidities 0.0938 , One morbidity 0.1811 , Two morbidities 0.2859 , Three morbidities 0.3354 , Four morbidities

Costs

The key costs included in the model are the cost of delivering the TARS intervention (£239.18; see Estimating the cost of delivering the TARS intervention) and the annual costs associated with prevalent morbidities.

For lung cancer, coronary heart disease and stroke, we adopted costs from Aveyard et al. 127 and inflated them to 2018/19 prices. The resulting cost estimates were £7248 for lung cancer, £1291 for coronary heart disease and £6093 for stroke.

For COPD, we took estimates from McLean et al. 112 and inflated them to 2018/19 prices. This resulted in estimated maintenance costs (excluding exacerbations) of £799 for mild COPD, £905 for moderate COPD, £1308 for severe COPD and £1811 for very severe COPD. The estimated cost of an exacerbation not leading to hospitalisation was £133, whereas the cost of an exacerbation leading to hospitalisation was £4186. Averaging over the distribution of COPD severity and accounting for the rates of exacerbations, we calculated an annual cost of £1785 for COPD.

Discounting

Future QALYs and costs were discounted at 3.5% per year, in line with the NICE reference case. 93 Life-years are presented without discounting.

Analysis and presentation of results

The lifetime costs and QALYs for those receiving the TARS intervention and those receiving brief advice (control) were estimated from the model up to a maximum age of 100 years. From these, we calculated the ICER, as shown in Equation 6, and the INMB, as shown in Equation 7. The INMB requires the specification of λ, the willingness to pay for 1 QALY. Throughout this report, we use a value of £30,000 per QALY for λ, such that if the INMB is positive, this indicates that the TARS intervention is cost-effective at a threshold of £30,000 per QALY.

⁽⁶⁾ ICER = Cost TARS − Cost Control QALY TARS − QALY Control ,

⁽⁷⁾ INMB = ( QALY TARS − QALY Control ) × λ − ( Cos t TARS − Cos t Control ) .

We conducted a base-case deterministic analysis in which all parameters took their ‘base-case’ values (typically maximum likelihood estimates), and in which the cohort comprised men aged 50 years. We also conducted a heterogeneity analysis, in which the age and gender of the cohort were varied to observe the relationship between age and gender and cost-effectiveness. We conducted one-way sensitivity analyses on all parameters, generally varying them between the limits of their 95% uncertainty intervals. We conducted detailed sensitivity analyses on parameters with a significant effect on cost-effectiveness. We conducted two probabilistic sensitivity analyses: in the first, the cohort was fixed as men aged 50 years (as in the base-case deterministic analysis); in the second, the joint distribution of age and gender was taken from TARS.

Results

In the base-case analysis, it was estimated that the TARS intervention would lead to marginal improvements in life expectancy and QALYs, at an increased cost of £236 (Table 29), leading to an ICER of £37,100 per QALY. The main contribution to increased costs was the cost of delivering the TARS intervention (£239).

TABLE 29 - Base-case cost-effectiveness results from the decision model

Notes

‘Base case’ corresponds to a man aged 50 years, with all model parameters taking their base-case estimate (typically maximum likelihood estimate). QALYs and costs are discounted at 3.5% per year whereas life-years are not discounted. INMB calculated with a willingness to pay of £30,000 per QALY.

Trial arm	Life-years	QALYs	Cost (£)	INMB (£000s)
SAU	27.69	14.03	10,575	410.40
Intervention	27.71	14.04	10,811	410.36
Difference	0.017	0.006	236	−0.0453

It is important to note that, in the base-case analysis, all input values are fixed at their best estimates. This means that it is assumed that the TARS intervention leads to a 1.12% absolute increase in quit probability. The results in Table 29 are conditional on the true effectiveness of the TARS intervention being 1.12% (and all the true values of other inputs being equal to their best estimates), even though the TARS has not established the effectiveness of the TARS intervention with sufficient precision to even conclude that it is positive.

To demonstrate the effect of smoking cessation in the model, we ran the model with a relative risk of 20 for sustained abstinence to 9 months. As shown in Figure 14, this means a very significant difference in the prevalence of smoking early in the model, but the prevalence tapers quite quickly as a result of the rate of spontaneous quitting and the risk of relapse, which is initially high. This translates into an improvement in life expectancy of 0.95 years (Figure 15), and affects the prevalence of the four modelled morbidities (Figure 16).

FIGURE 14.

Prevalence of smoking when relative risk for sustained abstinence at 9 months is 20.

FIGURE 15.

Overall survival when relative risk for sustained abstinence at 9 months is 20.

FIGURE 16.

Prevalence of modelled morbidities when relative risk of sustained abstinence at 9 months is 20. (a) Coronary heart disease; (b) COPD; (c) lung cancer; and (d) stroke.

Heterogeneity analyses

As shown in Figure 17, age has a significant effect on the cost-effectiveness of the TARS intervention, whereas gender has a smaller effect. The TARS intervention is most cost-effective for those aged 45–60 years, and least cost-effective for those younger or older than this range. This result is expected because younger smokers have more subsequent opportunities to quit (or relapse if they initially quit) before the morbidity and mortality risks of smoking increase sharply. For older smokers, there is less potential to benefit from quitting because less of the increased risk due to smoking can be reversed.

FIGURE 17.

Analysis of the effects of age and gender on the cost-effectiveness of the TARS intervention.

Sensitivity analyses

One-way sensitivity analyses were conducted for all parameters (generally varying between the limits of the 95% CIs). When a parameter was correlated with other parameters in the model, we recalculated their values in accordance with the parameter that was directly varied using the conditional expectation.

The results of the one-way sensitivity analyses are presented in a tornado diagram in Figure 18. The 20 parameters with the biggest impact on cost-effectiveness are shown. The risk difference for sustained abstinence (parameter name TARS_RD) has the biggest impact on cost-effectiveness, followed by the age at which people receive the TARS intervention, the discount rate for QALYs and the cost of the TARS intervention. Two parameters relating to how quickly the risk of coronary heart disease declines following smoking cessation were important (chd_rr_m_under5 and chd_rr_m_10_14). Two of the parameters for excess mortality due to smoking (xsm_beta_0 and xsm_beta_1) are also important for cost-effectiveness, but these parameters determine the quitting age at which 50% of excess mortality remains and the slope of the excess mortality curve at this point, rather than the magnitude of excess mortality (xsm_beta_2), which has less impact on cost-effectiveness over the range in which it is varied.

FIGURE 18.

Tornado diagram of one-way sensitivity analyses.

The relationship between the risk difference for prolonged abstinence and cost-effectiveness was explored further. We found an almost linear relationship between the risk difference and the INMB for the TARS intervention. For a man aged 50 years (the base case), the risk difference would need to be 1.35% for the TARS intervention to be marginally cost-effective at a cost-effectiveness threshold of £30,000 per QALY.

We estimated the slope of INMB to be £189 per 1% risk difference in the base-case analysis; however, different slopes were calculated for men and women of different ages. For women aged 30 years, the TARS intervention would need to result in a risk difference of at least 4.83% to be cost-effective at a threshold of £30,000 per QALY.

If the relative risk from TARS (2.267) is used instead of the risk difference, the model predicts that the TARS intervention will be significantly more cost-effective (see Appendix 32, Table 63), although it would not be considered cost-effective at a threshold of £20,000 per QALY for women aged 30 or 70 years, or men aged 70 years.

The effect of the rate of spontaneous quitting on cost-effectiveness was also examined, as the TARS intervention was investigated in a population of smokers not willing to quit at the point of recruitment. In this scenario, we removed the age-related components of the spontaneous quit curve and set the rate so that the proportion abstinent in the control arm at 9 months would be 0.9% (as in TARS).

In this scenario, the life expectancy and QALYs are reduced in both groups, compared with the base-case analysis, and costs are slightly increased (see Appendix 32, Table 64). The difference in QALYs between the control and TARS intervention groups expanded from 0.006 to 0.010, which led to the ICER reducing from £37,100 to £23,100 per QALY.

Probabilistic sensitivity analysis

The combined effect of uncertainty in all parameters was assessed through a probabilistic sensitivity analysis (Figure 19). The population was kept as men aged 50 years (as in the base-case analysis). The probability of the TARS intervention being cost-effective at £20,000 per QALY was estimated to be 0.158, increasing to 0.388 at a cost-effectiveness threshold of £30,000 per QALY.

FIGURE 19.

Probabilistic sensitivity analysis for men aged 50 years.

To assess the probability of cost-effectiveness in a representative population, we conducted a probabilistic sensitivity analysis in which the ages and genders were drawn from the 915 TARS participants (Figure 20). As expected, this resulted in a worsening of the cost-effectiveness of the TARS intervention.

FIGURE 20.

Probabilistic sensitivity analysis with age and gender varied according to the TARS population.

Chapter summary

Summary of findings

Of the sample recruited, 55% identified as female, the average age was 50 years and high levels of unemployment were reported. Sixty per cent of participants lived in one of the four most deprived UK deciles. The sample participants were moderately heavy smokers and fairly active, roughly comparable to the EARS pilot trial.

TARS was powered to detect a difference of 6% (i.e. 5% vs. 11%) in the proportion of smokers who achieved CO-verified 6-month floating prolonged abstinence, with a corresponding OR for this target effect size of 2.3, based on the EARS pilot trial and existing literature. Although the TARS primary analysis showed that the estimated odds of achieving the primary outcome in the intervention group were in line with the target OR (fully adjusted estimated OR of 2.30), the between-group difference was not statistically significant (95% CI 0.70 to 7.56), owing to the much lower than anticipated proportion of participants in both groups achieving CO-verified abstinence. We recorded 0.9% (n = 4) of participants in the SAU group with CO-verified 6-month floating prolonged abstinence between 3 and 9 months, compared with 2.0% (n = 9) of participants in the intervention group. When we also included participants who had not been CO-verified abstinent at 3 months but were at both 9 and 15 months, we found that the SAU and intervention arms had 2.2% (n = 10) and 3.1% (n = 14) CO-verified abstinence, respectively, which was also not significantly different.

We also considered the proportion of participants who achieved CO-verified 12-month floating prolonged abstinence. The estimated adjusted OR was 6.33, but with 0.2% (n = 1) and 1.3% (n = 6) in the SAU and intervention groups, respectively, this between-group difference also proved statistically not significant (95% CI 0.76 to 53.10; p = 0.089), which is unsurprising given the small numbers of participants involved in this analysis.

Secondary outcomes (smoking)

Although the TARS was powered to primarily detect differences in CO-verified 6-month floating prolonged abstinence, we also investigated intervention effects on a number of secondary smoking outcomes, which have been reported previously in studies focused on harm and smoking reduction involving behavioural support for smokers who do not want to quit. Given the focus of the TARS intervention on multiple behaviour changes, we also examined the intervention effects on PA.

The intervention had weak effects on both self-reported 7-day point prevalence and CO-verified abstinence at 3 months, but not at 9 or 15 months, compared with SAU. Between baseline and 3 months, the intervention group reduced the number of cigarettes smoked by 5.6, compared with the control group. There were stronger, statistically significant, intervention effects, compared with SAU, on the proportion of participants who reduced the number of cigarettes smoked per day by at least 50%, between baseline and 3 months (18.9% vs. 10.5%) and between baseline and 9 months (14.4% vs. 10%). Assuming that those lost to follow-up did not make a quit attempt, the proportion of participants who reported making a quit attempt of at least 24 hours was fairly similar in the intervention and SAU groups between both baseline and 3 months (11.8% vs. 8.1%) and 3 and 9 months (16.9% vs. 15.1%).

Moderators of intervention effects

When possible (i.e. with sufficient available data), modifiers of the intervention effect on smoking outcomes were explored. The only significant interaction (p < 0.01) involved the IMD (comparing effects for those in the lowest and those in the highest deciles), with the intervention effect on number of cigarettes smoked per day at 3 months being present for those in the most, but not in the least, deprived neighbourhoods. At 9 months, the intervention was still effective for those in the most deprived decile, but SAU became more effective in the least deprived neighbourhood. The TARS intervention, including during the EARS pilot trial, was designed to support smokers from more disadvantaged communities, perhaps who find changing smoking habits more difficult. The focus of the client-centred intervention on increasing a sense of competence, control and connectedness, while building trust and demonstrating empathy, were all valued aspects, based on interviews with participants and HTs in our process evaluation. We have not yet explored any moderator effects on our process survey items, but, if identified, this would link well to the logic model in which the intervention may be most useful for those who need most support to change without the pressure of more directive approaches from health professionals. Overall, however, the intervention effects on various outcomes were robust across different potential moderators.

Control group changes in smoking outcomes

The failure of TARS to show a statistically significant effect on the primary outcome of CO-verified 6-month (and 12-month) floating prolonged abstinence was possibly due to fewer participants than expected stopping smoking. In particular, the sample size calculation assumed that 5% of the SAU group would achieve the primary outcome. Although the sample size scenarios considered at trial design went as low as 3% for the SAU group, the observed 0.9% was much lower than anticipated (see Table 2).

Although there are no comparative data from previous studies for the primary outcome, a number of comparisons can be made for secondary outcomes for control/SAU group participants. Catley et al. 14 reported that, among control participants, 0% had 6-month CO-verified abstinence, compared with 1.8% and 4.7% with CO-verified 3- and 9-month point prevalence abstinence in TARS, respectively. Carpenter et al. 10 reported that 4% of control participants self-reported 7-day point prevalence abstinence at 6 months, compared with 2.9% (at 3 months) and 8.0% (at 9 months) in TARS. Klemperer et al. 12 reported control rates for self-reported 7-day point prevalence abstinence of 5% and 4% at 6 and 12 months, respectively.

Similarly, Carpenter et al. 10 reported that 16% of control participants had reported making a 24-hour quit attempt at 6 months, whereas Klemperer et al. 12 reported that 34% of control participants made a quit attempt lasting at least 24 hours up to the 6-month follow-up. In Catley et al. ,14 31% and 43% of control participants had made a quit attempt (in the previous 30 days) by the 3-month and 6-month follow-ups, but this was the only trial that included participants who smoked as few as one cigarette per day at the time of recruitment (compared with at least 10 daily cigarettes in the other studies and in TARS). In comparison, in TARS, 14% of control participants reported having made a quit attempt (of ≥ 24 hours) between baseline and 3 months, and 25% reported having made a quit attempt between 3 and 9 months.

Compared with the study by Glasgow et al. ,11 in which 19% of control participants achieved a ≥ 50% reduction in the number of cigarettes smoked per day between baseline and 12-month follow-up, TARS found that 10.5% of control participants did so at 3 and 9 months, respectively.

The EARS pilot trial reported that 4% of control participants had expired CO-verified abstinence at least 4 weeks after quitting; 6% made a quit attempt; 4% achieved self-reported point prevalence abstinence at 16 weeks; and 16% and 20% achieved at least a 50% reduction in the number of daily cigarettes smoked by weeks 8 and 16, respectively.

In summary, TARS reports less change in smoking outcomes in the control/SAU group than in other studies on roughly comparable measures.

Effects on physical activity

TARS is unique in that it had an additional focus on supporting changes in PA to complement behavioural support for smoking reduction. Clearly, the TARS intervention, with a multiple behaviour change focus, has not had adverse effects on smoking outcomes for smokers who are not motivated to quit. Although the EARS pilot trial,30 with 99 participants, showed no intervention effect on self-reported or accelerometer-recorded MVPA in the exploratory analysis, it appeared that a greater focus in training the HTs to support changes in PA may have led to the significant intervention effect on self-reported minutes of MVPA at 3 months; this equated to a weekly average of 79 minutes more than the control group. However, these effects were not evident at 9 months (mean difference 22 minutes), and it would appear that additional support for participants would be needed for longer-term change in PA.

Mediation of physical activity on smoking

A key question for TARS was if changes in MVPA would mediate intervention effects on smoking outcomes. As the results in Chapter 4 show, MVPA minutes did not significantly mediate intervention effects on the number of cigarettes smoked daily at 3 months or on achievement of a ≥ 50% reduction in the number of cigarettes smoked daily between baseline and 3 months or baseline and 9 months. There have been 24 randomised trials, of varying quality, on the effects of exercise interventions on smoking cessation22 for smokers wanting to quit, with only very limited certainty that exercise can be effective. TARS was not designed to test the separate effects of PA on smoking outcomes, but the analysis does not support the hypothesis that increasing PA mediated the effects on smoking; however, the very modest-sized effects overall limit our ability to test this. Subjective evidence from interviews with HTs and participants did indicate that some participants were finding various effective ways for using PA to help changes in smoking.

Intervention effects on smoking and physical activity process survey measures

The logic model informed the design of the HT-led intervention and the process survey measures we collected. It was encouraging to see that the intervention had positive effects on 10 out of 11 smoking process survey items, and on five out of seven PA process survey items, including one item that suggested that intervention participants became more positive about PA for controlling smoking. This is contrary to the mediational analysis discussed previously, in which PA did not mediate smoking outcomes, and contrary to the rating of HT sessions in which the integration of behaviours appeared to be less easy to deliver than some of the other competencies. The only smoking-related process measure that was not influenced by the intervention was the importance attached to quitting, and the dilemma for the HTs in supporting reduction and not pushing cessation was evident in qualitative work.

Process measures as mediators of intervention effects on smoking and physical activity

The increase between baseline and 3 months in smoking beliefs about the importance of smoking reduction; confidence to reduce; and use of action planning, coping planning and self-monitoring mediated changes in intervention effects on the number of cigarettes smoked daily up to 3 months. Changes in confidence to reduce and to quit, action planning, coping planning, self-monitoring and thoughts about quitting also mediated intervention effects on whether or not participants reduced their smoking by ≥ 50% up to 3 months, but, beyond 3 months, the effect of these process measures weakened, with only confidence in reducing and self-monitoring being marginally significant. Changes in PA beliefs, up to 3 months, regarding confidence to be active and self-monitoring PA mediated intervention effects on MVPA minutes at 3 months. Klemperer et al. 13 also explored possible mediators of changes in smoking outcomes and found that a 1-unit (on a scale of 0–5, from low to high) change in self-efficacy to quit from baseline to 4 weeks mediated intervention effects on whether or not a quit attempt (lasting ≥ 24 hours) took place over 6 months, but had no effect on the self-reported point prevalence abstinence rate at 6 months.

Health economics

The TARS intervention has been estimated to cost £239.18 per participant, on average. The majority of this cost is HT time, including contact time (£92.84), travel time (£53.02) and non-contact time (£71.69). The assumptions required to estimate non-contact time mean that it is plausible that the cost could vary between £200 and £300.

The review of cost-effectiveness analyses of targeted smoking cessation interventions (i.e. not using mass marketing or changes to taxation) found a number of studies evaluating pharmacological aids, counselling, electronic aids and complex interventions. Two published studies incorporated exercise. Taylor et al. 23 reported the results of the EARS trial, which was a pilot for this trial. The cost of the EARS trial intervention was estimated to be £192, with the principal reason for the cost difference being improved estimates of HT time associated with the intervention. Leung et al. 108 reported the incremental costs of an exercise–counselling intervention as NZ$428 [£228 when converted from 2012 New Zealand Organisation for Economic Co-operation and Development (OECD) purchasing power parity (PPP) to 2018 UK OECD PPP133]. Glasgow et al. 11 reported the costs of a counselling programme for smoking reduction as US$652 (excluding recruitment costs, approximately £526 when converted to 2018 UK OECD PPP); the differences in costs between the study by Glasgow et al. 134 and TARS are likely to be explained by the presence of additional components (tailored newsletters) and more intensive training and supervision (including supplementary training and live supervision).

The NHS and PSS resource use measured in TARS suggests that those receiving the TARS intervention may have incurred somewhat lower costs than those in the control arm, but after accounting for the cost of the intervention, total costs are expected to be higher for those receiving the TARS intervention by £173.50 over 9 months, starting from treatment allocation. This difference was not statistically significant.

Preference-based health-related quality of life was measured, which allowed QALYs to be estimated for trial participants. There was no statistically significant difference in QALYs between the groups.

A significant number of data for health economic analysis were missing. A sensitivity analysis using multiple imputation found that the TARS intervention resulted in net savings of £38.02 and a gain of 0.009 QALYs, but neither of these findings were statistically significant.

A decision-analytic model was constructed to estimate the lifetime costs and health outcomes based on the trial outcome of smoking cessation. The model incorporated the effects of smoking cessation on COPD, coronary heart disease, stroke and lung cancer; on excess mortality not due to those four conditions; and on quality of life.

It was estimated that the 1.1% absolute difference in the probability of prolonged abstinence to 9 months would translate into a small gain in lifetime QALYs, and small savings in lifetime costs from smoking-related diseases, but not enough to offset the cost of the intervention and lead to TARS being considered cost-effective (at a threshold of £20,000–30,000 per QALY).

The absolute difference in quit rates between the model arms and the assumptions around the rate of spontaneous quitting had a strong effect on cost-effectiveness. In a sensitivity analysis (in which spontaneous quitting was reduced to match the quit rate in the control arm of the TARS and maintained at this low level over the lifetime of the cohort), the TARS intervention was estimated to be cost-effective at a threshold of £30,000 per QALY, but not at £20,000 per QALY.

Strengths

TARS has a number of important strengths including the following: it involved a very robust primary outcome, a detailed mixed-methods process evaluation and a health economic analysis; rigorous intervention development with PPI input; manualisation of the intervention and practitioner training; good intervention engagement and fidelity (content, training, delivery and signs of enactment); and a good proportion of the sample was from socially deprived neighbourhoods. The logic model seemed to be working based on process measures (and interviews/session recordings). This is the first UK definitive trial of behavioural support for smokers not wanting to quit and adds to the research on multiple behaviour change interventions, especially involving PA and smoking. Given the prevalence of clusters of poor health behaviours, linked to low socioeconomic status, TARS is unique in its contribution to the understanding of intervention engagement and effects in this population.

Limitations

Although we identified an adjusted OR close to our pre-planned rates for the primary outcome, very low abstinence rates for the whole sample made it impossible to be confident in this effect in favour of the intervention. However, the low absolute quit rates achieved mean that, even if we had detected this effect, the TARS intervention does not represent a worthwhile addition to tobacco control interventions because it was not cost-effective.

We collected data mostly during face-to-face meetings with participants at baseline, and then relied on remote data capture at the follow-up assessments; it is unknown what effects this had on the results. To reduce participant burden in recruitment and completion of baseline data, we collected accelerometer data from only a sample of participants at 3 months, which made the assessment of intervention effects on PA less robust.

To detect intervention effects on secondary outcomes and to detect moderators and mediators only exploratory analyses were conducted. Although we calculated Cronbach’s alpha for the composite scales involving process survey items drawn from previous measures used with our randomised trials, we did not examine the factor structure of the scales.

A considerable proportion of participants did not return follow-up assessments; in this case, we applied the Russell Standard and the assumption that those with missing data were still smoking. We conducted a sensitivity analysis to check this assumption, but it is less clear how missing data influence other secondary outcomes.

The small number of participants achieving the primary outcome precluded many of the planned sensitivity analyses, with only the prespecified ‘best-case scenario’ analysis possible, which did not change the overall conclusion. Similarly, the effects of COVID-19 restrictions, which forced us to switch to saliva cotinine verification of self-reported abstinence, were considered descriptively, but the numbers tested using saliva cotinine-verified abstinence were too small to conduct any worthwhile further analysis.

In terms of the process evaluation, one limitation of note was that interviews with intervention participants took place after receipt of the intervention, but before the final follow-up; this could have influenced some of the trial outcomes. Moreover, males were slightly over-represented in the interviews as a result of the sampling methods, although it is unclear how this may have affected the analysis and reflections.

Implications for health care

It is clear from the perspective of the primary outcome (CO-verified 6- and 12- month floating prolonged abstinence) and associated health economic analysis that there was no evidence that the intervention was effective or cost-effective.

For smokers not wanting to quit, the HT intervention can change both smoking behaviours and PA for at least up to 3 months, with some evidence that the multiple behaviour change intervention can increase the proportion who report reducing their smoking by at least 50%.

TARS involved a detailed analysis of what the intervention involved, how the HTs were trained and delivered the intervention, and what effects this had on beliefs about smoking, PA and multiple behaviours for > 450 participants. Overall, there was good evidence of acceptability to participants and feasibility of scaling up the delivery of this intervention through standardised training and supervision.

The limited long-term intervention effect on CO-verified abstinence and qualitative work suggest that, for some participants, booster sessions to maintain harm reduction and smoking abstinence would have been valued. Delivering motivational support to reduce smoking, while at the same time encouraging quitting and abstinence, was highlighted as a challenge.

The data suggested that, although the sample participants were initially physically active (possibly from a greater proportion in manual occupations, and lower levels of car ownership, than the general population), the HT intervention still managed to increase short-term self-reported MVPA. However, as far as we can tell from the present preliminary analyses, intervention effects on MVPA do not appear to then modify changes in smoking behaviour, although some participant interviews did provide support that this was happening. The qualitative data provide some rich detail on how some, but clearly not all, participants used PA to support changes in smoking behaviour, both acutely and chronically.

There are a variety of UK community-based practitioners with overlapping roles with HTs, such as link workers, health coaches and those working in social prescribing initiatives, who may all be faced with providing support for multiple behaviour change, especially among disadvantaged groups. For smokers not wanting to immediately quit, it is possible to support short-term changes in PA and smoking, but we found no evidence that these changes result in sustained abstinence in smoking and longer-term PA, or that this intervention would be cost-effective. The findings should be integrated into future reports that follow The King’s Fund work on multiple health behaviour change. 135

The model-based cost-effectiveness analysis did not incorporate the effects of sustained reduction in smoking intensity or sustained increase in PA on long-term costs and health benefits. These effects would probably be in favour of the TARS intervention, as reduced smoking intensity and increased PA would lower the risks of smoking-related morbidity and mortality (reducing costs and improving quality-adjusted life expectancy). If these effects can be realised and sustained in real-life practice, then the value for money of the TARS intervention will be improved.

Future research implications

In seeking to recruit participants to the trial, we clearly advertised that the trial was not about quitting, and the trial was not overtly linked to NHS Stop Smoking Services. Further research is needed to explore the role of multiple behaviour change support (building on the present HT intervention) for smokers who want to reduce prior to quitting.

Further interventions should be tested in which approaches to aid longer-term changes in smoking and PA are added. The present intervention was client centred in that participants could elect to work on changing smoking and/or PA behaviours, in their own time and way (with manualised HT support). Although some TARS participants appreciated the opportunity to receive support for smoking reduction without the pressure to quit, this approach was insufficient to offer an overall effective and cost-effective approach to achieve sustained abstinence; alternative forms of support will need to be explored.

In populations who are resistant to quitting smoking but are open to reducing the amount they smoke and engaging in other positive health behaviours (such as increasing PA), an alternative modelling framework for health economic evaluation may be necessary that is less driven by prolonged abstinence and more driven by smoking intensity and measures of physical health. In such modelling, it will be important to consider the possibility of relapse towards prior health behaviours (i.e. returning to pre-intervention levels of smoking and PA), just as our economic evaluation considered the possibility of relapsing following smoking cessation.

Conclusions

TARS showed that the intervention was delivered with sufficient fidelity and influenced theoretically relevant constructs that led to short-term behaviour change in smoking and PA. However, we found no evidence that the intervention was effective in promoting long-term smoking cessation and it was not cost-effective in the short term or in terms of long-term health. We had assumed that the baseline abstinence rate would be much higher, so that a relative effect equivalent to the point estimate we observed would have given us 90% power. In the event, achieved power was much lower, but the trial evidence was sufficient to show that the intervention was not cost-effective. Consequently, this intervention cannot be recommended to health and social care providers as a way to promote prolonged smoking abstinence.

Contributions of authors

Adrian H Taylor (https://orcid.org/0000-0003-2701-9468) (Professor in Health Services Research) conceived the idea for the study with Tom P Thompson (https://orcid.org/0000-0002-6366-9743) (Senior Research Fellow, Addiction and Health Behaviour Research), Michael Ussher (https://orcid.org/0000-0002-0995-7955) (Professor of Behavioural Medicine), Paul Aveyard (https://orcid.org/0000-0002-1802-4217) (Professor of Behavioural Medicine), Colin Green (https://orcid.org/0000-0001-6140-1287) (Professor of Health Economics), Colin J Greaves (https://orcid.org/0000-0003-4425-2691) (Professor of Psychology Applied to Health) and Siobhan Creanor (https://orcid.org/0000-0002-7373-8263) (Professor of Medical Statistics and Clinical Trials).

Adrian H Taylor, Tom P Thompson, Michael Ussher, Paul Aveyard, Rachael L Murray (https://orcid.org/0000-0001-5477-2557) (Professor of Population Health), Colin Green, Colin J Greaves and Siobhan Creanor contributed to the final trial design and development of the protocol.

Adrian H Taylor, Tom P Thompson and Colin J Greaves developed the intervention, with Lynne Callaghan (https://orcid.org/0000-0002-0766-645X) (Senior Research Fellow, Psychological Research Methods), who supported PPI input.

Adrian H Taylor, Michael Ussher, Paul Aveyard, Rachael L Murray and Tess Harris (https://orcid.org/0000-0002-8671-1553) (Professor of Primary Care Research) were the principal investigators at participating sites.

Tom P Thompson led the process evaluation data collection, analysis and write-up with Adam Streeter (https://orcid.org/0000-0001-9921-2369) (Research Fellow in Medical Statistics) (mediation analysis), Jane Horrell (https://orcid.org/0000-0001-7284-7647) (research assistant), Lynne Callaghan (lead) and Lucy Cartwright (https://orcid.org/0000-0001-8733-1379) (research assistant) (qualitative methods), and with further input from Adrian H Taylor and Colin J Greaves.

Adam Streeter, Jade Chynoweth (https://orcid.org/0000-0002-2516-5923) (Research Assistant in Medical Statistics) and Siobhan Creanor provided the SAP, and Adam Streeter and Jade Chynoweth performed the statistical analysis.

Wendy Ingram (https://orcid.org/0000-0003-0182-9462) (Clinical Trials Manager) and Sarah Campbell (https://orcid.org/0000-0002-5974-5969) (Senior Clinical Trials Manager) were the trial managers at the PenCTU.

Colin Green developed the economic evaluation analysis plan with Tristan Snowsill (https://orcid.org/0000-0001-7406-2819) (Senior Research Fellow, Health Economics). Tristan Snowsill performed the health economic analysis with Colin Green.

Lisa Price (https://orcid.org/0000-0001-9478-3488) (Lecturer in Physical Activity and Health) was consulted on accelerometer data capture and processed the accelerometer data.

Jonny Wilks (https://orcid.org/0000-0002-0545-4146) (Clinical Trials Data Manager) was the data manager at the PenCTU.

Dan Preece (https://orcid.org/0000-0002-0559-7948) (Advanced Public Health Practitioner, Plymouth City Council) provided a public health perspective on usual community stop smoking service support.

All authors critically revised successive drafts of the manuscript and approved the final version.

This research has been conducted independently by researchers at the University of Plymouth, the University of Oxford, the University of Nottingham and St George’s, University of London, with the support of clinical research networks (CRNs) at each site and the Applied Research Collaboration South West Peninsula.

Data-sharing statement

Following publication of the primary results of the trial, access to available anonymised data may be granted depending on review of the data request and appropriate agreements being in place. Members of the Project Management Group, in particular the chief investigator in the first instance, with members of the PenCTU, medical statistics and health economics team, a representative of the sponsor and site principal investigators, will consider applications. Any disagreement will be considered by members of the TSC.

Disclaimers

This report presents independent research funded by the National Institute for Health and Care Research (NIHR). The views and opinions expressed by authors in this publication are those of the authors and do not necessarily reflect those of the NHS, the NIHR, the HTA programme or the Department of Health and Social Care. If there are verbatim quotations included in this publication the views and opinions expressed by the interviewees are those of the interviewees and do not necessarily reflect those of the authors, those of the NHS, the NIHR, the HTA programme or the Department of Health and Social Care.

Core competency 1: active participant involvement

Core competency 2: motivation-building for cutting down (quitting)

Core competency 3: motivation-building for physical activity

Core competency 4: self-monitor, set goals and discuss strategies to reduce smoking

Core competency 5: self-monitor, set goals and discuss strategies to set goals to increase physical activity

Core competency 6: review efforts to cut down smoking/problem-solving

Core competency 7: review efforts to increase physical activity/problem-solving

Core competency 8: integration of concepts – building an association between physical activity and smoking reduction

Core competency 9: identify and reinforce any identity shifts towards being a more ‘healthy person’ or ‘healthy living’

Core competency 10: managing social influences on smoking reduction

Core competency 11: managing social influences on physical activity

Core competency 12: referral to smoking cessation services

Was the issue of making an attempt to stop smoking raised and the response appropriately addressed (i.e. if desired, to make a referral/support access to NHS Stop Smoking Service or other specialist support available in the local area)?

Yes □ No □

_______________________________________________________________________

Below is some guidance on how these CCs may be scored as part of the research process when session recordings are being reviewed. If it is helpful, a scoring approach could also be used in supervision sessions.

The rating scale

The present 7-point scale (i.e. a 0–6 Likert scale) extends from 0, meaning the HT did not deliver the intervention element appropriately, either they did not do it well or did not do it sufficiently (low fidelity), to 6, meaning the element was delivered appropriately (high fidelity). Thus, the scale assesses a composite of adherence to the intended intervention method and skill of the HT. To aid with the rating of items, an outline of the key features of each item is provided at the top of each section above. A description of the various rating criteria is given in Figure 21. The examples are intended to be used as useful guidelines only, providing illustrative anchor points, rather than prescriptive scoring criteria.

FIGURE 21.

Description of the various rating criteria. The scale incorporates the Dreyfus64 system for denoting competence. Please note that the ‘top’ marks (i.e. near the ‘expert’ end of the continuum) are reserved for those HTs demonstrating highly effective skills, particularly in the face of difficulties (i.e. clients with high resistance to change, high levels of emotional expression and complex situational barriers). Please note that there are six competency levels but seven potential scores.