Moderators and mediators of the effect of nicotine preloading

Introduction

Chapter 3 showed no evidence that, in the primary analysis, preloading was an effective strategy for promoting smoking cessation in the current NHS setting. This is because preloading deterred the use of varenicline, the most effective smoking cessation medication, as a post-quit day medication.6 However, when adjusted for the use of varenicline, there was evidence of an independent effect of preloading as an effective strategy for promoting smoking cessation. However, in the future, it is plausible that this deterrent effect could be overcome by changing the guidelines or the everyday customs and practices of smoking cessation therapists. Therefore, it is appropriate to understand better how preloading may work and the people for whom it may work best. This may allow us to maximise the benefit for patients and, in those who choose preloading, to monitor the effectiveness of the strategy and discontinue preloading if benefits are unlikely.

Effect modification

This chapter addresses two questions. The first examines effect modification or moderation of the possible treatment effect of preloading. It is possible that preloading was effective only for a subgroup of the population of smokers seeking help to quit. Given its proposed mechanism of action on undermining the reward from smoking and the drive to smoke, we hypothesised that preloading would be more effective for people who showed evidence of higher dependence on cigarettes. This might be evidenced by a higher cigarette consumption at baseline (shown by cigarettes per day or exhaled CO concentration at baseline) or a higher dependence score (FTCD). Understanding this may help us target preloading to the group who stand to gain a greater than average benefit from preloading, thereby maximising the cost-effectiveness of the treatment approach.

Mechanism of action of preloading

The second question addresses the question of how preloading may work.

As discussed in Chapter 1, Mechanisms of action of preloading, our review of the mechanisms of action of preloading revealed scant evidence for any of these mechanisms explaining the possible effectiveness of this strategy. This is important for clinical practice. Nicotine preloading may not help everyone who is trying to stop smoking. Therapists could monitor whether or not preloading is achieving its intermediate effects and, if not, abandon the strategy early, saving resources on behalf of the NHS. In this chapter, we examine the possible mediating pathways between nicotine preloading and outcomes.

Method

Design

The trial protocol and main outcomes are described in Chapters 2 and 3. For the moderation analysis, we followed the same analysis strategy as outlined in Chapter 2, Statistical analysis, but added multiplicative interaction terms to the equation between two markers of dependence assessed at baseline and trial group. As markers of dependence we used FTCD, the standard measure of baseline dependence, and exhaled CO concentration. The latter is a measure of smoke intake in the preceding hours; more dependent smokers draw more heavily on their cigarettes or smoke more cigarettes per day and, hence, tend to have higher CO measurements.

For the mediation analysis, we followed the regression framework described by Kenny.41 The steps are as follows:

1. show that the preloading variable causes smoking cessation (Chapter 3)
2. show that preloading causes change in the mediator
3. show that change in the mediator is associated with smoking cessation
4. to establish complete mediation, show that controlling for change in the mediator abolishes the association between preloading and cessation.

Given that Chapter 3 showed that, on adjustment for the effect of post-cessation varenicline, there was evidence for the efficacy of preloading, this chapter deals with steps 2–4.

Outcomes

For the moderation analyses, the outcome was the primary outcome: Russell Standard 6-month abstinence.

We assessed the impact of preloading on potential mediators of effect outlined in Figure 5. These were measured 1 week after starting preloading or the control condition (3 weeks prior to quit day) and were:

• positive reinforcement – modified Cigarette Evaluation Questionnaire (mCEQ) satisfaction subscale42 and noticing whether cigarettes were more or less enjoyable than previously
• negative reinforcement – mCEQ reward subscale
• drive to smoke – self-reported change in urge strength and change in frequency of the urge to smoke combined, assessed using the Mood and Physical Symptoms Scale – Craving (MPSS-C) subscale;43 smoking stereotypy taken from the Nicotine Dependence Syndrome Scale;44 the mCEQ craving question; and a question used by Hajek et al.21 in a similar trial that asked participants to rate their urge to smoke compared with usual
• cigarette consumption – cigarettes per day and exhaled CO concentration
• symptoms of addiction – modified score on the FTCD,45 modified by removing cigarette consumption from the scale to ensure that symptoms of addiction were distinct from cigarette consumption.

FIGURE 5 Effect of preloading compared with control on the primary outcome, depending on baseline tobacco dependence. (a) FTCD score (0–3, 4–6 and 7–10); and (b) baseline exhaled CO concentration in p.p.m. (Figure not reproduced here; see the original article.)

The hypothesis is that, by undermining the learned addiction to cigarettes during the pre-quitting period, preloading reduces the intensity of craving and withdrawal during the post-quit period. For this, we examined the effect of preloading on the intensity of urges and mood symptoms at the end of the first week of abstinence. We used the MPSS-C and the Mood and Physical Symptoms Scale – Mood (MPSS-M) subscales to measure these. The results of assessments of craving and withdrawal symptoms in those who are smoking can be hard to interpret,46 so we analysed these once in those who had been perfectly abstinent and once in those who were continuing to try to be abstinent.

For our other three mediational hypotheses, we assessed the impact of the possible mediators measured at 1 week after commencement. These were as follows:

1. Medication adherence: days of use of post-quit day medication measured at the +1-week visit.
2. Confidence question: ‘How high would you rate your chances of giving up smoking for good at this attempt?’, measured on a five-point scale (with 1 being ‘not at all’ and 5 being ‘extremely good’).
3. Nausea and aversion: we assessed nausea using the mean scores for two questions derived from themes from participant interviews in a previous trial.47 These were ‘Over the past week how nauseous have you felt when you have seen cigarettes or lighters?’ and ‘Over the past week how nauseous have you felt when you have smelt cigarette smoke?’ and were measured on five-point scales from not at all to extremely. Aversion was measured using the aversion subscale of the mCEQ.

To maximise the sensitivity to detect mediation in the mediator analyses, our outcomes were Russell Standard 4-week and 6-month abstinence. Given that 4-week abstinence is more proximate to the mediators, it should have a stronger association with the mediators than 6-month abstinence.

Statistical analysis

The moderator analysis used the same model as for the primary analysis outlined in Chapter 2, Statistical analysis, but added, in turn, our two markers of dependence (FTCD and exhaled CO concentration) and the multiplicative term representing the baseline marker of dependence multiplied by the trial group indicator variable.

In all of the mediator analyses, we excluded people who had missing data on the mediator. In the analyses, we examined for evidence of mediation on variables assessed at –3 weeks (1 week after baseline) and at +1 week (5 weeks after baseline, 1 week after quit day).

For the second step of the mediation analysis, we assessed the effect of treatment group on mediators descriptively by comparing means between treatment groups and then by analysis of covariance, with adjustment for centre and for the baseline value of the mediator when applicable; thus, effectively examining the effect of preloading on change in the mediator. We checked the residuals to ensure that there was no marked deviation from normality.

We proceeded to the third step of mediation analysis only for those variables for which there was reasonable statistical evidence that preloading influenced the change in the mediator. We assessed the effect of mediators on abstinence outcomes at 4 weeks and 6 months using logistic regression. We looked at this first in unadjusted analysis and then with adjustment for treatment group, as advocated for mediation analysis by Kenny.41 As previously, these analyses were adjusted for stratification by centre.

In the fourth step of mediation, we examined whether or not potential mediators (i.e. those correlated with both the treatment group and the outcome) did indeed mediate the relationship between treatment group and each outcome; this was done by adding each to a logistic regression model including the mediator and the treatment group. Given that there was strong statistical evidence for the efficacy of preloading only when adjusted for use of varenicline, we used the model adjusted for varenicline as the primary model in this analysis. We added each potential mediator singly and then all of the mediators in the same model.

Results

Modification of the effect of preloading on abstinence by baseline tobacco dependence

There was no evidence that people who were more dependent on smoking received a greater benefit from preloading. The p-values for multiplicative interaction terms for the effect of preloading in those with higher dependence scores and a higher exhaled CO concentration were 0.831 and 0.171, respectively. Per unit increase in each variable, the ORs were 1.01 (95% CI 0.90 to 1.14) and 1.01 (95% CI 0.99 to 1.04), respectively (see Figure 5).

Effect of preloading on potential mediators

Mediators measured at –3 weeks

This assessment took place 1 week after baseline and 3 weeks before quit day.

There was evidence that preloading reduced both positive and negative reward (Table 10). Preloading should reduce the drive to smoke and the reduction of reward should also reduce the drive to smoke. There was evidence that three of the four measures of drive to smoke reduced because of preloading. The exception was withdrawal mood symptoms, but measures of drive to smoke specifically all reduced. If this model is correct, then consumption of cigarettes should decrease and this was indeed observed, with a mean modest reduction of 2.6 cigarettes per day (from a baseline mean of 18.9 cigarettes per day) and a reduction in exhaled CO concentration of 3.2 p.p.m. (from a baseline mean of 23.7 p.p.m.). Likewise, perhaps reflecting a decreased drive to smoke, symptoms of addiction (FTCD, excluding daily cigarette consumption) were also shown to have decreased.

Two alternative hypotheses were that preloading improves confidence in quitting and that preloading works through creating an aversion to smoking. There was no evidence that confidence in quitting improved because of preloading but there was strong evidence that both markers of aversion to smoking increased because of preloading (see Table 10). The medication adherence hypothesis could be tested only after quit day.

Mediators measured at +1 week

This assessment took place 1 week after quit day, and potentially after stopping preloading treatment, and 5 weeks after baseline.

As participants were trying to achieve abstinence and many were abstaining, we did not measure positive or negative reward from smoking cigarettes at this assessment. In the main hypothesis, reduced urges to smoke during preloading and a reduced level of smoking mean that people are experiencing cues to smoke that do not prompt smoking, thus beginning the process of extinguishing the learnt drive to smoke. This would manifest as reduced urges to smoke or craving experienced after quit day and there was evidence that this did occur, although there was no evidence that withdrawal mood symptoms were reduced. At this assessment, when participants were abstinent, it was not sensible to ask about any other potential mediators on this pathway.

As in the pre-quit period, there was no evidence that preloading increased participants’ confidence in their ability to quit smoking. Unlike in the pre-quit period, there was no evidence of a difference in aversion to cigarettes after the quit day (see Table 10).

The final mediating hypothesis is that preloading improves adherence to post-cessation medication but there was no evidence to support this (see Table 10).

Association between mediators and outcome (smoking abstinence)

We analysed the association between mediators and outcome for those variables for which there was evidence that preloading influenced the mediator (Table 11).

In the analyses assessing the main hypothesis for the mechanism of action, at –3 weeks, there was no evidence of an association between change in either positive or negative reinforcement from smoking and subsequent abstinence from smoking at 4 weeks or 6 months. Although two of the three measures of drive to smoke were not associated with smoking abstinence, one of these, noticing a change in the strength of urges, was associated with abstinence. Objective reduction in consumption, measured by a reduced CO concentration over the week, was associated with later abstinence, but self-reported reduction in cigarettes smoked per day was not associated with later abstinence. There was no evidence that change in symptoms of addiction was associated with later abstinence (see Table 11). At +1 week, during the abstinence period, reduction in urge strength was associated with later abstinence.

After the second step in the mediation analysis, the only surviving competing hypothesis to our main hypothesis was that the effect of preloading was mediated by creating an aversion to smoking. However, in this third step, there was no evidence that a change in aversion (measured by nausea and the mCEQ aversion subscale at –3 weeks) predicted abstinence.

Effect of preloading on the outcome, controlled for mediators

In the final step of the mediation analysis, we considered whether or not the strength of association between use of preloading and abstinence was diminished after adding three potential mediators. These mediators, all of which showed evidence of potential mediation in the previous steps, were (1) noticing that urges were weaker than usual while still smoking, (2) reduced exhaled CO concentration while still smoking and (3) reduced urge strength after abstinence. As the clear evidence of efficacy of preloading emerged only in the analyses adjusting for post-cessation varenicline use, we used this as the primary comparison for this final stage of mediation analysis. In the data on abstinence, we included all participants in the analysis with the presumption that people who dropped out were smoking. However, where people were not followed up, the value of their mediator was not imputed and so these participants were dropped from the analysis. To ensure that the ORs were comparable across analyses, the number with valid data was used on the mediator in all analyses, including those analyses that did not include the mediator.

There was evidence that each of the mediators explained the effect of preloading to some extent (Table 12). Controlling for a reduction in urge strength reduced the effect of preloading on abstinence by 47% for 4-week abstinence and 59% for 6-month abstinence. Controlling for a reduction in exhaled CO concentration during the pre-quit period reduced the effect of preloading on abstinence by 23% for 4-week abstinence and 14% for 6-month abstinence. Controlling for a reduction in craving after quitting reduced the effect of preloading on abstinence by 28% for 4-week abstinence and 21% for 6-month abstinence. Controlling for all three potential mediators reduced the effect of preloading on abstinence by 71% for 4-week abstinence and 78% for 6-month abstinence.