What Are the Effects of Varenicline Compared With Nicotine Replacement Therapy on Long Term Smoking Cessation and Clinically Important Outcomes?

1. Background

Smoking is the major avoidable cause of preventable morbidity and premature mortality in the UK and internationally. Smoking is also the principal cause of health inequalities and is responsible for most of the difference in healthy life-expectancy between the richest and poorest in our society and those with and without mental health problems. It has been estimated that smoking-related illnesses cost the NHS approximately £5bn per year. Varenicline has been shown to be the most clinically effective smoking cessation medicine for short-term abstinence in randomized controlled trials (RCTs). However, there is very little evidence for its long-term effectiveness and impact on clinical outcomes which are relevant to the NHS.

Although RCTs are considered the gold standard for the evaluation of the intended effects of medicines, they are less appropriate for determining unintended or unexpected beneficial or adverse effects for several reasons. Data from RCTs of varenicline typically have less than a year of follow-up, with relatively low power to detect effects on clinically-relevant outcomes, and are not informative about the longer-term consequences of varenicline treatment. Furthermore, these RCTs usually exclude smokers with chronic respiratory or heart disease and psychiatric illnesses, so their findings may not be generalisable to all patients. Observational studies using large primary care databases of electronic health records may be used to address these unanswered questions, as the investigators propose to do here.

The investigators recently investigated the short-term relationships between smoking cessation treatment and suicide, self-harm, depression and all-cause mortality using CPRD data. The investigators found no evidence of an increased risk of suicidal behaviour in patients prescribed varenicline. The investigators were concerned that the conventional observational estimates of effects of varenicline were biased by residual confounding by indication. If patients prescribed varenicline differed from patients prescribed other smoking cessation treatments prior to their first prescription, then the investigators could not be certain whether any differences in outcomes were due to the medication the patients received, or because of the pre-existing differences between the patients. In the investigators previous study the investigators tried to overcome confounding by indication using three statistical approaches: conventional multivariable-adjusted regression, propensity score matched analysis, and an instrumental variable analysis.

The investigators used physicians' prescribing preferences as instrumental variables, as described in Section 1, for the physicians' prescriptions of varenicline or nicotine replacement products. In the conventional observational analysis, after adjustment for a wide range of observable patient characteristics, patients prescribed varenicline had a 51% lower risk of all-cause mortality than those prescribed other nicotine replacement products in the nine months after their first prescription, (hazard ratio=0.49, 95% confidence interval (95%CI): 0.40 to 0.61). However, the investigators found little evidence of a reduction in all-cause mortality using instrumental variable analysis (risk difference per 1,000 prescriptions=0.7, 95% CI: -3.3 to 4.7). This suggests that the conventional multivariable-adjusted regression may have suffered from bias due to unmeasured confounding, as the effect sizes are implausible.