Parts of this text have been reproduced from Taylor et al.17 © The Author 2017. Published by Oxford University Press on behalf of the International Epidemiological Association. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited.
In this chapter, we estimate the effects of prescribing varenicline or NRT on patients’ likelihood of smoking cessation using multivariable-adjusted, propensity score and instrumental variable regression. Varenicline has been demonstrated to be efficacious in RCTs and network meta-analysis of RCTs. However, these trials had < 1 year of follow-up and may not have been representative of the population seen in primary care.
Population characteristics
Between 1 September 2006 and the end of follow-up, 1,875,980 smoking cessation prescriptions were issued to 287,079 patients. Of these, 220,136 patients met our eligibility criteria (see Figure 2 for a flow chart of exclusion and inclusion of prescriptions and patients into the study). Patients who were prescribed NRT received patches, gum, oral or nasal spray, inhaler, lozenges or microtabs, and 23% received more than one type of NRT at their first prescription. The proportion of current smokers who were prescribed NRT or varenicline in primary care has fallen between 2007 and 2015 (Figure 3). Nationally, the number of NRT prescriptions issued fell by 65% from 110,608 in February 2013 to 38,303 in January 2018. Prescriptions of varenicline fell by 59% from 71,551 in February 2013 to 29,645 in January 2018.25
FIGURE 3 The per cent of smokers prescribed NRT or varenicline between 2007 and 2015 in the CPRD. (Figure not reproduced here; see the original article.)
Table 1 shows the baseline characteristics of the patients who were prescribed varenicline and NRT. Patients who were prescribed varenicline were healthier in almost every way. They were younger, visited their GP less frequently, had fewer comorbidities, were less likely to have misused alcohol or drugs, were less likely to have a diagnosis of bipolar disorder, depression, a neurotic disorder, self-harm, a rare mental disorder, or were less likely to be prescribed antidepressants, antipsychotics, hypnotics/anxiolytics or other psychotropic medications. There were few differences by BMI or length of prior registration.
| Characteristic | NRT (N = 149,526) | Varenicline (N = 70,610) | Whole sample (N = 220,136) |
|---|---|---|---|
| Age at time of their first prescription (years),a mean (SD) | 46.4 (15.4) | 44.5 (13.2) | 45.8 (14.9) |
| Sex (female), % (n) | 53.7 (80,348) | 50.2 (35,466) | 52.6 (115,814) |
| IMD score, median | 12 | 12 | 12 |
| Number of GP visits 1 year prior to first prescription,a mean (SD) | 7.9 (7.4) | 6.3 (6.1) | 7.4 (7.0) |
| BMI (kg/m2),a,b mean (SD) | 26.4 (6.4) | 26.5 (5.9) | 26.4 (6.1) |
| Year of first prescription | 2009 | 2010 | 2009 |
| Days of history,a mean (SD) | 3158.7 (1892.1) | 3283.9 (1976.6) | 3198.9 (1920.5) |
| Comorbidity ever (Charlson Comorbidity Index), % (n) | 37.6 (56,274) | 31.9 (22,523) | 35.8 (78,797) |
| Alcohol misuse ever, % (n) | 8.3 (12,422) | 6.0 (4199) | 7.6 (16,621) |
| Drug misuse ever, % (n) | 3.1 (4595) | 1.9 (1357) | 2.7 (5952) |
| Bipolar disorder ever, % (n) | 1.0 (1464) | < 1.0 (160) | < 1.0 (1624) |
| Depression ever, % (n) | 35.0 (52,233) | 29.2 (20,615) | 33.1 (72,848) |
| Neurotic disorder ever, % (n) | 24.7 (36,921) | 20.1 (14,189) | 23.2 (51,110) |
| Self-harm ever, % (n) | 10.6 (15,903) | 8.7 (6169) | 10.0 (22,072) |
| Other rare mental disorder ever, % (n) | 6.9 (10,343) | 4.0 (2832) | 6.0 (13,175) |
| Antidepressant prescription ever, % (n) | 50.1 (74,921) | 43.1 (30,435) | 47.9 (105,356) |
| Antipsychotic prescription ever, % (n) | 20.0 (29,873) | 14.8 (10,459) | 18.3 (40,332) |
| Hypnotics/anxiolytics prescription ever, % (n) | 21.1 (31,513) | 17.6 (12,415) | 20.0 (43,928) |
| Other psychotropic medication, % (n) | < 1.0 (473) | < 1.0 (120) | < 1.0 (593) |
Differences in smoking cessation rates by treatment prescribed
Patients prescribed varenicline were more likely to be recorded as a former smoker at all follow-up periods (Figure 4). Three months after the first prescription, 18.8% and 14.1% of patients prescribed varenicline and NRT, respectively, were recorded as a former smoker. These proportions increased to 32.3% and 28.4%, respectively, by 4 years after the first prescription. Figure 5 and Table 2 illustrate the relative difference in quit rates between the two treatments over time. Patients prescribed varenicline were 47% [95% confidence interval (CI) 42% to 52%] more likely to quit after 3 months than patients prescribed NRT. The relative difference in quit rates fell over time; patients prescribed varenicline were 23% (95% CI 21% to 26%) more likely to have quit after 4 years than patients prescribed NRT. These differences slightly attenuated, but remained detectable, after multiple imputation and adjustment for a full range of confounders.
FIGURE 4 The proportion of patients who quit, by medication prescribed, at 3, 6 and 9 months and 1, 2 and 4 years after exposure (n = 220,136). Reproduced from Taylor et al.17 © The Author 2017. Published by Oxford University Press on behalf of the International Epidemiological Association. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. (Figure not reproduced here; see the original article.)
FIGURE 5 Patients prescribed varenicline were more likely to be non-smokers 3, 6 and 9 months and 1, 2 and 4 years after their first prescription than those prescribed NRT. Fully multivariable-adjusted logistic regression model using multiply imputed data: odds ratios and 95% CIs presented. The difference in smoking cessation rates peaks at 6 months and declines over the following 3.5 years (n = 220,136). Reproduced from Taylor et al.17 © The Author 2017. Published by Oxford University Press on behalf of the International Epidemiological Association. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. (Figure not reproduced here; see the original article.)
| Model | Odds ratio (95% CI) | |||||
|---|---|---|---|---|---|---|
| 3 months | 6 months | 9 months | 1 year | 2 years | 4 years | |
| Partial adjustedb | 1.47 (1.42 to 1.52) | 1.50 (1.46 to 1.55) | 1.44 (1.4 to 1.48) | 1.38 (1.35 to 1.42) | 1.30 (1.27 to 1.33) | 1.23 (1.21 to 1.26) |
| Fully adjustedc | 1.42 (1.38 to 1.47) | 1.46 (1.42 to 1.50) | 1.40 (1.36 to 1.44) | 1.34 (1.31 to 1.38) | 1.26 (1.23 to 1.29) | 1.19 (1.16 to 1.21) |
In almost every way we could measure, patients who were prescribed varenicline were different from those who were prescribed NRT. As a result, we were concerned that our multivariable results might suffer from residual confounding. Therefore, we estimated the effects of the medications using propensity score and instrumental variable methods. The patients’ characteristics were well balanced across the propensity score (see Report Supplementary Material 1, Figures S1 and S2). The covariate balance tests presented in Report Supplementary Material 1, Figure S3, suggest that the measured covariates were more balanced across the instrument than across the actual prescription. The effects of prescribing varenicline estimated using the propensity score and instrumental variable analysis were consistent with the multivariable-adjusted regression above (see Report Supplementary Material 1, Tables S1 and S2).
Heterogeneity in the effect of varenicline by socioeconomic position
Patients from more deprived areas were 9% (95% CI 8% to 10%) less likely to be prescribed varenicline after controlling for age and sex. Patients from areas in the least deprived quintile were slightly more likely to have quit between 3 and 12 months after their first prescription. However, these differences could be attributable to chance (p > 0.05). The longer-term (after 2 and 4 years) differences in relative quit rates by treatment were similar in areas of high and low deprivation (Figure 6). Therefore, there was little evidence that the effects of varenicline and NRT differed across socioeconomic groups. There was insufficient statistical power to detect treatment effect heterogeneity using instrumental variable analysis (see Report Supplementary Material 1, Tables S3 and S4). However, instrumental variable point estimates suggested that varenicline increased the likelihood of smoking cessation at all follow-up periods.
FIGURE 6 The effectiveness of varenicline stratified by SEP. Partially adjusted odds ratios and 95% CIs for the association of prescription of varenicline vs. NRT and smoking cessation at 3, 6 and 9 months and 1, 2 and 4 years after exposure, by the level of deprivation as measured by the IMD. IMD is an increasing measure of neighbourhood disadvantage. Reproduced from Taylor et al.17 © The Author 2017. Published by Oxford University Press on behalf of the International Epidemiological Association. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. (Figure not reproduced here; see the original article.)
Sensitivity analyses
Missing data
Patients prescribed varenicline and NRT had similar numbers of missing data. After 2 years, 20.3% of patients had no new smoking status records (see Report Supplementary Material 1, Table S5). The patients who were missing smoking status records were more likely to be younger and male, were less likely to visit the GP and had fewer comorbidities (see Report Supplementary Material 1, Table S6). We investigated whether or not the missing smoking status records affected our results using multiple imputation. The multiply imputed results were similar to our main results, which assumed that patients with no smoking status records in the follow-up period were continuing smokers (see Report Supplementary Material 1, Table S7).28
Varenicline versus nicotine replacement monotherapy versus multitherapy
Patients who were prescribed varenicline were more likely to have ceased smoking after 2 years than those who were prescribed any NRT [odds ratio (OR) 1.30, 95% CI 1.27 to 1.33]. Patients who were prescribed varenicline were more likely to have ceased smoking after 2 years than patients who were prescribed NRT monotherapy defined by route of administration (OR 1.32, 95% CI 1.29 to 1.36). Patients who were prescribed varenicline were more likely to have ceased smoking after 2 years than patients who were prescribed NRT multitherapy (OR 1.23, 95% CI 1.19 to 1.27). Patients who were prescribed NRT multitherapy were more likely to have ceased smoking after 2 years than those prescribed NRT monotherapy (OR 1.08, 95% CI 1.05 to 1.12). After adjustment for the same set of covariates as we included in the primary analysis, the OR of cessation at 2 years comparing varenicline with NRT multitherapy was 1.22 (95% CI 1.17 to 1.28). The fully adjusted OR for cessation at 2 years comparing varenicline with NRT monotherapy was 1.23 (95% CI 1.19 to 1.28). After adjustment, there was little evidence of differences in quit rates between those prescribed NRT monotherapy and NRT multitherapy (OR 1.01, 95% CI 0.97 to 1.05). These results suggest that patients who were prescribed varenicline were more likely to be non-smokers after 2 years than patients who were prescribed either NRT monotherapy or NRT multitherapy.
Treatment compliance
On average, patients who were prescribed varenicline were issued 2.8 [standard deviation (SD) 1.6] prescriptions for varenicline in the 3 months following their first eligible prescription, and received an average of 107.5 (SD 120.7) tablets; 49.7% (35,076/70,610) of these patients received a full course of varenicline (i.e. ≥ 12 weeks) (see British National Formulary, chapter 4.10.214). Patients who were initially prescribed NRT were issued 2.5 (SD 2.2) prescriptions for NRT on average during the 3 months after their first eligible prescription. Patients who were issued a single varenicline prescription only were less likely to cease smoking after 2 years than those who were prescribed any NRT (OR 0.59, 95% CI 0.57 to 0.62). Patients who were prescribed at least two varenicline prescriptions were more likely to cease smoking after 2 years than those prescribed any NRT (OR 1.63, 95% CI 1.59 to 1.67). Patients prescribed two or more varenicline prescriptions in the 2 months following their first prescription were more likely to cease smoking than those who were issued a single varenicline prescription (OR 2.80, 95% CI 2.67 to 2.92). After adjustment for the same set of covariates as we included in the primary analysis, the difference in cessation between patients issued a single varenicline prescription and those issued NRT increased (OR 0.55, 95% CI 0.51 to 0.58). The difference in cessation at 2 years between patients prescribed two or more varenicline prescriptions and any NRT attenuated (OR 1.51, 95% CI 1.46 to 1.57). The adjustment did not meaningfully alter the difference between patients prescribed two or more varenicline prescriptions and those prescribed a single varenicline prescription (OR 2.79, 95% CI 2.61 to 2.97). These results suggest that patients who were prescribed two or more varenicline prescriptions, who have a longer duration of treatment, were less likely to be smokers after 2 years than those prescribed either NRT monotherapy or NRT multitherapy.
Multiple quit attempts
Patients who were prescribed varenicline were more likely to quit smoking than patients who were prescribed NRT who had previously had any smoking cessation prescriptions (OR 1.27, 95% CI 1.22 to 1.31). After adjustment for the same set of covariates as included in the primary analysis, the association in patients previously prescribed smoking cessation prescription remained similar (OR 1.27, 95% CI 1.22 to 1.33). Patients who had never previously been prescribed smoking cessation treatment who were prescribed varenicline were more likely to cease smoking than those who were prescribed NRT (OR 1.32, 95% CI 1.29 to 1.36). This association attenuated after adjustment for the covariates included in the primary analysis (OR 1.23 95% CI 1.19 to 1.29). Patients who had previously received a smoking cessation prescription were less likely to quit (OR 0.90, 95% CI 0.88 to 0.92) than those who had never received a smoking cessation prescription. This difference attenuated after adjustment (OR 0.93, 95% CI 0.91 to 0.96). These results suggest that patients prescribed varenicline are more likely to have ceased smoking after 2 years, irrespective of whether or not they had previously been prescribed smoking cessation therapy. Patients who had previously received smoking cessation therapies were less likely to quit.
Comparison with other studies
Figure 7 compares our results with the previously published results from RCTs. Our estimates of the effectiveness of varenicline compared with NRT at 6 months were similar to, and considerably more precise than, the results estimated via the large RCT reported by Anthenelli et al.10 and a network meta-analysis conducted by Cahill et al.7 Combined, the studies suggested that, on average, patients prescribed varenicline were 44% (95% CI 33% to 55%) more likely to quit smoking after 6 months.
FIGURE 7 Random-effects meta-analysis comparing effect estimates (OR and 95% CI) from existing studies examining the effect of varenicline vs. NRT for smoking cessation at 6- to 12-month follow-up. Reproduced from Taylor et al.17 © The Author 2017. Published by Oxford University Press on behalf of the International Epidemiological Association. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. (Figure not reproduced here; see the original article.)