We conducted a historical prospective cohort study among participants who started treatment with varenicline or bupropion in 2007-10. The primary outcomes were the composite of any major cardiovascular event and its individual components acute coronary syndrome (myocardial infarction and unstable angina), ischaemic stroke, and cardiovascular death. The secondary outcomes were other serious cardiovascular events, the individual end points being ischaemic heart disease (including angina pectoris, ischaemic heart disease, and coronary revascularisation), heart failure, peripheral arterial disease, transient ischaemic attack, and cardiac arrhythmia. In most of the trials included in the meta-analysis that had found an increased cardiovascular risk, the treatment lasted for 12 weeks whereas the follow-up times during which cardiovascular events were recorded ranged from 24 to 52 weeks.5
Because this indicated that varenicline might also increase risk after the treatment finished, we set our primary time point of follow-up at six months after the start of treatment. Our secondary analyses included different follow-up times, ranging from six weeks to 24 months. We also carried out subgroup analyses by sex, duration of use, and in participants with and without pre-existing cardiovascular disease.
The source population was defined from the Danish Civil Registration System,11
comprising all Danish people aged ≥18 during the study period. Using the participants’ unique civil registration numbers, we linked individual level information on drug use, hospital contacts, causes of death, and potential confounders.
Users of varenicline and bupropion were identified from the National Prescription Registry.12
This nationwide registry holds information on all prescriptions filled at all Danish pharmacies from 1995, including the anatomic therapeutic chemical (ATC) code, date of filling the prescription, number of tablets, and tablet strength. We established a cohort of new users of varenicline (code N07BA03) and bupropion (code N06AX12), including those who filled a first prescription for either drug during the study period. We excluded individuals who had filled a prescription before 2007. In Denmark, bupropion is not approved for the treatment of depression. Although varenicline was marketed in Denmark in September 2006, the study did not start until 1 January 2007. This allowed the exclusion of participants who were early users as the first users of a newly marketed drug might be highly selected individuals who differ from later and more representative users of the drug.13
For study inclusion, participants had to have been registered in Denmark for at least two years.
Information on cardiovascular outcomes was obtained from the National Patient Registry.14
This nationwide registry holds information on all hospital contacts in Denmark, including all diagnoses and procedures, classified according to ICD-10 (international classification of diseases, 10th revision) and the Nordic Classification of Surgical Procedures (NCSP), respectively. Major cardiovascular events were identified from primary and secondary diagnoses registered during hospital admissions or at emergency departments. Other serious cardiovascular events were identified from registered primary diagnoses only and from records of surgical procedures. Cardiovascular deaths were identified from the Cause of Death Registry,15
which records all deaths in Denmark classified according to ICD-10. Table e1 in the appendix lists the ICD-10 and NCSP codes used for all outcomes.
The National Patient Registry has high validity in the identification of myocardial infarction, with estimated positive predictive value and sensitivity both >90% for a registered diagnosis.16
For ischaemic stroke, the positive predictive value has been estimated as 88-90%.17
For diagnoses of heart failure, peripheral vascular disease, transient ischaemic attack, and atrial fibrillation, the values were >90%, >90%, 60%, and >90%, respectively.17
Information on potential confounders (age, sex, place of birth and place of living; medical history and healthcare use; and use of other selected drugs) was obtained from the Civil Registration System, the National Patient Registry, and the National Prescription Registry, respectively (ICD-10 and ATC codes, table e2 in appendix). Missing values were replaced with mode imputation. The proportion of missing values was <1% for all potential confounders (table e3 in appendix).
After estimation of propensity score, users of varenicline were propensity score matched 1:1 to bupropion users,21
with greedy 5-to-1 matching technique.22
The propensity score was estimated with logistic regression, with all variables listed in tables 1and 2
, and additionally, all estimable two way interactions between demographic and healthcare use variables, included as predictors.
Table 1 Baseline demographic characteristics* of people using varenicline and bupropion to help with tobacco use cessation in nationwide registry based cohort study in Denmark, with follow-up from January 2007 to December 2010. Figures are numbers (percentage) (more ...)
Table 2 Baseline medical characteristics* of people using varenicline and bupropion to help with tobacco use cessation in nationwide registry based cohort study in Denmark, with follow-up from January 2007 to December 2010. Figures are numbers (percentage) (more ...)
Study participants were followed from the date of filling the first prescription for varenicline or bupropion. Treatment status was defined by the initial drug, and study participants were considered always exposed to the respective drug for the entire duration of follow-up. For the primary analyses, participants were followed until the date of censoring (death, disappearance, or emigration), end of study (31 December 2010), switching to the other drug, six months after start of treatment, or event, whichever occurred first. In the analyses of other serious cardiovascular events, an additional censoring criterion was the occurrence of a major event.
We used the Kaplan-Meier method to generate survival curves according to treatment status and compared the groups using the log rank test. We used Cox proportional hazards regression to estimate hazard ratios with 95% confidence intervals, with days since start of treatment as the time scale. The proportional hazards assumption was assessed by a Wald test for the interaction between treatment status and the underlying time scale. P values for comparisons between subgroups were similarly based on the Wald test. All statistical tests were two sided with P<0.05 indicating significance. The statistical analyses were performed with SAS software (version 9.3; SAS, Cary, NC).