Use of Novel Case-Only Methods to Investigate the Neuropsychiatric and Cardiovascular

Use of novel case-only methods to investigate the neuropsychiatric and cardiovascular safety of varenicline in the UK Clinical Practice Research Datalink

Investigators

Dr Kyla Thomas, Dr Neil Davies, Dr Ian Douglas, Professor David Gunnell and Professor Richard Martin

1. Lay summary

Smoking is a major cause of poor health and early death in the UK and worldwide. Varenicline is more effective than other stop smoking medicines. However, there are concerns that it may be associated with adverse effects such as suicide and heart attacks. In this study we will use a novel study design to investigate whether varenicline is associated with an increased or decreased risk of suicide, self-harm, heart attacks and death from specific causes (such as lung and heart disease) and all causes.

1. Background and Rationale

Smoking is the major avoidable cause of preventable morbidity and premature mortality in the UK and internationally, due to its causative role in cardiovascular disease, Chronic Obstructive Pulmonary Disease (COPD) and lung cancer [1, 2].Varenicline is the most effective smoking cessation medicine; for every 10 smokers who quit with single form nicotine replacement therapy (NRT) or bupropion, about 16 would be expected to quit with varenicline [3]. However, there are ongoing concerns about the cardiovascular and neuropsychiatric safety of varenicline. This has led regulatory agencies to issue safety warnings about varenicline’s possible adverse effects. Since 2009, the US Food and Drug Administration (FDA) has required the addition of a black box warning to the labelling of varenicline [4]; this is the agency’s strongest safety warning.

Since the original safety concerns, various study designs have been used to investigate these safety issues, including case reports (considered to be at the bottom of the evidence hierarchy), observational cohort studies and meta-analyses (considered to be at the top of the evidence hierarchy) [5]. Whereas studies using data from spontaneous reporting systems have reported an increase in psychiatric adverse effects such as suicide with varenicline use [6], large observational studies and meta-analyses of RCTs have not supported these findings [7-11]. Additionally, large meta-analyses have provided conflicting evidence regarding whether patients prescribed varenicline are at increased risk of adverse cardiovascular events such as myocardial infarction [12, 13]. However, there are concerns about the validity of findings using these study designs. Firstly, although RCTs are considered the gold standard for the evaluation of the intended effects of medicines, they are less appropriate for determining unintended adverse effects. Psychiatric endpoints such as self-harm and suicide are not well suited to experimental studies as the sample size requirements for such rare outcomes, in particular suicide, are prohibitively large [14]. Secondly, although observational studies that utilise large primary care databases are more likely to meet the sample size requirements for rare outcomes, they are prone to residual or uncontrolled confounding, in particular confounding by indication. Confounding by indication may occur if the indication or reason for treatment is also associated with the outcome of interest [15]. For example, the use of smoking cessation medicines may appear to be associated with an increased risk of suicide as smokers themselves have a higher risk of mental illness and suicide compared with non-smokers [16, 17].

Novel epidemiological designs such as case-only designs (which include the case crossover method, case time control method and self-controlled case series), may have greater power to detect rare adverse effects [18]. Because cases act as their own controls, these methods also benefit from the elimination of time invariant confounding factors [18]. Therefore, residual confounding should be minimised in case only studies compared with other conventional observational study designs.

1. Aims and Objectives

The main aim of this project is to investigate the neuropsychiatric and cardiovascular safety of varenicline using the case time control method in the UK CPRD. Nicotine Replacement Therapy (NRT) will also be investigated as a comparator drug. We will investigate whether varenicline is associated with an increased or decreased risk of the following outcomes:

• Suicide
• Non-fatal self-harm
• Myocardial infarction (MI)
• Death from all causes and the following specific causes: MI, lung cancer and Chronic Obstructive Pulmonary Disease (COPD)

1. Study design and type

We will use the case time control design, an extension of case crossover methodology to investigate whether varenicline is associated with an increased or decreased risk of the previously listed outcomes of interest. The case crossover method [19, 20] is a type of case only design which is epidemiologically and statistically comparable to matched case control analyses except the case serves as his/her own control. However, the method assumes no temporal changes in the prevalence of drug treatment. The case time control design [21] was developed to deal with bias caused by changes in the prevalence of exposure over time. It is a more suitable approach for this study as varenicline prescribing may have increased over time due to greater recognition of the drug’s effectiveness or improved marketing. Conversely, safety alerts may have led to a decrease in varenicline prescribing.

1. Inclusion criteria

Patients will be included if they were adult smokers from 1st September 2006 (when varenicline was introduced to the UK) onwards to the most recent date available in the CPRD. Smokers will be defined as patients who have a smoking record (enttype=4) which indicates current smoker (data1 field=1 “Yes”) or Read codes which indicate current smoking after the 1st September 2006. Algorithms to define smoking status will be based on those used in a previous study by Szatkowski and McNeill (2013) in The Health Improvement Network (THIN) database [22]

Records from patients classified as ‘acceptable’ by the CPRD from all up to standard practices at least 18 months prior to date of entry of each cohort (1st January 2005) will be included. Patient data are defined as “acceptable” by the CPRD if they meet minimum quality control standards, for example they have a valid GP registration period.

1. Exclusion criteria

Patients who registered at a practice less than 365 days before the first recorded prescription will be excluded to allow for high quality assessment of baseline data. We will exclude patients prescribed both nicotine replacement therapy and varenicline at the same time. In a previous CPRD analysis, this occurred for 0.25% of all prescriptions (unpublished data).

1. Study population and selection of comparison groups/controls

Cases are all smokers who have experienced one of the outcomes of interest described previously. For each outcome, controls will include smokers who have not experienced the specific outcome of interest, matched by age, sex, general practice and outcome date.

1. Exposures, outcomes and covariates

For each of the outcomes described in the Aims and Objectives, individuals who have experienced that specific outcome of interest will be included as cases in the analysis. As each individual will act as his/her own control for time periods prior to the event/outcome, time invariant confounding should be eliminated. CPRD Read codes will be used to identify self-harm (Appendix 2) and MI [23]. ONS mortality data will be used to identify suicides (see section on linked data below) and deaths from all causes and from MI, COPD and lung cancer. HES data will be used to identify hospital admissions for self-harm. For example, if the outcome is suicide, the current period will be defined as 90 days prior to the suicide and the reference period will be defined as 91 to 180 days prior to the suicide. A time period of 90 days has been chosen for the current period as the maximum recommended treatment duration for varenicline is 12 weeks (or 3 months); NRT can be prescribed for up to 6 months. Use of a particular smoking cessation medicine (identified by codes in Appendix 3) for at least one day in a given current period or reference period will mean that the patient is considered exposed to that medicine for the entire duration of that period. The primary analyses will focus on exposure to varenicline.

1. Data/Statistical Analysis

9.1 Main analyses

Each patient will form both halves of a matched pair, comparing exposure to varenicline in the current period (90 days prior to the outcome event) with exposure to varenicline in a single reference period (90 days before the current period). For the cases, this analysis will be analogous to the case crossover design. However, if varenicline were associated with a specific outcome, for example suicide, then the odds of being prescribed this medicine in the 90 days immediately before the suicide would be higher than the 90 days before that. Also, using the cases alone could lead to biased results if the rate of prescribing of the smoking cessation medicine had changed over the 180 day period. The case time control design removes the effect of time by performing the analysis in the group of matched controls in addition to the group of cases. By subtracting the effect estimate derived from controls from the estimate obtained from the cases, it is possible to estimate the effect of exposure to the smoking cessation medicine corrected for time trends for the exposure in the population.

For the case time control design, conditional logistic regression will be used on matched pairs of individual patients at the current and reference time periods. The conditional logistic regression model is given by:

Logit(R) =?0+ ?1.E + ?2.E.G

Where R is the probability of an outcome event (for example the risk of a suicide event), E is a binary indicator variable for the exposure to varenicline (1= exposed, 0= unexposed) and G is a binary indicator of case or control status (1=case, 0=control). The odds ratio associated with the exposure is derived from the interaction term between the exposure E and the case/control group G:

OR = exp(?2)

No cells with counts <5 will be reported (in accordance with CPRD disclosure policy).

9.2 Sensitivity analyses

Sensitivity analyses will be repeated with 30 days and 180 days prior to the event as the current period. The corresponding reference periods would be 31-60 days prior to the event and 181-360 days prior to the event.

9.3 Use of multiple reference periods

Multiple reference periods will also be used to increase the power of the analyses up to a maximum of four reference periods to one current period in the main and sensitivity analyses. For example in the main analyses exposure to varenicline in the current period (90 days prior to the event) will be compared with exposure to varenicline in four 90 day reference periods (91-180 days prior to the event, 181-270 days prior to the event, 271 to 360 days prior to the event and 361 to 450 days prior to the event.

9.4 Use of NRT as a comparator

The analyses described in sections 9.1 to 9.3 will be repeated replacing exposure to varenicline with exposure to NRT.

9.5 Missing covariates

Due to the within person nature of this study we do not anticipate any major issues; time invariant confounding will be minimised. It is unlikely that we will be able to reliably impute possible time varying confounders such as alcohol consumption or depression for such small periods of time. However, we believe this is a small limitation as it would be the case for any study in the CPRD, regardless of study design.

1. Use of linked data (if applicable)

Previous research has shown that suicide reporting using CPRD Read codes is unreliable [24]. Therefore linked ONS mortality data will be used to identify cases of suicide. Linked HES data will be used to identify hospital admissions for self-harm. Linked ONS mortality data will be used to identify death from specific causes such as MI and lung cancer. We will compare baseline characteristics including deprivation levels (measured by IMD) in smokers prescribed varenicline with those prescribed NRT.

1. Sample size/Power calculation

Odds ratios of 1.5 or greater will be considered to be clinically meaningful as they indicate a 50% or higher increased risk of the events of interest. Estimates obtained from the CPRD Gold Database (June 2014 version) show that 2280 individuals died from suicide between September 1st 2006 (the date when varenicline was first licensed) and January 10th 2012 (the current end date for ONS linkage in the June 2014 version); 12,847 died from MI in the same time period. In 2013 2.58% of the CPRD population were exposed to varenicline.

As the case crossover study is statistically similar to a matched case control study we used the Stata command sampsi_mcc to estimate power in Stata version 13. The theory behind this command is described in Dupont (1988) [25]. Using two time periods of 90 days (1 current period and 1 reference period) andassuming that the prevalence of exposure to varenicline in the reference period is 0.0258, the study will have 98% power at 5% significance to detect a doubling of the odds of suicide (OR=2) while being prescribed varenicline. Using 1 current period and 4 reference periods, the study will have 99.9% power at 5% significance to detect an OR of 2. Table 1 summarises the power estimates at 5% significance for different matching ratios and odds ratios.

Table 1 Power estimates at 5% significance for suicide and deaths from MI

1. Patient/user group involvement

There has been no patient/user group involvement in this study. However, the linked study (ISAC protocol 15_107) includes patient involvement.

1. Limitations of the study design, data sources and analytic methods

This analysis has the following limitations:

1. This is an observational study and is prone to residual confounding although the case only design should minimise time invariant confounding.

2. This study is restricted to products prescribed in primary care, so patients receiving smoking cessation products in NHS smoking cessation clinics or buying over-the-counter nicotine replacement products from pharmacies will be excluded. Patients receiving prescriptions for nicotine replacement products are likely to be poorer than those who choose to buy these products over the counter. Furthermore those visiting their GP for prescriptions of smoking cessation products may differ from those attending specific smoking cessation clinics, meaning findings may not be generalisable to the wider population of people taking smoking cessation medicines.

3. Suicide is a rare outcome so it is possible that we may not have sufficient power to analyse this outcome in this study.

4. It is possible that the case time control approach of adjusting for time trends in prescribing may re-introduce confounding into the study [26]{Greenland, 1996 #29315}{, 1861 #1}.

1. Plans for disseminating and communicating study results

Key findings will be collated to form evidence based recommendations which will be communicated to relevant groups, with the aim of improving the evidence base to inform advice to prescribers and patients. We will also aim to publish findings in peer reviewed journals and present our work at national and international conferences.

References

1.Peto, R., et al., Smoking, smoking cessation, and lung cancer in the UK since 1950: combination of national statistics with two case-control studies. BMJ, 2000. 321(7257): p. 323-9.

2.Jha, P. and R. Peto, Global Effects of Smoking, of Quitting, and of Taxing Tobacco. N Engl J Med, 2014. 370(1): p. 60-68.

3.Cahill, K., et al., Pharmacological interventions for smoking cessation: an overview and network meta-analysis. Cochrane Database Syst Rev, 2013. 5.

4.US Food and Drug Administration Information for Healthcare Professionals: Varenicline (marketed as Chantix) and Bupropion (marketed as Zyban, Wellbutrin and generics). FDA Drug Safety Newsletter, 2009. 2.

5.Strom, B.L., Study designs available for Pharmacoepidemiology studies, in Textbook of pharmacoepidemiology, B.L. Strom and S.E. Kimmel, Editors. 2006, John Wiley and Sons: Chichester, West Suusex, England. p. 13-24.

6.Moore, T.J., et al., Suicidal Behavior and Depression in Smoking Cessation Treatments. Plos One, 2011. 6(11): p. e27016.

7.Gunnell, D., et al., Varenicline and suicidal behaviour: a cohort study based on data from the General Practice Research Database. British Medical Journal, 2009. 339.

8.Meyer, T.E., et al., Neuropsychiatric events in varenicline and nicotine replacement patch users in the Military Health System. Addiction, 2013. 108(1): p. 203-10.

9.Gibbons, R.D. and J.J. Mann, Varenicline, Smoking Cessation, and Neuropsychiatric Adverse Events. Am J Psychiatry, 2013.

10.Thomas, K.H., et al., Smoking cessation treatment and the risk of depression, suicide and self-harm in the Clinical Practice Research Datalink: prospective cohort study BMJ, 2013. 347: p. f5704.

11.Thomas, K.H., et al., Risk of neuropsychiatric adverse events associated with varenicline: systematic review and meta-analysis. BMJ, 2015. 350: p. h1109.

12.Singh, S., et al., Risk of serious adverse cardiovascular events associated with varenicline: a systematic review and meta-analysis. Canadian Medical Association Journal, 2011. 183(12): p. 1359-1366.

13.Prochaska, J.J. and J.F. Hilton, Risk of cardiovascular serious adverse events associated with varenicline use for tobacco cessation: systematic review and meta-analysis. BMJ, 2012. 344.

14.Gunnell, D., J. Saperia, and D. Ashby, Selective serotonin reuptake inhibitors (SSRIs) and suicide in adults: meta-analysis of drug company data from placebo controlled, randomised controlled trials submitted to the MHRA's safety review. BMJ, 2005. 330: p. 385-388.

15.Walker, A.M., Confounding by indication. Epidemiology, 1996. 7(4): p. 335-6.

16.Miller, M., et al., Cigarette smoking and suicide: a prospective study of 300,000 male active-duty Army soldiers. Am J Epidemiol 2000. 151(11): p. 1060-1063.

17.Miller, M., D. Hemenway, and E. Rimm, Cigarettes and suicide: a prospective study of 50,000 men. Am J Public Health, 2000. 90(5): p. 768-773.

18.Nordmann, S., et al., Case-Only Designs in Pharmacoepidemiology: A Systematic Review. PLoS ONE, 2012. 7(11): p. e49444.

19.Maclure, M., The Case-Crossover Design: A Method for Studying Transient Effects on the Risk of Acute Events. American Journal of Epidemiology, 1991. 133(2): p. 144-153.

20.Delaney, J.A. and S. Suissa, The case-crossover study design in pharmacoepidemiology. Statistical Methods in Medical Research, 2009. 18(1): p. 53-65.