A Randomized Placebo-controlled Trial to Test a Genetically-informed Biomarker ForPersonalizing Treatment for Tobacco Dependence

Design and Participants

Participants were randomized, by NMR group, to one of three treatment arms: (1) nicotine patch (placebo pill + active patch); (2) varenicline (active pill + placebo patch); or (3) placebo (placebo pill + placebo patch) (Figure 1). Our primary aim was to contrast the efficacy of nicotine patchvs.vareniclineby NMR group (NMsvs.SMs). Aplacebocondition wasincluded to examineside-effectsof treatment by NMR group. The primary endpoint was point prevalence abstinence at the end of treatment (EOT) to estimate the pharmacologic effect by NMR group during the medication period. Six- and 12-month follow-up data are presented as secondary endpoints.

We conducted the clinical trial at four academic medical centers (University of Pennsylvania (UPenn), Centre for Addiction and Mental Health/University of Toronto, State University of New York at Buffalo, and MD Anderson Cancer Center); assessment of the NMR was performed at the University of Toronto. FromNovember 2010-September 2013, we recruited participants through advertisements for a free smoking cessation program. Eligible participants were 18–65 years old and reported smoking ≥10 cigarettes/day for ≥6 months (verified by carbon monoxide (CO) >10 ppm).

Exclusion criteria included: 1)use of non-cigarette tobacco products, e-cigarettes, or current smoking treatment; 2) history of substance abuse treatment, current use of cocaine or methamphetamine, or >25 alcoholic drinks/week; 3) medical contraindications (pregnancy, history of cancer, kidney or liver disease or transplant, clinically significant cardiac dysrhythmias, stroke, angina, heart attack, or uncontrolled hypertension); 4) history of DSM-IV Axis 1 psychiatric disorder or suicide risk score on the MINI International Neuropsychiatric Interview (MINI)>1,or current major depression; 5) current use of anti-psychotics, stimulants, opiate medications, anti-coagulants, rescue inhalers, antiarrythmics, or medications altering CYP2A6 activity (e.g., monoamine oxidase inhibitors, tricyclic antidepressants); and 6) inability to provide informed consent or any condition that could compromise safety.

Randomization and Masking

A biostatistician, independent of the study investigators, developed the randomization procedure which was integrated into a centralized data management system. Subjects were randomized to the treatment arms in a 1:1:1 ratio. Randomization was stratified by baseline NMR status and study site, andblocked in blocks of 12 patients (4/treatment/block) to ensure approximate balance.Participants, study investigators, and personnel (except for the biostatistician and senior data manager) were masked to treatment arm allocation and NMR status. Data were unmasked for analysis following collection of all 6-month follow-up data.

Procedures

The institutional review boards at all sitesapproved the protocol. Participants eligible at telephone screening completed an in-person medical exam and psychiatric history, completed self-report measures of demographics and smoking history, and provided blood samples for the NMR assay. NMR results, reported within 7 days, wereused for final eligibility determination. SMs were oversampled to create a study sample comprising roughly equal numbers of SMs and NMs. For the first 25% of participants enrolled, the NMR cutoff for stratification (and oversampling of SMs) was 0·26 based on a prior clinical trial of NRT.¹⁵However, to ensure achievement of recruitment goals, the Data and Safety Monitoring Board made a determination (masked to study outcomes) to increase the stratification cut-off to 0·31 (redefining SMs as NMR < 0·31 and NMs(including rapid metabolizers) as those with NMR > 0·31). This cut-off was based on graphical representation of the NMR in the screened population (Supplemental Figure 1A), and was used inall analyses.

Following an in-person1-hour pre-quit group counseling session at the local clinical site (week-1), varenicline (or matching placebo) was initiated using the guidelines for one week precessation dose titration.Nicotine patch (or matching placebo) was initiated on the morning of the target quit date (TQD). Brief (~15 minute),protocol-driven,telephone counseling was delivered by two counselors at UPenn(weeks 0 (TQD), 1, 4, 8). Counseling focused on skills to quit and avoid relapse, instructions on the use of medication, and medication compliance. Staff at UPenn collected mid-treatment data (e.g., withdrawal, side-effects) by telephone. Self-reported smoking status was assessed using a standard timeline follow-back procedure,¹⁸ and biochemically verified (see below).

TheUPenn Investigational Drug Service distributed medication to the clinical sites. Active patches were purchased from GlaxoSmithKline(Nicoderm CQ®) and identical placebo patches were purchased from Rejuvenation Labs (Salt Lake City, UT).Participants received 11 weeks of patches to match the duration of varenicline following the TQD: 21 mg (6 weeks), 14 mg (2 weeks), and 7 mg (3 weeks). Pfizer manufactured active varenicline and matching placebo pills. Vareniclinewas delivered for 12 weeks (1 week pre-TQD) as in past trials:¹⁷Days 1-3 (0.5mg once daily); Days 4–7 (0.5mg twice daily); and Days 8–84 (1.0mg twice daily).

Measures

We assessed demographics, smoking rate, and nicotine dependence (Fagerström Test for Nicotine Dependence).¹⁹The MINI determined lifetime history of Axis I psychiatric diagnoses.For the NMR assay, cotinine and 3’hydroxycotinine were assessed by LC-MS; limits of quantification were<1 ng/ml whole blood.²⁰

The primary endpoint was7-day point prevalence abstinence at EOT (week 11). This outcome was chosen based on guidelines for smoking cessation trials.²¹Abstinencewasdefined as no self-reported smoking (not even a puff) for at least 7 days prior to the telephone assessment, with in-person verification for those self-reporting abstinence (CO < 8ppm).²¹ As is the convention, participants lost to follow-up were considered smokers.²¹To estimate the pharmacologic effect by NMR, subjects should be on medication; therefore, the EOT quit rate was the primary end point. Six- and 12-month quit rates were secondary endpoints to assess whether pharmacogenetic effects persisted after treatment was discontinued.

Self-reported withdrawal symptoms were measured using the Minnesota Nicotine Withdrawal Scale (pre-quit and weeks 0, 1, 4).²²A self-report checklist measured the severity of common side-effects (none [0]tosevere[3]), at weeks 0 (TQD), 1, and 4, and was summed to create a side-effects index (we refer to these as “side-effects”, rather than adverse events to distinguish these responses from the open-ended serious adverse events (SAEs) reported by participants).¹⁵ Weassessed self-reported use of pills and patches at each visitand collected unused patches and pill blister-packages to confirm self-reports.²³

Statistical Analysis

To test the primary hypothesis of an NMR-by-treatment interaction at EOT, we estimated a logistic regression model using all data. This model included parameters for the odds ratios (ORs) for the treatment effects (nicotine patch vs. varenicline) within SMs and NMs. One measures the interaction effect as the ratio of odds ratios (ORR), calculated by exponentiating the coefficients corresponding to the interaction terms in the logistic model. These models were repeated for 6- and 12-month cessation outcomes. We also used longitudinal logistic regression (General Estimating Equations, GEE) to examine the NMR-by-treatment interaction incorporating alltime points. All models controlled for study site; we also tested our multivariate models controlling for sex, cigarettes per day, nicotine dependence, and race.Self-reported side-effects (continuous measure from the checklist) were examined for each active treatment (vs. placebo) within each NMR group using GEE models including time-point, and adjusted for study siteand pre-quit levels of side-effects; the NMR-by-treatment interaction was estimated as a difference of differences (beta coefficients, which estimate mean differences in side effect severity).²⁴The correlation structure used in the GEE model for side effects utilized subject specific random effects. The model was linear and the measures were continuous. Assumptions were assessed and the distribution was non-normal. However, we verified that we obtained the same results using bootstrap methods (with exact confidence intervals) that are not sensitive to violations of normality.Lastly,we conducted receiver operating curve (ROC) analyses with abstinence as the binary response and NMR as the continuous predictor, separately in the three treatment arms. We tested for heterogeneity among ROC values using the methods of DeLong and colleagues²⁵as implemented in Stata ROCCOMP.

Our original target sample was 1350. The Data Safety and Monitoring Board conducted a masked interim futility analysis, based on the conditional power method,²⁶ in Feb 2013. In addition to this analysis, we recalculated the power and determined that an N of 1200 would provide adequate power.Specifically, the sample size of 1,200 provided 80% power to detect an ORR of 3·2 for the NMR-by-treatment (nicotine patch vs. varenicline) interaction at EOT. In models that included the placebo arm, the full ITT sample of 1246 participants(662 SM; 584 NM) was used.In models that compared varenicline to nicotine patch, the ITT sample of 838 was used. Stata (Version 13; Stata Corporation; College Station, TX) was used for the analyses.

Role of Funding Source

No funder had a role in study design, collection and analysis of data, writing of the manuscript, or in the decision to submit this manuscript for review. The corresponding author had complete access to all data.

Sample Characteristics

Retention rates at EOT exceeded 70% and retention did not vary across treatment arms or NMR group (Figure 1). The treatment arms did not differ on demographic and smoking history variables(Table 1). SMs were younger (p=0·02), less likely to be Caucasian (p<0·0001), more likely to be male (p=0·001), and smoked fewer cigarettes/day (p<0·001),vs.NMs, as reported previously.²⁷Sex differences in the NMR groups are expected due to estrogen effects on nicotine metabolism rate,²⁷ and the ethnicity difference is expected due to differences in the frequency of reduced/inactive CYP2A6 alleles. The average NMR for SMs was0·20 (SD=0·07) and for NMs was 0·50 (SD=0·18). Supplemental Figure 1 shows the distribution of NMR in the screened population(n=1733) and in theITT sample (n=1246)where SMs were oversampled.

Abstinence(Figure 2)

The NMR-by-treatment interaction was significant in the GEE model that included all time points (ORR)=1·96; CI=(1·11, 3·46); p=0·02). An effect for time point (OR=0·64; CI=(0·52, 0·80); p<0·001) and a time point-by-treatment interaction (ORR=0·58; CI=(0·42,0·79); p=0·001) indicated that varenicline's efficacy decreased significantly over time (compared to nicotine patch). We also conducted a validation analysis in which we re-ran the analyses in four different ways: Consider all dropouts to be smoking (analysis by GEE); including only cases who completed EOT, 6-, and 12-month follow-up (GEE); considering dropouts at any time-point as missing (GEE); and considering dropouts as missing data (analysis by mixed model, valid under missing-at-random dropout). The results were similar under all four models, with ratios of ORs ranging from 2·00 to 2·07, and p values ranging from 0·01 to 0·04.

As predicted, at EOT, varenicline was more efficacious than nicotine patch in NMs (OR=2·17; CI=(1·38, 3·42), p=0·001), but not in SMs (OR=1·13; CI=(0·74, 1·71), p=0·56), yielding a significant NMR-by-treatment interaction (ORR=1·89; CI=(1·02, 3·45); p=0·04). In a model including the placebo group as a reference, the interactioneffect was similar (p=0·05). Of the covariates, only nicotine dependence score predicted quitting significantly (OR=0·83; CI=(0·76, 0·90); p<·001).Likewise, at 6-months, varenicline was more efficacious than nicotine patch in NMs (OR=2·17; CI=(1·4, 3·4), p=0·001), but not in SMs (OR=1·12; CI=(0·74, 1·71), p=0·57); theNMR-by-treatment interaction was statistically significant (ORR=2·07; CI=(1·01, 4·22); p=0·05). Of the covariates, only nicotine dependence score predicted quitting significantly (OR=0·83; CI=(0·76, 0·91); p<·001).The interaction effect at 12-months was not significant (ORR=1·78; CI=(0·83, 3·80); p=0·14).The quit rates are as follows for EOT ([SMs]: placebo 17·2%, nicotine patch 27·7%, and varenicline 30·4% versus [NMs]: placebo 18·6%, nicotine patch 22·5%, varenicline 38·5%), 6-months ([SMs]: placebo 14·4%, nicotine patch 21·6%, and varenicline 19·1% versus [NMs]: placebo 12·9%, nicotine patch13·6%, and varenicline 22·0%), and 12-months ([SMs] placebo 13·4%, nicotine patch 19·4%, and varenicline 14·1% versus [NMs]: placebo 10·9%, nicotine patch 13·1%, and varenicline 16·0%).

The number needed to treat (NNT) was calculated using conventional methods,²⁸contrastingnicotine patch vs. placebo, and varenicline vs. placebo, at EOT. Among NMs, nicotine patch yielded an NNT of 26·0 (CI=19·7,32·3) while varenicline yielded an NNT of 4·9 (CI=4·7,5·1). For SMs, the NNTswere 10·3 (CI=9·4,11·2) and 8·1 (CI=7·5,8·7), respectively.

The ROC analyses revealed that the area was 0·51 (95% CI=(0·43, 0·58)) for placebo, 0·54 (0·47, 0·60) for nicotine patch, and 0·54 (0·48, 0·60) for varenicline. Although effects were in the predicted direction, comparisons of ROC areas between treatment arms were not significant.

Subjective Measures

For varenicline (vs. placebo), a significant NMR-by-treatment interaction was observed in side-effects (summary from the self-report checklist; β=−1·06; CI=(−2·08,−0·03); p=0·044). This reflected greatersummary side-effects reported on varenicline (vs. placebo) for SMs (β= 0·61; CI=(−0·10,1·32); p=0·09),but notfor NMs (β=-0·44; CI=(−1·19,0·30); p=0·24). Descriptive (post-hoc) item-level analysis indicated that, in SMs, varenicline led to significant increases in nausea (χ²=18·7, p=0·0003) and abnormal dreams (χ²=13·0, p=0·005); in NMs, varenicline led to significant increases in nausea (χ²=15·7, p=0·01), but decreases in irritability (χ²=15·4, p=0·001), anxiety (χ²=11·2, p=0·01), and attentional disturbance (χ²=11·3, p=0·01). For side-effects on nicotine patch (vs. placebo), the NMR-by-treatment interaction was not significant (β=−0·17; CI=(−1·21, 0·86); p=0·74).

SAEs, defined as any adverse event without regard to causality that resulted in death, was life-threatening, required hospitalization, or resulted in disability/incapacity, were determined by site physicians. There were 16 (3·9%), 22 (5·3%), and 11 (2·6%) SAEs on placebo, nicotine patch, and varenicline, respectively.Treatment arm effects or NMR-by-treatment interactions on SAE counts were not significant (see Supplemental Table 1 for side-effect counts by treatment and NMR group and Table 2 for summary side-effects values). There were no NMR-by-treatment interactions forwithdrawal symptoms or medication adherence (Ps >0·10).On average, 62% of participants used ≥80% of the pill dose recommended and 63% used ≥ 80% of the patches recommended, comparable to previous reports.^29,23

Discussion

In this biomarker-stratified randomized clinical trial, varenicline was superior to nicotine patch for NMs, but had equivalentefficacy for SMs. SMs, but not NMs, reported more overall-sideeffectson varenicline (vs. placebo). Although SMs were oversampledto be roughly 50% of the ITT sample, 40% of smokers attending the study intake wereclassified as SMsby the 0·31 cut point (illustrated in Supplemental Figure 1A). Thus, matching treatment choice based on the NMR could provide a viable clinical strategy for optimizing quit rates for all smokers, while minimizingside-effects for SMs.

As expected, the significant pharmacologic effect by NMR decreased after treatment was discontinued. A large decrease in quit rates on varenicline was observed over time, consistent with prior reports.¹⁷ Due to therelapsing nature of tobacco dependence, the current paradigm of short-term treatment has been challenged by clinical trials of extended duration therapy.²³Aplacebo-controlled trial comparing6-monthsof extended duration nicotine patch vs.the standard 8-weeks duration ofnicotine patch showed that SMs achieve significant benefit from extended therapy with 6-month quit rates of ~50%.³⁰ An important question is whether NMs would benefit from extended varenicline therapy.

An improved understanding of the mechanisms underlying NMR associations with treatment response could help to refine the use of this biomarker in clinical practice. The higher quit rates with nicotine patch among SMsvs.NMs are not due to differences in plasma nicotine levels,¹³nor are these differences due to nicotine withdrawal as shown here. Moreover, the NMR is not associated with nicotine dependence in most studies,¹⁵ and controlling for baseline cigarettes per day or dependence did not alter our findings.

Other neuropharmacological mechanisms can be considered. Because NMs smoke more cigarettes per day than SMs,³¹ conditioned smoking responses may be stronger in this group. NMs have enhanced responses in the brain's dopamine reward circuitry when viewing smoking cues, compared to SMs.³²Further, NMs have greater daily fluctuation in blood (and presumably brain) nicotine concentrations than SMs, which could contribute togreaterreward from smoking.³³ This could explain why, for NMs, both varenicline and bupropion are more efficacious than nicotine patch. Although bupropion and varenicline have differing mechanisms of action, both are non-NRT medications that increase dopamine levels in brain reward pathways.

The present study had both strengths and limitations. It is the largest pharmacogenetic study of tobacco dependence treatment and the first to use prospective stratification. The rate of loss to follow-up did not differ by treatment arm or NMR group. The mixed ethnicity of participants and the absence of gender or race interactions with NMR and treatmentsuggest that the NMR works well in both Caucasian and African American smokers; however, a limitation is that few Hispanics or Asians were included. As in most smoking cessation trials, we excluded individuals with major psychiatric and medical comorbidities, limiting the generalizability of our findings. Lastly, the quit rates are lower than some previous studies.^{17, 23} This may be due to the high unemployment rate in our sample (37%). Tobacco use is associated with unemployment, and was reported to increase during the recent economic decline,³⁴coinciding with our study.

With respect to the clinical implications, it is important to recognize that varenicline may be superior to the nicotine patch for smokers overall and that the side-effects of varenicline are generally mild and tolerable. Thus, varenicline could be prescribed to all smokers without consideration of NMR, especially in countries such as the United Kingdom, which do not caution use for certain subgroups of smokers. However, varenicline does have a “black-box warning” in the United States. Further, our data indicate that SMs do not achieve greater benefit from varenicline relative to nicotine patch, and are vulnerable to more side effects than NMs.

Thus, in addition to the imperative of increasing the utilization of treatments for nicotine dependence,³⁵ our data suggestthat treating NMswith varenicline, and SMs with nicotine patch could provide a viable clinical approach. Extending the duration of use of these treatments could potentially sustain the benefits of tailoredtreatment.³⁰ The NMR is practical for clinical use because it is unaffected by the time of day of sampling, is stable at room temperature, is not dependent on time since last cigarette among ad libitum smokers, and is stable within a smoker over time.^{11, 20, 36} While the assay results were reported in under one week at $50/sample, it is conceivable that a point-of-care test could be developed and implemented in clinical practice. However, using the NMR to individualize treatment may include additional costs and time. Our findings also underscore the notion that tobacco dependence is aheterogeneous condition and pharmacotherapies are not equally effective for all smokers.

Systematic Review

In our search of PUBMED for the term “nicotine metabolite ratio” and “smoking cessation” (or “tobacco/nicotine dependence treatment”), we located five retrospective analyses of data collected from randomized trials testing associations of 3’hydroxycotinine/cotinine values with response to smoking cessation treatments. None of these studies prospectively stratified randomization to alternative treatments, nor did any include the newer medication varenicline. All five papers are cited within the manuscript.

Interpretation

Our findings showed that varenicline was superior to nicotine patch for normal metabolizers of nicotine, but had equivalent efficacy for slow metabolizers. Slow, but not normal metabolizers, reported more overall side-effects on varenicline (vs. placebo). These results support the potential clinical validity of the NMR as a biomarker to guide the choice of therapy for individual smokers. To optimize quit rates for all smokers while minimizing side-effects, our data support treating normal metabolizers (60% of smokers in the population) with varenicline, and slow metabolizers with nicotine patch.