Improving efficiency of initial tests for efficacy in smoking cessation drug discovery

Abstract

Introduction: One obstacle to rapid development of new smoking cessation medications is the inefficient early clinical evaluation of the efficacy of novel drugs, which inform us as to whether or not to proceed with the greater expense and time of more formal clinical trials. The vast majority of novel drugs fail to show efficacy for cessation only after substantial resources have been spent and, thus, are largely wasted.

Areas covered: The author reviews the general limitations in the current typical procedures for initial tests of cessation efficacy in novel drugs. Small, randomized clinical trials often have good validity but may have practical limitations in achieving adequate statistical power to test novel versus placebo treatment conditions. Lab tests of acute drug effects on abstinence symptoms, during brief enforced cessation periods, are practical but have limited clinical predictive validity.

Expert opinion: Initial efficacy testing may be more efficient if done using innovative crossover designs that evaluate brief ‘practice’ quit periods for both active and placebo treatments within the same smokers, recruiting those high in quit motivation. Because this approach would require far fewer subjects and a shorter duration of testing, results could be obtained more rapidly and inexpensively to indicate that a novel drug may, or may not, be sufficiently efficacious as to warrant the greater costs and time of formal randomized clinical trials.

Keywords::

• experimental design
• medication screening
• nicotine dependence
• smoking cessation

1. Introduction

Worldwide, nearly one billion adults now smoke tobacco [1], still the greatest preventable cause of mortality with about 5 million deaths per year, or 1 every 6 s, and expected to cause > 10 million annual deaths by the middle of this century [2]. In the US, most smokers state a desire to quit, but only a minority will then actually make an attempt to quit permanently [3]. Moreover, just 5% of those quit attempts, most of which are unaided, are successful 1 year later [4]. This success rate rises only to 20 – 30% with even the most effective cessation medication, varenicline [5]. Therefore, in addition to greater use of other tobacco control strategies, more effective treatments are needed if the public health toll from smoking is to be reduced.

In recent decades, dozens of drugs have been evaluated for initial evidence of efficacy in smoking cessation [6,7], or early Phase II studies using US FDA terminology. Among the novel or repurposed medications tested for efficacy in smoking cessation are many antidepressants (including bupropion, doxepin, fluoxetine, imipramine, moclobemide, nortriptyline, paroxetine, sertraline, selegeline, tryptophan and venlafaxine), several antianxiety drugs (such as buspirone, diazepam, meprobamate, metoprolol, oxprenolol), known nicotine agonists/antagonists (nicotine replacement, mecamylamine, lobeline, cytisine, varenicline, etc.), and other drugs (e.g., ondansetron, baclofen, clonidine, naltrexone, rimonabant), in addition to experimental compounds. Yet, as with development of drugs for other treatment indications [8], fewer than 10% of these have been FDA-approved to date for clinical use as first-line cessation treatments: nicotine replacement therapy (gum, patch, etc.), bupropion and varenicline.

Most of these clinical ‘failures’ to demonstrate efficacy essentially waste time and resources that ideally should be focused on developing more promising novel candidate drugs. A more efficient early screening procedure would be one that obtains a valid answer as to the likely clinical efficacy of a novel drug (versus a placebo or other comparison) using the smallest participant sample and/or shortest duration of testing possible. More efficient use of resources could enable ineffective drugs to be so identified as soon as possible and dropped from further testing [9]. The goal would be to avoid continuing on in drug development and thereby committing even larger resources and longer time in order to test the drug’s clinical efficacy in larger numbers of patients (e.g., late Phase II, III trials). Thus, early Phase II screening is designed to inform whether or not to proceed to the larger late Phase II assessments of the medication for cessation efficacy in randomized clinical trials.

2. Limitations to current early Phase II evaluation approaches

A common early Phase II study is a small (e.g., ‘pilot’) clinical trial of smokers trying to make a permanent quit attempt who are randomized to active drug versus placebo (or some other comparison) treatment conditions. This testing is often valid in predicting that drug’s efficacy in subsequent larger Phase III trials, but only when the sample size is sufficiently powered (i.e., about 100 per group, as in the first trial of bupropion vs placebo; [10];). This is usually not the case since most pilot trials have fewer than 50 per treatment condition group [7]. As a consequence of inadequate sample size, many small trials of promising novel drugs may fail to show efficacy, while some trials of ineffective drugs may show spuriously positive results due to idiosyncrasies in the small sample. Yet, even early Phase II trials that are adequately powered can still be time-consuming and expensive propositions for testing (by definition) novel compounds of unknown efficacy for cessation.

A second approach, which is quicker and less costly but also of limited validity, is to evaluate initial evidence of efficacy for cessation by assessing drug responses during brief (often < 1 day) tobacco abstinence in non-quitting smokers being paid to abstain temporarily. These smokers are far more common and easier to recruit in sufficient numbers [3]. Here, study measures generally assess surrogate markers of efficacy, mostly whether the drug blunts abstinence-induced symptoms, such as self-reported withdrawal or craving, or attenuates smoking reward, assessed by the delay or reduction in smoke intake during brief ad lib cigarette access [11]. Such studies often take advantage of the within-subject crossover design, thus repeating most session procedures twice, for active versus placebo conditions, ideally administered blind and in counter-balanced order. Subjects in studies that test relief of symptoms are instructed to abstain for some time to enable withdrawal and craving to emerge. This period of abstinence must be biochemically verified or otherwise closely monitored since continued smoking would confound assessment of craving and withdrawal relief due to novel drug or placebo. In studies testing a drug’s ability to curb smoking behavior, non-quitting subjects are typically administered the medication or placebo and assessed on amount of ad lib smoking or delay to begin smoking when provided an alternative reinforcer. Their subsequent smoking behavior and self-reported responses (e.g., ‘liking’ of the cigarettes) determine whether smoking reward is attenuated by active drug versus placebo [12,13].

However, as tested with this approach, acute drug effects on symptom relief or ad lib smoking behavior often are not strongly predictive of cessation outcome with the drug, the objective of early Phase II testing, for several reasons. First, clinical therapeutic effects of drug are minimized in smokers who are not interested in making a permanent quit attempt, the typical study population here, compared to those making a quit attempt [12]. Moreover, abstinence symptom relief in smokers who do try to quit is imperfectly related to subsequent quit success. Withdrawal is a reliable consequence of a quit attempt, but its severity may be a weak mediator of abstinence and poorly predictive of subsequent cessation success [13]. Related symptoms, notably negative affect, are similarly reliable effects of quitting and somewhat predictive of quit duration [14]. Yet, negative affect relief is relatively modest as an index of a drug’s efficacy, as illustrated by the fact that nearly all of the many antidepressants tested for cessation efficacy have not been successful. Abstinence-induced craving also predicts cessation success [15] but may require more than a full day of tobacco deprivation, and it is not necessarily attenuated by first-line cessation medications in short-term tests of non-quitting smokers [16], showing tenuous predictive validity with this approach. Cue-induced craving, or very brief increases in smoking urge due to viewing smoking-related stimuli (‘cues’), is more practical as an acute measure during sessions in a within-subjects design but has clearly failed to show clinical value in predicting success in a quit attempt and is also not influenced by first-line cessation medications [17]. Assessing acute abstinence symptoms may be useful for gauging potential mechanisms of a novel drug’s efficacy for cessation but should not be the primary outcome of interest (unless the drug is specifically designed to aid withdrawal relief and not necessarily smoking cessation per se). Alternatively, because withdrawal symptoms usually ameliorate within a few weeks after quitting, drugs directly aimed at withdrawal relief may aid initiation of abstinence but not longer term cessation since risk of relapse extends well beyond a few weeks post-quit [18].

In the tests of brief ad lib smoking behavior due to drug versus placebo, once again first-line medications have not shown sensitivity in attenuating such behavior [12]. As discussed elsewhere [12,13], these studies typically assess amount of ad lib smoking or the delay in onset to smoking during hour-long access to cigarettes after overnight abstinence in non-quitting smokers administered active or placebo drug, often in a within-subjects design. A longer delay before initiating smoking behavior usually is reinforced with a small monetary reward, and the assumption is that a potentially effective medication will lengthen this delay. Whether the duration of delay to smoking in a brief session relates to outcome of a quit smoking cessation attempt is unclear. Furthermore, reduction in smoking amount is not consistent with complete smoking abstinence, the treatment goal of cessation medications.

3. Potentially more efficient combined approach

A more efficient, yet valid, early test of a drug’s potential cessation efficacy may involve some optimal combination of features from these two general approaches, especially in regard to the choice of outcome measure, study participant characteristics and study design.

3.1 Primary outcome measure

Until an acute drug response that can robustly index long-term quit success is found, the best surrogate outcome measure of drug efficacy in aiding smoking cessation is likely to be complete abstinence for at least 24 h, which can be biochemically verified by expired-air carbon monoxide (CO < 5 ppm). Consequently, this abstinence needs to be a dependent measure of the drug response, rather than a study condition imposed for testing other drug responses such as withdrawal relief (e.g., where participant’s study payment is contingent on following instructions to temporarily abstain). Yet, the dichotomous nature of quit status during treatment and follow-up (i.e., quit/not quit) requires large subject samples for adequate statistical power to detect differences due to drug conditions [19], as noted. However, the continuous measure of number of quit days during initial treatment, statistically more powerful than a dichotomous measure, may also predict subsequent cessation outcome result. Clinical research shows that number of quit days during the first week of a formal trial predicts quit status at the end of 2-month treatment and 6-month follow-up [20], and quitting during the first 1 – 2 weeks predicts long-term cessation outcome [21,22]. This suggests number of quit days early in treatment may be a valid surrogate outcome measure for tests of a novel drug’s efficacy in aiding long-term cessation.

3.2 Participant characteristics

Because smokers with no intrinsic interest in trying to quit have no motivation to abstain for at least 24 h, number of days quit is not a sensitive measure of drug effects in such smokers and requires testing those who are at least seriously contemplating a permanent quit attempt. (To attract those with high quit interest for a study of short-term drug effects, a benefit of study participation could be free treatment to help make a permanent quit after study completion.)

3.3 Study design

Another key procedure for enhancing statistical power to compare drug effects with only a modest subject sample (i.e., greater efficiency) is to employ a within-subject, crossover design, which allows each participant to act as his or her own control, minimizing error variance [19,23]. This design cannot be used in a traditional clinical trial, since equating the initial baseline prior to each drug condition requires all participants to be non-abstinent, even when the first condition successfully aids abstinence. However, a crossover design could be used if each drug condition is administered for a short duration of attempting to quit (e.g., 1 week), separated by a ‘washout’ period involving resumption of ad lib smoking. Each crossover drug condition could be presented to participants as a brief ‘practice’ quit attempt in which they can learn how to cope with withdrawal symptoms and other difficulties with initiating and remaining quit, so that they feel better prepared to maintain continuous abstinence when they make a permanent quit attempt (after the study).

3.4 Initial evaluations of a crossover Phase II procedure

A crossover procedure consisting of these outcome measures, subject characteristics and study design has been developed and evaluated for early Phase II testing of cessation efficacy in novel drugs, employing week-long quit attempt periods during each drug condition with smokers intending to quit soon [13]. In terms of test sensitivity and specificity [24], the procedure has been shown to be sensitive to drug efficacy using all three first-line cessation medications as model drugs, versus placebo, as well as specific in detecting lack of efficacy in a drug known to be ineffective for smoking cessation (modafinil). Its greater efficiency is such that fewer than 50 smokers participating over about 6 weeks of testing (depending on the required dose run-up and washout periods) are likely sufficient to evaluate potential efficacy of a drug for aiding cessation in the within-subjects comparison of drug conditions.

The reason this procedure can be effective with a much smaller sample size is due to the typically high within-subject correlation of days quit in response to each drug condition (even placebo), so that error variance is controlled [23]. The formula to estimate the sample size required for a randomized between-subjects trial that has the same statistical power as a within-subjects crossover study is N_Randomized/N_Within = 2/(1−rho), in which rho designates the within-subject correlation of responses to medication versus placebo from the within-subjects study [23]. If the quit day responses to each treatment condition are completely unrelated (so rho is 0), then a between-subjects trial requires simply twice the sample size of a crossover study. However, as the correlation of quit day responses to the two conditions becomes stronger (increases in rho), the error variance is reduced in a within-subjects design. Consequently, the sample size that would be needed for a between-subjects trial of statistical power comparable to the crossover study design has to increase. In each of the studies using this crossover procedure [13], rho was ∼ 0.6 for days quit on placebo versus active (bupropion, varenicline). Thus, to equate the power of a crossover within-subjects study with 50 subjects, a randomized between-subjects trial would require 5 times as many, or 250 subjects (i.e., 250/50 = 2/[1-0.6]), or at least 120 smokers in each of the placebo and active drug conditions (similar to [10]).

3.5 Potential applicability of approach to preclinical procedures

Finally, enhancing the predictive validity of preclinical evaluations of potential efficacy in novel drugs, even earlier in the drug development process, is far more challenging than this early Phase II clinical testing. Yet, some elements of the above approach may be applicable to the drug development testing procedures used in animal studies [11]. For example, taking steps to minimize error variance by reducing heterogeneity of groups for statistical comparison has been recommended [25], analogous to our crossover design for brief clinical tests of drug conditions, in which participants act as their own controls. Also relevant are the notions that: i) tests of potential therapeutic drug effects on substance use should minimize forced abstinence in animals; and ii) amount of substance use behavior should be a primary-dependent measure [26]. These points are consistent with understanding the limitations in assessing symptom relief during required abstinence in non-quitting smokers, and with the use of daily abstinence as the primary outcome measure in clinical studies. Heightening an animal’s motivation to refrain from substance self-administration by making alternative reinforcers available [27] could be similar to testing efficacy of drugs for cessation ability in smokers with a high interest in quitting soon.

4. Conclusion

The high probability of negative results in early Phase II testing of novel drugs for evidence of efficacy in smoking cessation argues for a more efficient initial clinical evaluation in the drug discovery process. Such a procedure would improve the use of overall development resources by enabling ineffective drugs to be identified quickly and with less cost so they could be dropped from further clinical testing, which necessarily requires even larger numbers of patients and longer study durations. That way, most drug development efforts could continue to be focused on novel drugs with more promise of potential smoking cessation efficacy. A more efficient early Phase II screening procedure to inform whether or not to proceed to larger clinical assessments for confirming a medication’s efficacy is one that takes advantage of the practicality of a within-subject crossover design to compare active drug versus placebo on short-term quit ability in the same smokers. This design minimizes error variance, allowing for a much smaller sample size than in the traditional between-subjects designs of randomized trials, but it requires careful recruitment of smokers already preparing to quit soon, to provide a sample that is sensitive to the proposed mechanism of a drug’s efficacy for cessation.

5. Expert opinion

The two current primary approaches to early Phase II testing of novel drugs for evidence of efficacy in smoking cessation generally have opposite strengths and weaknesses that impact their validity and efficiency. The first, small clinical trials, has good predictive validity when adequately powered but can be somewhat costly and time-consuming for tests of whether a novel drug has sufficient promise as a cessation medication. This commitment of time and costs poses practical risks if the drug is found to be ineffective, which unfortunately is very likely given the prior track record of early Phase II testing of efficacy for smoking cessation in novel drugs, as noted. The second approach, lab assessments in non-quitting smokers of smoking behavior or reward, or acute symptom relief during brief instructed abstinence, is relatively inexpensive and quick to conduct but often lacks predictive validity. Notably, compared to the eventual intended clinical population of those preparing to make a quit attempt, these acute lab effects of drug are attenuated in the typical study population, paid smokers who are not interested in making a permanent quit attempt. Moreover, abstinence symptom relief from drug in smokers who are ready to quit is imperfectly related to subsequent quit success, further explaining the lack of clinical validity of this approach and arguing for a focus on actual quit behavior as the main outcome measure.

In the author’s opinion, combining the predictive validity of clinical trials with the practical advantages of brief lab tests of drug effects, as suggested here, can likely lead to a more rapid and less costly answer as to whether or not more resources should be devoted to a drug’s further development as a potential new smoking cessation medication. In subjects with a high interest in permanently quitting after the study, a drug that does not increase ability to quit smoking compared to placebo during separate short-term ‘practice’ quit attempts does not warrant further efficacy evaluations, while drugs that do increase ability to quit may meet a threshold for evidence of efficacy to justify additional study. Such a crossover procedure may not be practical for some novel compounds such as those requiring lengthy dose run-up or washout periods (e.g., vaccines). However, the huge and continuing public health burden caused by persistent tobacco dependence demands accelerated clinical testing of promising new compounds arising from basic research observations, or repurposed existing medications, to foster their development as effective pharmacotherapies for smoking cessation. Rapid determination of those lacking efficacy can help focus limited resources on the remaining compounds that do have evidence of efficacy for aiding cessation.

Finally, because this crossover procedure is designed to evaluate relative treatment conditions, it may be applicable to other testing in the drug development process. Thus, comparing different doses of a single novel drug, not just active versus placebo, may aid identification of the optimum dosing regimen to use in further clinical testing. Similarly, comparing different active drugs could be useful to gauge their relative efficacy for smoking cessation. Moreover, this type of crossover procedure could be adapted to test initial efficacy in novel drugs to treat other problems of drug abuse, since cessation of persistent drug use is the objective in those tests as well.

Declaration of interest

K Perkins is employed by the University of Pittsburgh and has served as a consultant for Embera Neurotherapeutics that is unrelated to the current research. This paper was supported by NIH grant UH3 TR000958 from the National Center for Advancing Translational Sciences. He has no other relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript apart from those disclosed.