|Home | About | Journals | Submit | Contact Us | Français|
Initial screening of medications for efficacy in treating drug dependence may be accomplished more efficiently by using novel approaches that combine the practical advantages of within-subject laboratory studies with the clinical validity provided by clinical trials. A priori selection of functional gene variants associated with the pharmacokinetic or pharmacodynamic effects of a medication may aid this effort by controlling for individual variability as to clinical response or adverse effects; however, there are limitations to this approach, and these should be carefully considered.
The development of a new medication is a long and expensive process, with a new compound generally costing $2 billion and taking 14 years to move from drug discovery through human testing to US Food and Drug Administration approval.1 Costs escalate with each successive phase of clinical evaluation. At least one-third of the drugs in phase III, the final and most expensive clinical testing phase, fail to gain Food and Drug Administration approval and therefore do not become commercial products, representing a major waste of time and resources2 and needlessly exposing patients in clinical trials to potentially harmful compounds. More efficient use of drug development resources requires improvement of procedures to determine the likely efficacy of drugs in the patient population of interest. The earlier in drug development that ineffective compounds are identified, the smaller will be the waste in resources. An initial screening procedure (e.g., early phase II) that could identify clinically ineffective compounds with near-perfect accuracy prior to late phase II and phase III trials could save up to $700 million per compound and shorten its development duration by 5 years.1
Traditional randomized clinical trials, although valid, are often impractical for carrying out initial tests of the efficacy of medications aimed at treating drug dependence. This is due to the costs involved in recruiting sufficiently large numbers of subjects to enable detection of differences in the standard dichotomous outcome for such trials (i.e., abstinent vs. relapsed) between the active and placebo arms.3,4 In this paper, we first discuss potential new approaches to more efficient screening of novel medications to treat drug dependence, with the focus on nicotine dependence, and then consider how these procedures can incorporate genetic information.
In general, medications to treat addiction act through one or both of two mechanisms: relieving adverse symptoms of drug abstinence, such as withdrawal and craving, and reducing the direct reinforcing effects of acute drug use. As part of early screening of medications to treat drug addiction, a variety of human behavioral pharmacology paradigms have been used to test medication effects on acute responses related to these two mechanisms.3,4 Such acute responses may be viewed as intermediate markers of efficacy.5,6 Most models evaluate the efficacy of an addiction medication in terms of a patient’s self-reported levels of craving and withdrawal,7,8 as well as the observed affective and cognitive symptoms during acute withdrawal.9 A few studies have focused on the second mechanism—the blunting of the addictive drug’s reinforcing effect—by assessing the effects of the medication on acute drug self-administration behavior,10 drug-lapse behavior,11 and drug-, cue-, and stress-induced reinstatement of drug use.12 Neuroimaging-based markers of medication effects also have potential for aiding in the screening of addiction medications.13,14 Importantly, these studies usually recruit subjects who are smokers and are not currently interested in quitting smoking.
Studies that use intermediate markers of medication efficacy tend to cost less because they are shorter in duration and allow for smaller sample sizes. This is possible because of the greater statistical power afforded by within-subject comparisons and use of continuous outcome measures. However, although this is a very practical approach, effects of medications on most of these intermediate markers have not yet been shown to yield accurate predictions of the medications’ efficacy for drug abstinence in phase III clinical trials, the purpose of initial screening for efficacy. 3,10
We are conducting programmatic research to develop and validate a new type of procedure for initial efficacy screening of medications for treating nicotine and other addictions.15,16 This procedure seeks to combine the clinical validity of a clinical trial with the practical advantages of brief laboratory-based studies. In contrast to the acute intermediate markers used in human behavioral pharmacology studies (e.g., craving and amount of drug self-administration), our main dependent measure is daily smoking abstinence, somewhat similar to the measure in clinical trials. However, the number of days of abstinence provides a continuous outcome that enhances statistical power over comparisons of a dichotomous outcome at follow-up in clinical trials.17 On the other hand, we employ a within-subject crossover design, which is typical of laboratory studies but not of clinical trials, to compare days of abstinence during use of active medication vs. placebo. In short, subjects’ smoking behavior (as assessed by expired-air carbon monoxide) and related responses (e.g., craving, withdrawal) are assessed intermittently during 1 week of ad libitum smoking (baseline), 1 week of medication run-up (where necessary), and daily during 1 week of a simulated quit attempt while on the full medication regimen. Subjects then resume smoking, and this 3-week sequence is repeated for the other medication condition(s), administered in a double-blinded manner and in counterbalanced order. The difference in days of abstinence during the weeklong simulated quit attempt of each medication condition is the comparison of primary interest. Our assumption is that a medication that increases days of abstinence in a simulated quit attempt will aid long-term abstinence.
Our first studies assessed the sensitivity of the procedure in detecting efficacy. This was done by testing medications already approved for smoking cessation by the US Food and Drug Administration, specifically the nicotine patch15 and the α4β2 nicotinic receptor partial agonist varenicline.16 However, our ultimate goal was to devise a screening procedure to detect the efficacy of novel drugs. In both studies, we found that smokers high in intrinsic quit motivation (i.e., those planning to make a permanent quit attempt soon, but not immediately) provided a sensitive sample for detecting medication efficacy; however, smokers high in extrinsic quit motivation (those who received payment to abstain each day) did not. Our approach may be applicable to the screening of medications to treat other types of drug dependence, but the generalizability of the procedure has not yet been directly tested. It may be less applicable to neuropsychiatric conditions that require longer duration of treatment to observe improvement.
Each of these studies identified medication efficacy for smoking abstinence with a sample of fewer than 60 subjects who were high in intrinsic quit motivation—far fewer than would be needed in a comparable clinical trial. The enhanced power of our procedure is due particularly to the within-subject comparison of medication effects,17 in contrast to the between-subjects comparisons in clinical trials.18 An estimate of the sample size required for a randomized between-subjects trial with the same power as a within-subjects crossover study is provided by the formula, NRandomized/NWithin = 2/(1 − ρ), where ρ is the within-subject correlation of responses to medication vs. placebo from the within-subjects study.17 If ρ is zero (i.e., abstinence responses to each condition are unrelated), then a between-subjects trial simply requires twice the sample size of a crossover study. As ρ increases, however, the sample size of a between-subjects trial with power comparable to the crossover study must increase. In our study on varenicline, ρ was 0.6. Therefore, in order to match the power of a crossover, within-subjects study with 60 subjects, a randomized between-subjects trial would need five times as many, that is, 300 subjects (i.e., 2/(1 − 0.6))!
An alternative paradigm to test intermediate indexes of medication efficacy is illustrated by our recent study9 using a within-subject crossover design to evaluate the effects of short-term (13 days) treatment with varenicline vs. placebo. Intermediate markers of medication efficacy in relieving abstinence symptoms were assessed after 72 hours of smoking abstinence while on medication or placebo. Subsequently, a programmed lapse in abstinence (restarting of smoking) in the laboratory was used to assess the effects of the medication on acute smoking reward and on the subsequent ability to remain abstinent from smoking for a 7-day period after the lapse. Consistent with its known clinical efficacy in promoting smoking cessation, varenicline reduced self-reported withdrawal, craving, and negative affect during abstinence and improved sustained attention and working memory, as compared with placebo. Varenicline also reduced smoking reward during the programmed lapse and increased the duration of abstinence, although the latter affect was dependent on treatment order (i.e., whether the medication phase of the study preceded the placebo phase). An independent study adapted this paradigm, incorporated functional magnetic resonance imaging during a working memory task, and documented varenicline’s effects on cognitive performance and brain activity.19 Further validation is necessary to determine whether such intermediate marker approaches are robustly predictive of clinical efficacy or whether they are better suited to examining mechanisms of action of candidate medications.
Because these alternative approaches have not yet been shown to predict the efficacy of novel medications in large clinical trials, smaller (early phase II) clinical trials may continue to be necessary for initial screening of medication efficacy. However, there may be ways to improve the efficiency of these early trials, such as by using adaptive designs to identify optimum parameters for testing a medication’s effects, including the most appropriate dosage regimen or the optimal duration of therapy.20,21 In adaptive designs, data accrued early in a trial are used to modify the trial procedures for subsequently recruited subjects, so as to produce the most sensitive possible test of the medication. Another example is the use of alternatives to placebo-controlled designs when the interest is in comparing the effects of a new compound with those of an existing drug. These alternative strategies can reduce the number of treatment arms from three to two and include superiority (i.e., better than existing drug), equivalence (i.e., similar to existing drug), or noninferiority (i.e., not significantly worse than existing drug) designs.22 However, these designs, particularly the superiority design, may actually require more subjects per treatment arm than placebo-controlled trials, thereby limiting their practicality.
Incorporating genetic information into early human screening studies for medication efficacy may further enhance the sensitivity and predictive validity of such studies. Individual differences in drug response can reduce the efficacy signal of a medication in early development, and incorporating pharmacogenetic information into trials may help control this variability in response to the medication. While there are many potential applications of pharmacogenetics to the field of drug discovery, including the use of genomic data to identify novel therapeutic targets,23 we focus here on the potential utility and limitations of incorporating genetic information into medication screening studies early in clinical testing.
One such strategy is to utilize genotyping during subject recruitment to acquire a more homogeneous sample (thereby reducing variability in clinical outcomes and increasing statistical power) or to overweight the sample with subjects who are more likely to exhibit a particular clinical response. Alternatively, sample stratification based on genetic information may be used to ensure balance in clinical outcomes between alternative treatment conditions. Uhl and colleagues24,25 proposed that a panel of single-nucleotide polymorphisms in multiple genes can be used to assign smokers predisposed to high or low quit success in equal proportions to the different treatment arms of a smoking cessation pharmacotherapy trial. Their simulation data suggest that such an approach can decrease variance and increase the magnitude of the medication response, with associated cost reductions. The clinical utility of this approach awaits empirical confirmation.
Along with a priori genetic testing in the example above, post hoc genotyping may have utility for improving the efficacy signal for medications in development.23 In post hoc analyses of the results from our earlier crossover study comparing the effects of nicotine patch vs. placebo patch, we found that medication effects, in terms of a greater number of days of abstinence, were predicted by genetic variability in the nicotinic acetylcholine receptor β2 subunit (CHRNB2).26 This finding was consistent with the results from a larger pharmacogenetic clinical trial that identified this SNP within a systems-based analysis of smoking cessation.27 Notably, the effects of gene–medication interaction on the efficacy of the medication were found to be significant even though the sample comprised only 156 subjects. This demonstrates the increased statistical power of our approach because of the within-subject, rather than between-subject comparisons of medication effects.18 At the same time, such post hoc findings should be considered hypothesis generating, with the potential to inform a priori selection of candidate genes for larger-scale pharmacogenetic trials.
Finally, an important application of genetics in the field of medication development is for the identification of subjects who are likely to experience adverse drug reactions.28 However, this approach may not be well suited to the laboratory paradigms described here because longer duration of medication use and very large sample sizes are required in order to detect rare adverse events, which tend to be associated with alleles that occur only infrequently in the population. In such an eventuality, access must be available to large databases from multiple studies, which can be mined for adverse drug reactions, although differences across studies with respect to medication dosing regimens may make this challenging.
There are limitations to the approaches discussed here—most importantly, the potential for false-positive results when post hoc genotyping is employed.29 Per the norms for research studies, a priori rather than post hoc selection of candidate genes is likely to provide more reliable findings in general. It is important to select functional variants that have robust prior associations with clinical outcome and clear biological significance to a medication’s pharmacokinetic and pharmacodynamic effects. Likewise, for medication-screening paradigms employing intermediate markers of clinical outcomes, it is critical that the effects of the medication on such markers predict the medication’s efficacy for the ultimate clinical outcome measure (i.e., drug abstinence).3,4,6,10
Some of the issues discussed here are also pertinent to the development of medications for other neuropsychiatric conditions, such as disorders of mood and cognition. Indeed, paradigms to assess the cognition-enhancing effects of novel medications, combined with a priori genetic testing for variations in pharmacodynamic targets, have yielded promising results.30 With the expanded use of neuroimaging in the development of psychiatric drugs,31 the role of genetic variations and the related caveats discussed in this paper will become increasingly important.
Preparation of this article was supported by National Institutes of Health grants U01 DA020830 and P50CA 143187.
Conflict of Interest
K.A.P. has served as a consultant for Cypress Bioscience. C.L. has served as a consultant for and received research funding from AstraZeneca, GlaxoSmithKline, and Pfizer.