Abuse and Therapeutic Use of Benzodiazepines and Benzodiazepine-Like Drugs

James H. Woods, Jonathan L. Katz, and Gail Winger

Since their introduction over 30 years ago, benzodiazepines have largely replaced older sedative–hypnotic agents in most countries. Because these drugs are used primarily for their therapeutic effects, abuse and misuse of benzodiazepines are best conceptualized in the context of their appropriate use. This chapter will therefore review pharmacological, clinical, and epidemiological studies of both appropriate use of benzodiazepines and their liability for abuse.

Research has of course continued in the effort to develop new anxiolytics with lesser sedative effects or liability for abuse. The most fruitful of these efforts have continued to focus on the benzodiazepine receptor system. In the context of what is known about established benzodiazepines, this chapter will consider some newer sedative/anxiolytic drugs (i.e., zopiclone, zolpidem, abecarnil, and bretazenil) that act on the benzodiazepine receptor.

There have been several recent reviews of the pharmacology of benzodiazepines. For a broader discussion of the basic pharmacological mechanisms of action of these drugs, the interested reader is referred to reviews by Haefely (e.g., see refs. 29, 30, 31). Hollister et al. (35) have recently published a comprehensive review of therapeutic uses of benzodiazepines, to which this chapter refers. With respect to the abuse liability of benzodiazepines, this chapter draws extensively from our previous reviews (76, 77).

Because of editorial limitations on the number of references that can be cited in this chapter, we refer to literature considered in our earlier reviews (76, 77) or in that of Hollister et al. (35) as follows: Rather than citing individual studies, we generally cite review articles, including numbers of the pages on which the relevant studies were discussed. We regret that we cannot cite all individual studies.

The first marketed benzodiazepines, chlordiazepoxide and diazepam, have a 1,4-diazepine ring that contains a 5-aryl Substituent Ring and is fused to a benzene ring (Fig. 1). This has become known as the classic 1,4-benzodiazepine structure. A number of modifications of this structure have produced compounds with similar binding characteristics and, typically, similar spectra of action. Midazolam, the shortest-acting of the benzodiazepines, has an imidazo ring fused to the diazepine ring. The benzodiazepine antagonist flumazenil also possesses an imidazo ring in this position. The triazolobenzodiazepines, including alprazolam and triazolam, are more recently developed anxiolytic and sedative compounds, which have a triazolo ring fused to the diazepine ring.

A number of newer anxiolytic or sedative compounds act at the benzodiazepine site or at a subset of this site, but they do not share the classic benzodiazepine chemical structure. One of the first such compounds to be developed was zopiclone, a cyclopyrrolone. The imidazopyridine class has also yielded sedative compounds, of which zolpidem is an example. Also among the newer sedatives is abecarnil, which has a beta-carboline structure. Another example of the newer sedative compounds to be considered here is bretazenil, which is a tetracyclic 1,4-benzodiazepine.

Benzodiazepine agonists and other agonist ligands at the benzodiazepine site achieve their therapeutic effects by enhancing the actions of the inhibitory neurotransmitter gamma-aminobutyric acid (GABA) at its receptor. The benzodiazepines have a binding site on the GABA receptor, which forms a channel through the membrane and opens and closes to control chloride flow into the cell. When benzodiazepine agonists are on their receptor site, GABA produces a more rapid pulsatile opening of the channel, and the flow of chloride is increased. The central GABA receptor, known as the GABAA receptor, consists of at least four subunits; three of these—alpha, beta, and gamma—each contains three to six variants. The multiplicity of variants suggests that there are a number of different GABA receptors, but the subunit makeup of the native receptor has not yet been determined.

Two GABA receptors have been identified anatomically and pharmacologically. These receptors—variably called type I and type II, benzodiazepine I and benzodiazepine II, or omega I and omega II—are located throughout much of the central nervous system (CNS). The omega I site has been associated with the alpha-1 subunit, whereas the omega II site appears to be heterogeneous, located on receptors with alpha-2, alpha-3, and alpha-5 subunits (75). The ratio of omega I to omega II binding sites is greater in the cerebellar and cerebral cortices, whereas omega II sites predominate in the spinal cord. Pharmacological studies indicate that the 1,4-benzodiazepines bind with relative nonselectivity to both omega I and omega II sites. The triazolobenzodiazepines tend to have a greater affinity for omega I and II receptors than do the other benzodiazepines, and they are more potent. Zopiclone, despite its unusual chemical structure, has a binding profile much like that of the classic benzodiazepines. Zolpidem, however, binds with much greater affinity to the omega I site (e.g., see ref. 49), and there is evidence that abecarnil may have some specificity for the omega I site as well (69).

Most of the benzodiazepines currently available for therapeutic use are considered to be full agonists at the benzodiazepine site; a full benzodiazepine agonist is a drug that produces the maximum effect in all biological assays, although it occupies less than the maximum number of benzodiazepine receptors. Recently, partial agonists at the benzodiazepine site have been identified; a partial benzodiazepine agonist is a drug that produces less effect than a full agonist when it occupies the same number of benzodiazepine receptors. Benzodiazepine antagonists are drugs with affinity for the benzodiazepine modulatory site but no efficacy. A unique aspect of the benzodiazepine-related modulatory site on the GABAA receptor is that it is bidirectional; there are agents that bind here that decrease the effects of GABA at its site on this receptor, and thus have effects opposite to those of classical benzodiazepines.

There is of course great interest in the possibility of developing compounds that offer the therapeutic effects of the classic benzodiazepines, but with less risk of the adverse effects associated with these drugs. Identification of site-selective compounds, such as zolpidem, is one strategy that has been tried in this pursuit. Another is to develop compounds with a partial agonist profile, either at both omega I and omega II receptors (e.g., bretazenil) or selectively at the omega I receptor (e.g., abecarnil). As discussed in this chapter, agents with either site-selective or partial agonist activity appear to have behavioral profiles that differ in interesting ways from those of the classic benzodiazepines (see also GABA and Glycine and Genetic Strategies in Preclinical Substance Abuse Research).

Benzodiazepines have been found useful in a remarkably wide and varied array of clinical applications. Most traditional clinical use has been based on their anxiolytic, hypnotic, anticonvulsant, and antispastic effects. Other, possibly related effects demonstrated in clinical trials and practice include antipanic, antidepressant, amnestic, and anesthetic effects.

In view of the diversity of these effects, it is remarkable that, as demonstrated by numerous clinical comparisons, benzodiazepines are distinguished much more by their similarities than by their differences. This supports the hypothesis that the activity of agonists at the benzodiazepine receptor is in fact the dominant mechanism underlying the clinical effects of these drugs.

The most important differences with respect to clinical use pertain to relative potency and to onset and duration of action. When equipotent doses of the various agents can be used, however, they tend to exhibit similar effects. For example, early studies had found that the older benzodiazepines were generally ineffective in treatment of panic disorder. Thus when the newer agent, alprazolam, showed antipanic efficacy, this was speculatively attributed to the compound's novel triazolo ring. However, alprazolam is also relatively more potent than the older agents; more recent studies have found that, when used at higher dosages, older benzodiazepines are effective against panic disorder as well (35, pp. 16S–17S).

With respect to clinical utility, differences pertaining to onset and offset of action are probably most important. In some uses, including various medical and psychiatric emergencies, rapid onset of action is of course a necessity. Short duration of action may be useful (e.g., for outpatient surgical and diagnostic procedures), although longer duration of action may be desired (e.g., in treatment of sleep-maintenance disturbances or for seizure control). Differences of onset and offset of action also have important implications with respect to adverse effects as discussed below.

The following brief description of therapeutic effects of benzodiazepines summarizes the conclusions of the recent review by Hollister et al. (35), which is certainly the most comprehensive review of the clinical uses of these drugs. A large number of placebo-controlled studies have demonstrated the efficacy of benzodiazepines in treatment of anxiety disorders, including generalized anxiety disorder (GAD) and panic disorder; benzodiazepines are also useful in facilitating the behavioral treatment of phobias (35, pp. 2S–23S). They are also effective in management of anxiety and other symptoms of psychological distress associated with various medical disorders (35, pp. 64S–72S). As shown in sleep laboratory and controlled clinical trials, benzodiazepines are effective in treatment of disturbances of falling asleep and of maintaining sleep (35, pp. 23S–51S). Alprazolam has been shown effective in treatment of major depressive disorder of mild or moderate severity. Although they do not appear to have specific antimanic activity, clonazepam and lorazepam provide rapid control of manic episodes (35, pp. 23S–51S). Used alone or in combination with neuroleptics, benzodiazepines have proved valuable for management of various psychiatric emergencies involving agitation or hostility. They also provide acute relief of catatonic symptoms, and they are sometimes useful "neuroleptic-sparing" adjuncts in treatment of schizophrenia (35, pp. 72S–81S).

Objective measures demonstrate the rapid and dramatic resolution of symptoms of many convulsive and spastic disorders with administration of benzodiazepines. Intravenous diazepam is frequently life-saving in various convulsive emergencies, such as status epilepticus or tetanus spasms. Benzodiazepines afford prompt control of neonatal seizures resulting from perinatal hypoxia, and they provide effective prophylaxis in children susceptible to prolonged seizure activity. Chronic administration of benzodiazepines as adjuncts to other anticonvulsants can significantly improve control of many cases of epilepsy (35, pp. 81S–94S). Benzodiazepines frequently bring substantial relief of spasticity associated with chronic conditions, such as multiple sclerosis and paraplegia resulting from spinal trauma. They are also of benefit in some patients with cerebral palsy, particularly younger patients with athetoid movements (35, pp. 100S–107S). The drugs are often effective in treatment of involuntary movement disorders, including restless legs syndrome, choreas, intention myoclonus, and some dyskinesias and dystonias associated with use of neuroleptic medications (35, pp. 94S–100S). It has long been recognized that benzodiazepines are effective in managing acute withdrawal from alcohol; they reduce the incidence of potentially fatal complications, such as alcoholic delirium, seizures, and hyperpyrexia (35, pp. 107S–113S).

Benzodiazepines are important adjuncts in medical and dental procedures. When administered prior to surgical anesthesia, they reduce anxiety, provide sedation, facilitate anesthetic induction, and produce amnesia for the events surrounding induction; they also often reduce the required doses of anesthetic agents. They provide safe and effective sedation for mechanical ventilation following cardiac surgery. Lorazepam and other benzodiazepines can help to control nausea and vomiting associated with cancer chemotherapy. Benzodiazepines are commonly used for their anxiolytic, sedative, and amnestic effects in a wide range of stressful diagnostic procedures (35, pp. 118S–135S).

In many of their uses, benzodiazepines are required for only short periods of treatment; as discussed below, the vast majority of actual use of these drugs lasts for only a few weeks. However, some disorders for which benzodiazepines are indicated are recurrent or chronic. Although the effects of long-term benzodiazepine treatment have not been systematically studied, the results of a number of studies provide some interesting suggestions about long-term effects. Therapeutic effects of benzodiazepines are usually or often sustained over months or years, without the need for increased dosage, in treatment of GAD and panic disorder (35, pp. 19S–21S). Relief of insomnia has been documented during periods of regular nightly use for up to 6 months (35, p. 138S). Although benzodiazepines are not indicated as sole therapy for chronic convulsive disorders, such as epilepsies, because they are known to lose effect fairly rapidly in a large proportion of cases, it is of interest that tolerance is not seen to develop in all cases; in some patients, the drugs have continued to control seizures for years (35, pp. 85S–88S). Similar evidence of interindividual variation in tolerance development has been described in benzodiazepine treatment of spastic and dyskinetic disorders (35, pp. 95S–99S). This evidence that tolerance does not develop to these effects in some individuals is compelling in that the sustained benefit can be verified objectively. However, there has been no attempt to study this interindividual variation in tolerance development directly.

Zopiclone is the oldest of the "new" compounds considered here, having been introduced in the late 1970s. Accord ingly, there have been many more studies of the clinical effects of this drug than of the others. Most controlled studies of zopiclone have compared the hypnotic effects of a 7.5-mg dose with those of benzodiazepine hypnotics (13). Sleep laboratory studies, in which zopiclone was compared with flurazepam 30 mg (35, p. 25S), triazolam 0.25 mg (35, p. 32S), or nitrazepam 5 mg (35, p. 29S), showed each compound to be effective, with no significant differences between them on any parameters of effect. In studies using subjective ratings, zopiclone was equivalent to flurazepam 30 mg (35, p. 25S), triazolam 0.25 mg (35, p. 32S), temazepam 20 mg (35, p. 38S), and nitrazepam 5 mg (35, p. 29S) on most or all measures. Among other comparative studies using patient questionnaires, some found zopiclone to be superior to benzodiazepines on some parameters (35, pp. 29S, 32S–33S), whereas others found the benzodiazepines to be superior (35, pp. 32S–33S, 36S).

Sleep laboratory studies have also provided objective evidence of the hypnotic efficacy of the newer agent, zolpidem (15, 39); the latter study found that zolpidem produced fewer alterations of sleep "architecture" than did flunitrazepam. Other controlled studies have also shown the efficacy of zolpidem as compared with triazolam (60), flunitrazepam (23), or placebo (67). Zolpidem also appears to be useful as a preanesthetic medication, producing good sedation as well as anterograde amnesia (14, 56).

In the single placebo-controlled study of the anxiolytic efficacy of abecarnil reported to date (3), patients with GAD who received 3–9 mg daily for 3 weeks were significantly more improved than controls, according to global evaluations as well as scores on the Hamilton Anxiety Rating Scale. Higher doses of abecarnil were not more effective and were associated with a higher incidence of sedative side effects.

The most common side effects of benzodiazepines in routine clinical use are manifestations of excessive depression of the CNS; adverse effects on other physiological systems are rare. Sedative side effects most frequently include drowsiness, muscle weakness, lightheadedness, vertigo, ataxia, dysarthria, diplopia, blurring of vision, confusion, apathy, and vertigo (e.g., see refs. 57 and 66). The relative risk of such effects varies with individual patient susceptibility; because of pharmacokinetic changes associated with aging, for example, the elderly may be at increased risk (27).

A large number of experimental and clinical studies have attempted to assess the liability of benzodiazepines to impair psychomotor performance. Various tests of performance in normal, anxious, and insomniac subjects have shown effects of single doses of benzodiazepines within the therapeutic range. When administration is repeated over several days, these effects diminish. Effects over much longer periods of time have not been adequately studied. There is remarkably little agreement among studies with respect to the types of performance most susceptible to the effects of benzodiazepines (76, pp. 290–300; 77, pp. 207–214).

With respect to the newer compounds considered here, a significant number of studies have assessed the effects on psychomotor performance of zopiclone, and somewhat fewer studies have assessed such effects of zolpidem. As with most of the benzodiazepines that have been studied, zopiclone at therapeutic doses produced a decrement in performance on various psychomotor tests when these tests were administered shortly after treatment (25, 40). When the drug was given at night, performances on the following morning were usually not disrupted (32, 54, 58; but also see refs. 10 and 12). A similar profile has been demonstrated for zolpidem (7, 19, 23, 50). These findings are consistent with the relatively short durations of actions of these drugs. Only one study (17) has assessed effects of abecarnil on performance. Performance was disrupted by 20 and 40 mg of abecarnil, with recovery at 24 hr after treatment; when the drug was administered over 9 days, patients developed some tolerance to these effects. Bretazenil has been reported to produce dose-related disruptions in psychomotor performance; however, the slope of the dose–response curve was not as steep as that for diazepam or alprazolam (64).

Among the experimental behaviors affected by benzodiazepines are performances in simulated or real automobile driving. However, whether such drug effects actually contribute to driving accidents can be established only by case–control epidemiological studies of drivers involved in accidents. The few studies of this kind that have attempted to address the question have not found evidence that benzodiazepines contribute to automobile accidents (76, pp. 308–315; 77, pp. 229–233). This conclusion has been substantiated in a recent study comparing accident victims that were and were not responsible for the accidents; there was no difference between these groups with respect to presence of benzodiazepines in blood (8).

A number of recent epidemiological studies have examined the association between use of benzodiazepines and the risk of falls and/or hip fractures in elderly populations. Some, but not all, of these studies have found a significant positive association (77, pp. 233–237). In sum, the research suggests that benzodiazepines can contribute to the risk of falls among elderly patients, although the extent of this contribution alone is apparently not great.

The greatest recent advance in our understanding of the psychomotor effects of benzodiazepines comes from increasingly sophisticated research on human recall. It can now be shown reliably that acute doses of these drugs can markedly impair recall, especially delayed recall (77, pp. 215–216). Further study is needed to establish whether and to what extent tolerance to any or all of the drugs' amnestic effects may develop. Benzodiazepines produce decrements in recall in elderly subjects, but these decrements are not greater than those produced in younger subjects; however, because of baseline deficits in recall in the elderly, the additional impairment produced by benzodiazepines may represent a more severe compromise (77, p. 219). Further research is needed to pursue important recent suggestions that benzodiazepines may vary substantially in their effects on recall (77, pp. 216–217).

A number of early case reports of hostile behavior in patients taking benzodiazepines prompted several experimental investigations of the older, longer-acting agents, with conflicting results. Some of the case reports described increased hostility and aggression following ingestion of these drugs (76, pp. 316–317). It is possible that these reports reflected phenomena similar to those described in the numerous and notorious case reports of untoward behavioral reactions to the short-acting, relatively more potent agent, triazolam. On the basis of these reports, triazolam has been subject to various official restrictions in several countries. Regardless of their numbers, however, case reports cannot substitute for controlled studies.

Very few controlled studies have inquired into the incidence or nature of daytime distress during and after periods of nightly administration of triazolam. Some studies found certain adverse effects during and/or after treatment, including increased anxiety (38, 52) and apparent idiosyncratic reactions such as panic attacks, personality changes, and delusional episodes (1); others have found no increased anxiety (11, 45, 61) or hostility (61) during triazolam treatment or upon discontinuation. Further controlled research is certainly needed to establish whether triazolam and/or other short-acting benzodiazepines can produce such adverse effects, and, if so, what their nature and incidence might be.

The relative safety of the benzodiazepines is most clearly evident in cases of overdose. Even massive overdoses, if taken without other CNS depressants, are almost never fatal. Morbidity and mortality associated with drug overdoses declined dramatically as benzodiazepines replaced older sedative–hypnotics; this has been one of the most important and least controversial advantages of this drug class (26, 57; 76, pp. 377–385; 77, pp. 302–314).

There is as yet no reliable evidence regarding the incidence or characteristics of overdoses of the newer compounds considered here. It will be especially interesting to learn whether overdoses of zolpidem are associated with less risk than are overdoses of other hypnotics.

In view of the extensive worldwide experience with benzodiazepines during the past 30 years, epidemiology should be our most important guide in assessing their risks, including their liability for abuse. This is particularly appropriate in view of the remarkable agreement among a great many diverse sources of information about how these drugs are used, which permits a high degree of confidence in the reliability of the patterns described.

In most countries for which data are available, sales of benzodiazepines increased steadily from the time of their introduction until the mid- to late 1970s; during this period they largely displaced the barbiturates (76, pp. 324–334). Sales then declined substantially until the early 1980s, after which there was some increase throughout the decade. In the 1980s, sales of benzodiazepine hypnotics increased far more rapidly than sales of benzodiazepine anxiolytics (77, pp. 239–250). Cross-national surveys conducted in 1981 indicated that an average of 12% of the adult populations of a number of Western European countries and the United States reported using anxiolytics (of which more than 80% were benzodiazepines) during the previous year (4). National data indicating trends in use are available for the United States, where the overall prevalence of annual use of anxiolytics declined from 11.1% of the adult population in 1979 to 8.3% in 1990; use of hypnotics meanwhile remained stable, at about 2.5% of adults (5).

As benzodiazepines originally came to be widely used, and indeed long thereafter, it was commonly perceived that this widespread use of psychoactive drugs was a uniquely contemporary phenomenon. This perception led many observers to assume that the benzodiazepines were overprescribed and overused, without consideration of the prevalence of the disorders for which the drugs were indicated. However, a historical review of United States prescription surveys indicates that benzodiazepines have not accounted for a greater proportion of prescriptions than have other sedative–hypnotics over the course of at least a century. Similarly, prescription data for the United Kingdom show that sedative–hypnotics accounted for about 15% of all prescriptions dispensed in 1949–1951 (chiefly bromides and barbiturates) as well as in 1975 (chiefly benzodiazepines). It seems reasonable to assume that the relative stability of sedative–hypnotic consumption is a function of the relatively stable morbidity that motivates use of these drugs (76, pp. 321–324).

Surveys comparing drug use and psychiatric morbidity in the general population have found that actual use of anxiolytics is generally appropriate, in that users report high levels of emotional distress. On the other hand, the vast majority of people afflicted by psychiatric problems do not seek or receive treatment (76, pp. 346–348). Thus, the problem is not contemporary overmedication, but rather the continued history of undertreatment of psychiatric illness.

Although rates of use of benzodiazepines vary widely across geographic areas, the demographics and patterns of use within populations are strikingly similar. In virtually every population studied, women receive about twice as many prescriptions for these drugs as do men. Also, use of anxiolytics increases to a peak prevalence in people aged about 50–65 years and declines somewhat in older people, whereas use of hypnotics is most frequent in the oldest age range (77, pp. 254–260).

Most people who receive benzodiazepine prescriptions use the drugs for relatively short periods of time—for example, 4 weeks or less. Patients tend to use lower doses than those prescribed, and to reduce use over time. However, a substantial minority of those who receive benzodiazepine prescriptions continue to use the drugs on a regular basis for longer than a year (76, pp. 355–361).

As suggested above, elderly patients receive a disproportionately large fraction of benzodiazepine prescriptions. They are also more likely than younger patients to use benzodiazepines on a daily basis and to continue use for long periods of time, often for years. Surveys show that long-term users are likely to be elderly patients with multiple chronic physical disorders; they receive prescriptions for benzodiazepines concurrently with multiple other medications (76, pp. 356–357, 367–369; 77, pp. 273–275).

In the United States and some Western European nations, which have historically led trends in drug use, rates of use of anxiolytics have declined in recent years, whereas use of hypnotics has remained stable or increased (77, p. 262). Because hypnotic use is particularly prevalent among older patients, the stability of use of these drugs may largely reflect a contingent of elderly patients who use hypnotics regularly for long periods. Although the last few years have brought a considerable increase in epidemiological information about older patients' use of benzodiazepines in general, there has been little attention specifically to use of hypnotics among these patients; because hypnotic use may be basically distinct from anxiolytic use in many respects, this represents an important focus for future study.

Recent years have also seen a dramatic increase in many countries in use of the benzodiazepines introduced in the late 1970s and later, particularly those with short elimination half-lives; this increase has come at the expense of a proportional decline in use of the older compounds, particularly those with long half-lives (77, pp. 240–250). An important result of this shift is that many patients are now taking single daily doses of benzodiazepine hypnotics with short half-lives, and may thus be exposed to the risk of interdose withdrawal or rebound (discussed further below); the liability for such risks should be assessed in animals as well as in humans. Interview surveys with patients using the short-acting hypnotics could also be designed to provide useful insights into the correlates and effects of this regimen.

Despite the wide availability and extensive medical use of benzodiazepines, there has been very little misuse or recreational use of the drugs among adults or youths in the general population (76, pp. 371–373; 77, pp. 286–297). There has been some nonmedical use of the drugs among populations of drug abusers, though benzodiazepines have usually not been the primary drugs of abuse (76, pp. 373–375; 77, pp. 297–302). These findings from epidemiological research parallel the results of experimental studies (described below) that have demonstrated no preference for benzodiazepines in normal subjects. In subjects with histories of sedative abuse, although there are virtually no reinforcing effects of doses within the therapeutic range, modest reinforcing effects are seen at higher doses.

Benzodiazepines are found with some frequency in overdose surveys, usually in combination with other drugs (76, pp. 377–385; 77, pp. 302–314). When the frequency of overdose cases is examined in relation to the volume and patterns of prescriptions, the frequency of cases involving benzodiazepines is substantially lower than that of other prescribed drugs (e.g., analgesics), and the relative frequency of cases involving individual benzodiazepines is generally proportional to their respective medical availability (76, p. 384; 77, p. 311). These drugs are rarely implicated in fatal overdoses (76, pp. 383–384; 77, p. 311). Overdoses involving benzodiazepines are most likely to result from suicide attempts rather than from accidental consequences of recreational use; in this respect, these overdoses are like those typical of other psychotherapeutic agents and distinct from those typical of benchmark drugs of abuse (76, pp. 384–385; 77, pp. 308–311).

Survey data from the United States have documented continuing declines in nonmedical use of benzodiazepines in the general population. Periodic household surveys by the National Institute on Drug Abuse have shown that, between 1985 and 1992, annual rates of nonmedical use of tranquilizers decreased by more than half among adolescents (12–17 years old), young adults (18–25 years old), and adults (26 years of age and older) (72, 73); in 1992, between 1% and 3% of a national sample reported nonmedical use within the previous year, and 0.6% or less within the previous month (73). Periodic surveys in the United States and Canada have shown that nonmedical use of tranquilizers among youth has declined fairly steadily since the late 1970s (36, 37; 77, pp. 288–289); in 1992, the monthly prevalence of such use among United States high school seniors and college students, respectively, was 1.0% and 0.6% (36, 37). Data from the Drug Abuse Warning Network (DAWN) show that the frequency of mentions of benzodiazepines in overdose cases in the United States has been decreasing since the mid-1970s (77, pp. 308–309).

To date, there is no appreciable epidemiological evidence regarding abuse of the newer compounds considered here. It will surely be some considerable time before we know, for example, how abuse of these drugs may differ from abuse of the established benzodiazepines.

We define the abuse liability of a compound as its capacity to produce psychological dependence (which we prefer to address in terms of objective measures of drug taking), or physiological dependence, in conjunction with the capacity to alter behavior in a manner that is detrimental to the individual or to his or her social environment. The following section reviews experimental and clinical studies of the reinforcing effects of benzodiazepines and of their potential to produce physiological dependence.

When a subject takes (or "self-administers") a drug, the pharmacological effects are a consequence of the behavior. If this behavior subsequently increases in frequency, or if frequent self-administration is maintained, the behavior is considered to have been reinforced by the drug. This psychological process is an essential determinant of the abuse liability of a drug. Studies of reinforcing effects of drugs have largely supplanted studies of "psychological dependence" and have proven to have predictive value in identifying drugs of abuse.

In self-administration studies imposing a wide variety of experimental conditions, benzodiazepines appear only marginally effective as reinforcers. There is little or no preference for benzodiazepines in oral forms, even in physiologically dependent subjects (77, pp. 158–159, 162). Studies in which multiple responses are required for each intravenous injection, which represent relatively stringent measures of reinforcing effects, have demonstrated that benzodiazepines maintain response rates above those maintained by vehicle alone; however, these rates are typically lower than those maintained by reference drugs such as cocaine, codeine, or several barbiturates (76, pp. 258–260; 77, pp. 161–162).

Some studies have suggested that relative onset of action may be a critical factor in determining whether benzodiazepines will function as reinforcers (76, pp. 259–260; 77, p. 162). This suggestion, from studies comparing several compounds, has not been supported by systematic evaluation of all benzodiazepines; nevertheless, the possibility is important and should be pursued directly in experimental studies. There is also some evidence that self-administration of diazepam might be maintained more readily in subjects trained to self-administer barbiturates than in those trained to self-administer stimulants (76, p. 259; 77, pp. 161–162). The influence of drug use history on the reinforcing effects of benzodiazepines is an important question that has also been raised in human studies (see below) and that deserves systematic study in animals. Similarly, as described below, clinical observations suggest that reinforcing effects of benzodiazepines are increased in patients undergoing withdrawal. Relevant studies in animals do not support this conclusion (77, pp. 159 and 162). However, these studies were not designed explicitly to assess reinforcing effects of benzodiazepines during withdrawal; such studies are clearly needed.

Only a few studies have examined the reinforcing effects of the newer compounds in animals. Zopiclone self-administration was maintained in rhesus monkeys at rates higher than those maintained by vehicle (79). Bretazenil did not maintain responding in cynomolgus monkeys trained to self-administer pentobarbital (47). In baboons trained to self-administer cocaine, abecarnil failed to maintain responding (62), whereas zolpidem maintained self-administration at rates similar to those obtained with cocaine or methohexital and considerably greater than those shown previously with most other benzodiazepines (28, 78). Thus, drugs (such as zolpidem) that act as full agonists on a subset of benzodiazepine receptors appear to have marked reinforcing effects, whereas drugs that act as partial agonists may have reduced reinforcing effects.

Evidence in humans pertaining to the liability of benzodiazepines for abuse is largely limited to experimental and epidemiological studies. Few of these studies have examined indices of inappropriate use in clinical populations, or in conditions of therapeutic use. This section considers experimental evidence of abuse liability of benzodiazepines in normal and anxious subjects and in subjects with histories of sedative abuse.

A number of studies have consistently shown that, when given a choice between diazepam and placebo, normal volunteers prefer to take placebo. Even anxious subjects tend to choose placebo over diazepam, at least if they are not actively seeking treatment for their anxiety. Anxious subjects seeking treatment were more likely to choose diazepam over placebo, although only a minority always selected the active drug (76, pp. 261–262; 77, pp. 164–165). These findings suggest that it is unlikely that most use of these drugs in patient populations is associated with a significant risk of abuse.

On the other hand, patients undergoing withdrawal following abrupt discontinuation of diazepam show an increased tendency to self-administer the drug (77, p. 165). This suggests that physiological dependence to benzodiazepines, and the withdrawal signs resulting from their discontinuation, may maintain ingestion of these drugs. However, studies of dependent patients (see below) have found no tendency to escalate doses. Self-administration of benzodiazepines during withdrawal appears to reflect an effort to relieve distressing symptoms; after the withdrawal syndrome abates, patients do not exhibit continued "craving" to resume benzodiazepine use (43).

Perhaps the most interesting recent finding is that subjects who were moderate users of alcohol (consuming an average of 11 drinks weekly), in contrast to lighter drinkers, consistently chose to take diazepam rather than placebo (16). Previous studies had shown consistent preference for benzodiazepines only in populations of sedative abusers (76, pp. 261–267; 77, pp. 164–167). It will be of great interest to determine the specific conditions associated with diazepam selection in these experiments with moderate drinkers and to explore whether moderate use of other classes of drugs (e.g., other sedatives or stimulants) might also be associated with increased preference for benzodiazepines.

Subjects with histories of sedative abuse show a preference for benzodiazepines over saline or vehicle; this preference is less than that induced by short- or intermediate-acting barbiturates. Sedative abusers also express greater "liking" for benzodiazepines than for placebo (76, pp. 263–265). Studies of sedative abusers often examine effects of only a single dose of each drug tested, which may be a limitation of this research. In addition, studies of sedative abusers have increasingly tended to focus on subjective effects to the exclusion of behavioral effects; although there appears to be a reasonably good correlation between drug-liking and drug-taking in these subjects, it remains important to observe actual drug selection and ingestion in order to assess reinforcing effects. Also, because studies of this population have indicated that reinforcing effects of benzodiazepines tend to decrease over time (76, p. 263), experimental observations of subjective and reinforcing effects of these drugs should be assessed for a period sufficient to obtain stable evaluations of effects.

There have been few studies of the abuse liability of the newer benzodiazepine receptor ligands in humans. Forty inpatients recently detoxified from alcohol were trained to distinguish 0.25 mg triazolam from 3.75 mg zopiclone. When given a choice of which to take, 25 chose triazolam and 15 chose zopiclone. The preference for triazolam appeared to be due to the perception that this drug had greater antianxiety effects. The drugs produced nearly equal reports of hypnotic effects and euphoric effects (6).

Effects of several doses of triazolam and zolpidem were studied in subjects with histories of sedative abuse. At the two highest doses tested (0.5 and 0.75 mg triazolam, and 30 and 45 mg zolpidem), subjects reported "drug-liking" greater than that with placebo, with no significant difference between the drugs. Interestingly, drug-liking as rated the next day was not different from placebo for either drug; this finding supports the need for assessment of such effects over time. Neither zolpidem nor triazolam produced increases in the MBG scale of the ARCI questionnaire, a scale that traditionally measures drug-induced "euphoria" (19).

Sellers et al. (64) compared the effects of several doses of diazepam, alprazolam, and bretazenil on subjective measures in sedative abusers. Although each drug produced effects greater than those of placebo on most scales, the effects of diazepam and alprazolam were dose-related, whereas those of bretazenil were not clearly related to dose. On the basis of the subjective effects measured, bretazenil appeared to have less liability for abuse than did diazepam or alprazolam.

Physiological dependence is a condition of the organism, induced by drug treatment, that results in a time-limited withdrawal reaction when treatment is discontinued or when an antagonist (e.g., flumazenil) is administered.

Several studies in animals have shown that large doses of benzodiazepines can produce physiological dependence. These studies have also provided some information about the extent to which this physiological dependence conforms to the general rules derived from studies of this phenomenon with opioids, barbiturates, and ethanol. The general rules suggest that withdrawal signs are more frequent or of greater magnitude (a) following administration of doses with greater effects, (b) following treatment for longer periods of time, and (c) following continuous rather than periodic drug administration.

Most studies of benzodiazepines have found results consistent with the first general rule (68; 76, pp. 271 and 276; 77, pp. 173–174). However, some findings suggest that the magnitude of the withdrawal may reach some asymptote beyond which increases in dose have no effect, or have different effects on a composite withdrawal score (76, pp. 271 and 276) or on individual withdrawal signs (68).

The rule that the frequency or intensity of withdrawal should vary as a function of the duration of treatment has been borne out consistently in studies of benzodiazepines (76, pp. 271 and 276–277; 77, p. 174).

Whether benzodiazepine dependence is more frequent or of greater magnitude following continuous rather than periodic administration has not been systematically investigated. It seems particularly important to conduct such studies since the advent of short-acting compounds, which are often used clinically as hypnotics (i.e., in single daily doses). This has introduced the phenomenon of repeated intermittent, rather than continuous, exposure to the agonist. Some animal studies have indicated that dependence can develop to single daily exposures to midazolam (20), although this compound is eliminated within a few hours. However, studies have not addressed the possibility, suggested by some clinical observations, that once-daily administration of short-acting benzodiazepines might produce repeated episodes of acute dependence and withdrawal. The behavioral and biochemical correlates and consequences of this pattern of benzodiazepine use should be pursued systematically.

Results of several studies have suggested that benzodiazepines might differ in their potential to produce physiological dependence. Of particular interest is a series of studies of precipitated withdrawal by Martin et al. (48), which suggest that the withdrawal syndromes following exposure to different benzodiazepine agonists may consist of overlapping but distinct constellations of signs. Following chronic treatment with several agonists in dogs, flumazenil was administered and the nature and intensity of withdrawal was assessed by scoring individual signs. Withdrawal was most intense in subjects treated with diazepam, less intense with flunitrazepam or halazepam, lesser still with nordiazepam or alprazolam, and least for those treated with oxazepam. Seizures were most frequent in dogs treated with alprazolam, diazepam, or flunitrazepam, less frequent with nordiazepam, and least frequent with halazepam or lorazepam. The investigators identified three different syndromes associated with different agonists, characterized by relative frequency of seizures and relative magnitude of withdrawal scores. They suggested that differences among these syndromes may be due to differences in the mechanisms and sites of action of the benzodiazepines or their metabolites.

The capacity of some of the newer sedative/anxiolytic agents to produce physiological dependence has been evaluated to a limited extent in animals. Following chronic treatment with abecarnil subcutaneously at doses that elevated seizure threshold, flumazenil produced few signs of withdrawal in dogs, and only one of seven subjects showed a lowered threshold for pentylenetetrazol-induced seizures after treatment was discontinued (41). In a subsequent study in dogs, abecarnil and diazepam were chronically administered subcutaneously at doses with approximately equal anticonvulsant effects; antagonists produced signs of withdrawal in all subjects treated with diazepam, but signs of precipitated withdrawal were much less frequent in subjects treated with abecarnil (42; see also ref. 65). In baboons given abecarnil by continuous intragastric infusion, two of four subjects exhibited "mild" signs of withdrawal upon administration of flumazenil and after discontinuation of 6–8 weeks of treatment (62). Steppuhn et al. (71) examined mice treated with subcutaneous depot injections of abecarnil or diazepam at doses producing equivalent time courses and receptor occupancy. Only the mice treated with diazepam showed signs of withdrawal after discontinuation. This study reveals the importance of relating response to receptor occupancy; by including this comparison, the authors could strongly suggest an intrinsic difference in the potential of these drugs to produce physiological dependence.

Flumazenil produced clear withdrawal signs in squirrel monkeys treated with diazepam, but not in those treated with bretazenil (46). In another study (47), squirrel monkeys treated with a range of doses of bretazenil or alprazolam were challenged with the benzodiazepine partial agonist, sarmazenil (Ro 15-3505); sarmazenil produced convulsions in a smaller proportion of the subjects treated with the various doses of bretazenil compared with alprazolam. Because bretazenil is more potent than alprazolam in producing a range of pharmacological effects, these findings led the authors to conclude that it has a greater potential to produce physiological dependence.

Moreau et al. (51) compared bretazenil with triazolam, alprazolam, and diazepam in convulsion-prone DBA/2J mice. After 1 week of continuous treatment using osmotic minipumps, sarmazenil produced signs of withdrawal that varied in frequency with dose of triazolam, alprazolam, or diazepam; withdrawal signs were not observed in animals treated with bretazenil or vehicle alone. These investigators also conducted in vivo studies of receptor occupancy, which established that both bretazenil and alprazolam were bioavailable in the CNS; this represents important evidence that the difference between the drugs in producing dependence is a function of differences in their intrinsic efficacy, rather than their relative CNS bioavailability. Additional evidence of an intrinsic difference in the dependence potential of these drugs was provided by a further study in which mice were treated with doses of bretazenil and alprazolam that were roughly equivalent multiples of their anticonvulsant ED50 values; sarmazenil precipitated withdrawal in mice treated with alprazolam, but not in mice treated with bretazenil.

In a procedure for rapid evaluation of dependence-producing effects of benzodiazepines, after 3 days of treatment mice were injected with flumazenil and the threshold for electric-shock-induced seizure was determined. Flumazenil substantially reduced seizure threshold following treatment with several benzodiazepine agonists and related compounds, including chlordiazepoxide, diazepam, flurazepam, alprazolam, triazolam, midazolam, and zopiclone, as well as the partial agonists bretazenil and Ro 17-1812. In contrast, seizure threshold was not altered after treatment with zolpidem, tracazolate, or CL 218, 872. In addition, although bretazenil and alprazolam had similar ED50 values for inhibition of in vivo binding, the dose of bretazenil required to produce any sign of dependence was 10 times greater than that of alprazolam. Similarly, the ED50 values of the full agonist triazolam and the partial agonist Ro 17-1812 were similar, though 10-fold-higher doses of Ro 17-1812 were required to produce signs of dependence (74). These data are consistent with the conclusion that drugs with limited efficacy are also limited in their liability for producing physiological dependence.

In another study in mice, zolpidem or midazolam was administered for 10 days at pharmacologically equivalent doses, based on time course and effectiveness in suppression of locomotor activity. Latency to isoniazid-induced convulsions was used as an index of withdrawal. Both precipitated and spontaneous withdrawal was observed in mice treated with midazolam; neither was observed in those treated with zolpidem (55).

In contrast to the above studies, Griffiths et al. (28) obtained a result that might indicate a physiological dependence on zolpidem. Baboons were trained to earn food pellets during and after periods of zolpidem self-administration. When zolpidem was replaced with vehicle, five of seven baboons substantially decreased their responses for food. The suppression of this behavior proved to be time-limited and may therefore have represented withdrawal; to confirm this interpretation, it would have been useful to determine whether the effect could have been reversed by resumption of treatment. In earlier studies, Yanagita (80) rated the magnitude of withdrawal from zopiclone, diazepam, and nitrazepam in rhesus monkeys. Discontinuation of chronic administration of zopiclone and nitrazepam produced withdrawal of intermediate intensity, whereas withdrawal from diazepam was rated as severe. The author noted that the lower intensity of withdrawal ratings of zopiclone may have been due to its relatively short time course (76, pp. 272–273).

A number of studies have suggested that benzodiazepines may differ in their potential to produce dependence or in the characteristics of the dependence they produce. For example, results of studies by W. R. Martin and co-workers suggest that the withdrawal syndromes following administration of different benzodiazepine agonists may consist of overlapping but distinct constellations of signs. Studies with the beta-carboline abecarnil and the imidazopyridine zolpidem, as well as studies with benzodiazepine partial agonists, suggest that these compounds may have less potential to produce dependence than do classical benzodiazepine agonists. In addition, as noted above, there have been suggestions that the partial agonists are not effective as reinforcers.

It has also been suggested that the onset of withdrawal might be more rapid and the intensity of withdrawal might be greater following discontinuation of short-acting benzodiazepines than after discontinuation of long-acting compounds. However, when the short-acting benzodiazepine midazolam was administered for a period equivalent to the duration of the effects of a single dose of the long-acting chlordiazepoxide, spontaneous withdrawal from the two regimens was comparable in intensity (12). Further studies that equate the benzodiazepines according to all dosing parameters, except their speed of elimination, will be of importance in verifying these results. Meanwhile, the available evidence from animal studies is not consistent with clinical data (see below) indicating that withdrawal is more intense following termination of short-acting as compared with long-acting benzodiazepines. However, animal and human studies appear to concur that the onset of withdrawal is more rapid following discontinuation of the short-acting agents (77, pp. 178–180 and 187–195).

Finally, it may be noted that an increasing number of behavioral procedures purport to assess rebound anxiety using animal models (77, pp. 183–184). In the majority of these procedures, the models used have not been validated according to standard pharmacological criteria for establishing behaviors as withdrawal phenomena. Moreover, the behaviors measured have not been shown to be functionally equivalent to human anxiety; nor have they been shown to have any utility for prediction of clinical outcomes. Few of these studies to date have made a substantial contribution to the understanding of benzodiazepine withdrawal.

A withdrawal syndrome following abrupt discontinuation of high doses of chlordiazepoxide was first described in 1961 (33). The possibility that dependence could develop at therapeutic doses of benzodiazepines was recognized only around 1980 (66; 76, pp. 279–285). Although high-dose benzodiazepine dependence surprised no one familiar with the pharmacology of sedative drugs, the phenomenon of dependence at therapeutic doses had not been anticipated (34); systematic study of barbiturates and other older sedatives had shown that dependence developed only at doses far above the therapeutic range (18, 22). Although dependence on therapeutic doses of benzodiazepines has now been documented in numerous studies and remains a focus of widespread concern, our understanding of this phenomenon has not advanced appreciably.

The studies that established the potential of benzodiazepines to produce physiological dependence at therapeutic doses also found that not all patients using these drugs became dependent; that is, some did not show signs of withdrawal when the drugs were discontinued (76, p. 280). More recent studies have supported this finding, although they have not clarified what proportion of users do become dependent nor have they confirmed the determinants of dependence (77, pp. 195–198). Most studies of therapeutic-dose dependence examined the effects of discontinuation in patients who had used the drugs for relatively long periods, and it was often assumed that the risk increased with duration of use; studies in animals supported that view (76, pp. 279–285). However, more recent human studies have demonstrated withdrawal after relatively brief periods of use (e.g., 2 weeks), and some studies comparing discontinuation after different periods of treatment have found no difference in the incidence of withdrawal (53; 77, pp. 197–198). Other factors that have been considered as possible determinants of the development of physiological dependence have included duration of drug action, magnitude of dose, patients' prior drug use, age, and personality traits; although some of these factors have influenced the development of dependence in animals, none has yet been shown to be clearly related to the risk of dependence in patients (77, pp. 204–205). As described previously, the development of tolerance to therapeutic effects of benzodiazepines evidently varies widely among individuals. This suggests the importance of examining individual variables as they may interact with the pharmacological determinants of benzodiazepine dependence.

Upon abrupt discontinuation of benzodiazepine treatment, dependent patients are likely to experience increased anxiety and/or insomnia. Other characteristics of the benzodiazepine withdrawal syndrome include alterations in taste and smell sensations, tremor, restlessness, gastrointestinal distress, sweating, tachycardia, and mild systolic hypertension. The syndrome and associated discomfort are usually mild, reaching peak severity in 2–10 days and abating within 4 weeks after discontinuation (59, 66; 76, pp. 289–290). Some recent studies have suggested that symptoms and signs of benzodiazepine withdrawal can persist for many months or for years, but the available evidence is not convincing (66; 77, p. 205).

Although it has not been established whether dependence is more or less likely to develop with short- versus long-half-life benzodiazepines, recent studies have shown that withdrawal from the short-acting drugs develops more rapidly and may be more intense than withdrawal from benzodiazepines with longer durations of action. It has also been found that effective discontinuation of short-acting benzodiazepines, to minimize the risk of rebound symptoms and other effects of withdrawal, requires a particularly gradual and prolonged tapering regimen and that, at least in the short term, patients discontinued from these drugs are more likely to resume use than are patients discontinued from long-acting agents (77, pp. 187 and 192–195).

Despite concerns about the risk of physiological dependence on benzodiazepines, there has been little consideration of ways in which dependence might be prevented from the onset of treatment. Clinical authorities have long recommended that use should be interrupted by occasional "drug holidays," which would permit reassessment of the need to continue treatment and might also reduce the risk of dependence development; however, the effect of drug-free intervals on the development of dependence has not been studied under controlled conditions. Another possible approach to prevention has been suggested by studies of chronic benzodiazepine treatment in animals; periodic injections of a benzodiazepine antagonist significantly reduced the intensity of the subsequent withdrawal syndrome (24). The suggestion that it might be possible in this way to "reset the clock" for the initiation of dependence needs to be pursued through studies of a variety of benzodiazepines in both animals and humans.

The potential of zopiclone to produce physiological dependence in humans has not been extensively studied. A review by Bianchi and Musch (9) found that in 25 studies of zopiclone including assessments of possible withdrawal phenomena, only a small minority found such effects. Nervousness, anxiety, or vertigo appeared as possible withdrawal signs in seven (1.6%) of 441 patients treated with zopiclone, whereas possible withdrawal signs were observed in 6%, 7%, and 10% of patients treated with flunitrazepam, triazolam, and flurazepam, respectively. In comparative trials of hypnotic use, rebound insomnia was observed following discontinuation of nitrazepam, flurazepam, and triazolam, but not after discontinuation of zopiclone. Among patients with insomnia associated with GAD, nightly treatment with zopiclone was associated with less daytime anxiety than was treatment with nitrazepam (2) or triazolam (21).

The dependence potential of zolpidem has likewise received little study in humans. However, the available data are consistent in reflecting no rebound insomnia or withdrawal signs in insomniac patients receiving zolpidem daily for periods from 7 to 180 days (44, 63, 67).

Abecarnil was administered in three different dosage ranges to patients with GAD. Following 3 weeks of treatment, insomnia, anxiety, and other possible signs of withdrawal were observed among patients who received active medication, but not among a placebo control group. The incidence of withdrawal signs was positively related to the dose administered (3).

Benzodiazepines have been on the market for over 30 years and have achieved remarkable clinical success. The introduction of these drugs in clinical practice represented a very significant advance over previous generations of sedative–hypnotics. Some problems of course remain, associated with both abuse and therapeutic use of benzodiazepines. From our current perspective, it appears that it may be difficult to overcome these problems.

Therapeutic use has largely shifted from the older, longer-acting benzodiazepines to the shorter-acting agents that became available more recently. However, it is not clear that this represents an advance in the appropriate use of these drugs. This massive shift in prescribing should be regarded as a compelling stimulus to well-controlled assessments of the relative advantages and disadvantages of the longer- versus the shorter-acting agents in a variety of clinical applications.

Similarly, it is not clear that efforts to develop newer sedative–hypnotics, such as those considered in this chapter, will readily resolve the kinds of problems associated with established benzodiazepines. There are multiple genetic forms of the benzodiazepine receptor. It will be no easy matter to find ligands selective for them. Moreover, if such ligands can be identified, there is a long and arduous path, from in vitro pharmacology through pharmacodynamics, to establish whether they bear pharmacological characteristics that are relevant to therapeutic or adverse effects. Although the prospect of this work is daunting, it is certainly one of the most exciting challenges for the future of benzodiazepine research.

The partial agonist compounds bretazenil and abecarnil may well represent the most interesting leads to date. The available information on both of these compounds suggests that they may represent significant progress toward reducing the risks associated with reinforcing effects, subjective effects associated with abuse liability, and potential to produce physiological dependence. Human trials of abecarnil are likely to provide new challenges to animal pharmacology and the concept of partial agonism, as well as information about therapeutic effects of this compound (e.g., see ref. 70).

The amalgamation of new molecular developments in receptorology and research toward development of new ligands for the benzodiazepine receptor should prove to be one of the most productive avenues for the growth of knowledge in this area of therapeutics.

The authors thank Kaim Associates, Inc., for support in collection, organization, and management of the literature reviewed and for administrative and editorial assistance.

published 2000