Issues in the Long-Term Treatment of Anxiety Disorders

Edward Schweizer, Karl Rickels, and Eberhard H. Uhlenhuth

The goal of this chapter is to review issues relating to the long-term treatment of anxiety disorders. The review does not pretend to be exhaustive, but it has been written with the intention of critically addressing clinically important aspects of long-term anxiolytic therapy. The scope of the review is limited to panic disorder with or without agoraphobia, generalized anxiety disorder (GAD), and social phobia. Most of the emphasis is placed on drug therapy. Space limitations, and the limitations of the authors themselves, have dictated that results from psychotherapy studies could only be highlighted. It was felt, though, that the rapidly expanding literature of controlled psychotherapy research simply could not be ignored, because their preliminary results stake a claim to the achievement of sustained efficacy post treatment.

The review will first undertake to establish whether there is any need for long-term or maintenance anxiolytic therapy. Long-term therapy can only be justified if there is evidence that the course of illness of panic/agoraphobia, GAD, and social phobia are frequently chronic. The longitudinal nature of these disorders will be evaluated in terms of retrospective reports, prospective studies of community samples (ideal but not generally available), and follow-up studies of treated samples. Acute (short-term) treatment studies will also be reviewed: High rates of relapse or recurrence in the months and years following acute treatment suggest that such treatment is inadequate, and that continuation or maintenance treatment may be beneficial. Establishing how beneficial leads us to a brief review of recent research on the degree of disability and impairment in quality of life caused by the anxiety disorders. The justification section will conclude with a brief review of the extent to which the natural history of the three anxiety disorders is complicated by Axis I and Axis II comorbidity. Long-term treatment planning cannot be undertaken without considering these high rates of comorbidity.

Next will come a brief review of published evidence for the efficacy of long-term treatment of each of the three anxiety disorders. Evidence for long-term efficacy will be reviewed for the various classes of antidepressants, as well as for the benzodiazepines and buspirone. Long-term treatment will be conceptualized, in keeping with the terminology employed in the affective disorders literature, as either "continuation" treatment or "maintenance" treatment. Continuation treatment consists of the extension of drug therapy for 6 months beyond the acute phase in an attempt to prevent relapse, defined as a return of the current episode of anxiety. Maintenance treatment consists of the long-term use of drugs whose goal is to prevent a recurrence of a subsequent episode of anxiety in a patient who has demonstrated high chronicity, a high rate of recurrence, and/or significant levels of severity or disability associated with previous episodes.

The putative benefits of long-term (continuation or maintenance) therapy, insofar as they have been established, will be contrasted with the potential risks attendant upon such therapy. The abuse liability, dependence, and withdrawal risks associated with the benzodiazepines will be especially emphasized. Miscellaneous issues relating to long-term therapy, such as predictors of chronicity, safety and benefits of combination therapy (drug–drug or drug–psychotherapy), compliance, and intermittent medication strategies, will be reviewed next. The chapter will conclude with a review of methodologic issues in studying long-term treatment, suggest some promising research designs, and identify areas for future research.

Only belatedly has it been recognized that the anxiety disorders tend to be chronic and/or recurrent conditions. This recognition has not yet reshaped our basic treatment approach, which still focuses almost exclusively on the acute control of symptoms and only secondarily acknowledges that treatment may need to be either continued or reinitiated. Treatment planning for bipolar disorder, for schizophrenia, and, recently, for unipolar depression is premised on the chronicity of each illness, whether a maintenance or an intermittent drug treatment strategy is ultimately utilized.

But what is the evidence for chronicity for panic disorder (with or without agoraphobia), GAD, and social phobia? The evidence comes mostly from cross-sectional and retrospective assessments of duration of illness and, much less frequently, from prospective studies. These latter prospective studies are not long-term, but generally consist of patient samples followed naturalistically post treatment, in which case they provide evidence not only for the chronicity of the disorder, but also for the insufficiency of acute therapy alone.

In the ECA community survey (51), the median age of onset was 23, and the mean duration of panic disorder was 7.1 years for the subset of patients whose panic was in remission. The duration of panic in the majority (81%) who continued to report panic was apparently not computed, because the illness was still ongoing. Treatment studies find a similar mean duration of illness (ongoing at the time of evaluation) in the range of 5–12 years (4, 56, 59). The retrospectively assessed age of panic onset in these treatment studies was in the mid-twenties, comparable to the results obtained in the ECA community survey, suggesting that the reported duration of illness was long not simply due to a sampling bias with respect to the type of patient applying for treatment in a drug study. Overall, the course of illness of panic disorder appears to be chronic in the majority of patients, but with many reporting periods of remission lasting 6 months or longer.

For GAD, the ECA community survey reported a median age of onset in the early twenties (51). For the good outcome cases that had remitted, mean duration was 4.5 years, with 40% reporting durations of illness of longer than 5 years. For currently active cases of GAD, the mean duration of illness to date was reported as 8.5 years. Duration of illness for still-active cases of GAD entering both drug and psychotherapy treatment studies (45, 47) ranges from 5 to 15 years, confirming that GAD has high chronicity. Several researchers (5, 29) have noted that the clinical course of GAD is both more chronic and more unremitting than panic disorder. Noyes et al. (35!popup(ch127), in one of the few prospective studies (of anxiety neurosis), reported that 48% of patients continued to have moderate-to-marked symptoms at the 4- to 9-year follow-up.

Much less information is available about the course of illness of social phobia, but available data from both community (23, 54) and patient samples (19, 24) suggest an age of onset in the mid to late teens, with a chronicity that is equal to or greater than that of panic disorder, with mean duration of illness exceeding 5 years.

AGE OF ONSET: EVIDENCE FROM ADOLESCENT POPULATIONS

As has been mentioned, panic/agoraphobia appears to have a mean age of onset, estimated retrospectively, to be in the early to mid twenties, GAD in the early twenties, and social phobia in the mid teens. Because retrospective recollection is prone to bias, in the direction of either over- or underestimation, it would be a useful corrective to study anxiety directly in its early incarnations. This is especially true in light of the lack of reliable course of illness data from prospective community samples followed over time. Fortunately, recent research provides data on the onset of panic, social phobia, and GAD in contemporaneous child and adolescent populations. Structured, prospectively conducted DSM-III-R assessments of a random community sample of 1710 adolescents found a lifetime prevalence rate, to date, of 1.2% for panic disorder, 0.6% for agoraphobia, 1.5% for social phobia, 4.3% for separation anxiety disorder, and 1.2% for overanxious disorder (23). A similar community survey of adolescents (70) found prevalence rates of 0.6% for panic disorder and 3.7% for GAD. Other community prevalence estimates place the rate of panic attacks post puberty as high as 12–13%. The panic/agoraphobia, GAD, and social phobia prevalence rates for adolescents are in the range of rates reported for adults, suggesting that the age of onset and duration of illness estimates gleaned from retrospective reports in adults are, if anything, an underestimate.

THE ADEQUACY OF SHORT-TERM (ACUTE) TREATMENT

Accumulated evidence across multiple controlled trials suggests that the high-potency benzodiazepines (the most studied is alprazolam) are effective drugs for the short-term treatment of panic disorder, either with or without agoraphobia (e.g., see refs. 4, 12, and 58). Similar short-term treatment efficacy has been reported for the tricyclic antidepressants (12, 58), the MAOI antidepressants (59), and, in pilot studies, the SSRI antidepressants (8).

Only a few studies, though, have examined clinical and drug treatment status after naturalistic follow-up 1–3 years post acute study treatment. Studies reporting outcome after short-term benzodiazepine therapy find that many patients experience transient rebound anxiety symptoms to a level above their pretreatment baseline (38, 56). By 1–3 years, approximately 50% or more of acutely treated patients have had a recurrence (34), and at least 50% have resumed treatment, either with a benzodiazepine or with an antidepressant.

Acute treatment with antidepressants appears to have a similar clinical and drug treatment outcome at follow-up, with reports of relapse ranging from 25% to 70% (30, 36). Less data are available at follow-up for the MAOI antidepressants, but here again clinical experience suggests that relapse rates are comparable. The similar relapse rates after both antidepressant and benzodiazepine therapy suggest the need for continuation therapy. They also suggest that relapse is not simply an iatrogenic byproduct of benzodiazepine therapy.

We are not aware of any published research that reports reliable relapse/recurrence rates for GAD patients treated with antidepressants. With azapirones, however, Rickels and Schweizer (48) found GAD patients treated with buspirone for 6 months to have a lower relapse at 3 years post treatment than did patients treated with clorazepate.

A few studies have examined relapse/recurrence rates after acute benzodiazepine therapy. Rickels et al. (42) reported an 81% anxiety recurrence rate at 1 year for patients who had received 4 weeks of benzodiazepine therapy. In another study, Rickels et al. (44) treated 138 GAD patients for a minimum of 6 weeks with diazepam. A recurrence rate of 63% was reported at 1 year, with 50% of the patients sustaining improvement for at least 3 months. Overall, it appears that 60–80% of GAD patients (at least the ones applying for drug therapy) require additional treatment by 1 year.

Another index of the inadequacy of acute benzodiazepine treatment is the percent of patients who relapse in the week or two immediately after discontinuation of acute treatment. In a 4-week treatment study utilizing both a long-and a short-half-life benzodiazepine, 30% of patients had relapsed by 2 weeks (46). In fact, approximately 20% achieved HAM-A scores that were equal to or higher than their pretreatment baseline, indicating a rebound anxiety that, it has been speculated, might be an early precursor of the benzodiazepine withdrawal syndrome. Other studies have reported similar rates of rebound anxiety after short-term benzodiazepine therapy, ranging from 25% to 44% (17). It is unclear what percent of patients experience rebound anxiety as a transient and self-limited phenomenon, and in what percent it serves to trigger a recrudescence of the underlying generalized anxiety.

The published outcome research for the psychotherapy of GAD is hampered by small sample sizes and low-power, frequently inadequate inclusion/exclusion criteria and characterization of the study populations, as well as a variety of other methodological problems. Therapies that appear to have the most well-established efficacy are cognitive therapy (with various degrees of behavior therapy added in) and anxiety management therapies. Despite poor power, the tentative evidence to date consistently suggests some degree of acute efficacy for the psychotherapies (6, 9, 16, 40). Of special interest is that 3- to 12-month follow-up suggests that GAD treated by cognitive or anxiety management therapy may be unusually effective in sustaining improvement, with less than one-third of patients relapsing in the immediate 6- to 12-month follow-up period. If these results are confirmed, especially in drug versus psychotherapy trials, it would be a significant advance in the long-term management of GAD.

Much less treatment research has been conducted on social phobia. Treatment experience with MAOI antidepressants consists of at least two open-label studies (63, 64) and three controlled studies (19, 24), all of which suggest efficacy over 8–16 weeks of acute treatment. Only one of these studies (69) examined efficacy during continuation therapy for up to 1 year. This latter study found efficacy to be sustained. None of the studies examined rates of relapse or recurrence after MAOI discontinuation.

Clonazepam has also shown pilot evidence of efficacy, but only in open-label studies (13). Fluoxetine and buspirone have both also been used successfully in short-term pilot studies (55). Again, no data were reported for post-drug relapse or recurrence rates, so no conclusions can be drawn concerning whether there is a role for continuation therapy. Anecdotal experience, though, with all its shortcomings, does suggest, at least for the MAOIs and clonazepam, that relapse frequently occurs within 3–6 months of discontinuing short-term treatment.

It can be concluded with some degree of confidence (albeit with methodological caveats) that the evidence is fairly strong and very consistent and that the three anxiety disorders all have substantial chronicity. But to what extent is this chronicity associated with significant distress, disability, and impairment in functioning and quality of life? In the end, judgments about the benefits and risks of long-term anxiolytic therapy must be placed in this broader context.

The evidence regarding disability and quality of life is not substantial, but appears to be fairly consistent as far as it goes. For panic disorder, significant psychosocial impairment appears to be a common feature of the disorder. In the 1979 National Survey of Psychotherapeutic Drug Use, significant proportions of respondents with anxiety syndromes reported impaired role performance due to psychological and other factors: for panic/agoraphobia, 41% and 25%; for other phobias, 33% and 23%; for general anxiety, 22% and 45% (66). The rates for panic/agoraphobia were comparable to those for major depression in the same study: 47% and 31%. Substantial proportions of respondents with anxiety syndromes utilized treatment during the year prior to the survey: panic/agoraphobia, 62%; other phobias, 32%; general anxiety, 50%; compared to 26% of the general population (67).

The ECA survey reported that approximately 58% of their community sample of persons with panic disorder had some degree of financial disability, whether welfare, disability, or social security payments (28, 51). Massion et al. (29), reporting on 234 panic patients in an outpatient setting, found approximately 50% to be employed full time and found 27% to be receiving some form of public assistance. In the ECA community survey, 86% had sought medical outpatient evaluation and treatment in the previous 6 months (compared to 8% of persons without panic), and 51% had sought mental health evaluation and treatment. In keeping with these figures, 35% of the community ECA sample rated their physical health as fair to poor, and 38% rated their emotional health as fair to poor (28). In the ECA survey, 7% of panic patients without comorbid depression reported a history of suicide attempts, while Massion and colleagues (29) found a rate of 3%. In a large group of high utilizers of medical outpatient care, Katon et al. (20) found a 12% rate of current panic disorder and a 30% rate of lifetime panic. This overrepresentation of panic disorder is especially high in medical subspecialty clinics, where rates of panic range from 20% to 50% for medically unexplained episodes of chest pain, tinnitus, dizziness, irritable bowel, and chronic fatigue (7, 27).

For GAD, the best information currently available probably comes from the ECA community survey (51) and from a multisite survey of psychiatric outpatients (29). The overall picture for GAD is one of significant psychosocial impairment (albeit somewhat less severe than is observed with panic disorder), with 38% of ECA subjects and 71% of outpatients characterizing their emotional health as fair to poor; 27% and 25%, respectively, were receiving disability payments, and only about one-half worked full time. Of those who were working, 38% of the patient population had missed at least 1 week of work in the past year due to their anxiety. The ECA survey found a strong correlation between occupational status/income and GAD, with presence of GAD being associated with (a) a threefold greater likelihood of working at a low occupational level and (b) a more than twofold greater likelihood of earning less than $10,000 per year. Medically, 35% of the ECA sample with GAD, and 23% of the patient population of Massion et al. (29), reported that their physical health was fair to poor. This latter finding is corroborated in an outpatient medical setting by Katon et al. (20), who found that 24% of all high utilizers of medical outpatient care had a diagnosis of GAD.

The early onset and the high degree of chronicity, disability, and medical and psychiatric morbidity associated with GAD have led some researchers to speculate that GAD might not be an independent diagnosis, but instead a trait or vulnerability factor that predisposes to later problems. Whatever its status, the chronicity and disability associated with GAD make it an unlikely candidate for a cure or sustained remission after short-term drug therapy.

Quality-of-life issues and the degree of disability associated with social phobia have been much less well studied. Unfortunately, social phobia was not analyzed as a separate part of the ECA survey, so good community-based information is not available.

As can be seen, the anxiety disorders are not only chronic in nature, but are often accompanied by a wide range of moderate impairments in both psychosocial functioning and quality of life. The impairment and early onset of these disorders raise questions about what developmental distortions they may produce and what benefits early treatment may provide.

One of the most important issues in the long-term treatment of the anxiety disorders is the presence of comorbidity. Any long-term anxiolytic treatment strategy must take account of the high rates of comorbidity that appear to develop during the longitudinal course of panic, GAD, and social phobia—and that complicate these disorders, in fact, from their very inception. Comorbidity rates for adolescent anxiety disorders have been reported to range from approximately 20% to 60%, with higher rates recorded for patient samples that are under treatment (21).

It is well-established that comorbidity is significantly higher for patients seeking treatment than for persons not in treatment in the community. Nonetheless, even community surveys suggest a high rate of comorbidity, especially for GAD. In the ECA survey (51), 1-year prevalence estimates, reflecting current comorbidity, showed that approximately 25% of patients with a diagnosis of GAD also suffered from either panic or major depression, while an additional 30% (approximately) suffered from another Axis I diagnosis, leaving only about 40–45% of patients with "pure" GAD (at least from an Axis I standpoint, though the extent to which social phobia might have been an undetected source of comorbidity is unclear). This figure reports only 1-year prevalence rates for comorbidity, and as such is a likely underestimate because anxiety has been shown frequently to evolve into either depression or panic (10) over time.

Patient surveys show even higher rates of comorbidity, though the samples studied are obviously much smaller. Massion et al. (29) found that of 123 psychiatric outpatients diagnosed with GAD, 51% had a history of comorbid panic disorder. When these patients were edited out, 46% of the remaining patients had a history of comorbid major depression and 27% had social phobia. Overall, only 1.6% (2 of 123 patients in a psychiatric treatment setting) were felt to have "pure" GAD, with no history of comorbid depressive or anxiety disorders. Similarly, Brown and Barlow (10) have reported high comorbidity rates for outpatients with a principal diagnosis of GAD. Thirty-six percent suffered from concurrent panic disorder with or without agoraphobia, 29% suffered from concurrent social phobia, and 29% suffered from concurrent major depression or dysthymic disorder. In another study (37) of outpatients who had been recruited through the media, 34% had a history of major depression and 17% had a history of social phobia (a panic disorder diagnosis was a reason for exclusion).

The ECA community survey (51), as well as the Zurich cohort of Angst (2), found that 73% of patients with panic disorder had other comorbid conditions. Massion et al. (29), in their outpatient sample of 294, reported the following rates of comorbidity, depending on whether the patient was diagnosed with panic alone versus panic with agoraphobia: social phobia, 6% versus 15%; major depression, 27% versus 22%; OCD, 10% versus 7%. Twenty percent of panic patients suffered from comorbid GAD, but these patients were excluded from the previously reported comorbidity rates. Brown and Barlow (10), in a sample of 232 outpatients with a panic disorder diagnosis (both with and without agoraphobia), reported comorbid social phobia ranging from 6% to 36%, comorbid GAD ranging from 20% to 36%, and comorbid major depression ranging from 7% to 36%, largely depending on the presence and severity of the associated agoraphobia. In a sample of patients with panic disorder alone versus panic with agoraphobia who were recruited for a trial of psychotherapeutic medication, substantially higher rates of comorbidity were found; for example, social phobia, 26% versus 64%; GAD, 32% versus 68%; major depression, 26% versus 68% (61).

Comorbidity patterns for social phobia are less well-studied. The ECA community survey offers no data on social phobia. Brown and Barlow (10) reported the following comorbidity rates among 76 outpatients with a principal diagnosis of social phobia: panic disorder, 9%; GAD, 17%; major depression, 11%; and dysthymic disorder, 13%.

Substance abuse and Axis II personality disorders constitute two other important, but often neglected, categories of comorbidity which may complicate the natural history of the anxiety disorders and which must be taken into account when planning treatment interventions over time (39, 62). Studies of alcoholic patients suggest a range of 25–45% for comorbid anxiety disorders, with social phobia and panic/agoraphobia being the most prevalent (60). Conversely, prevalence rates of alcoholism in anxiety disorder populations range from 15% to 40% (29, 37, 61, 63). The ECA community survey found, not surprisingly, much lower rates of alcohol dependence/abuse. Van Ameringen et al. (68) have reported a 28% rate of alcoholism in a group of outpatient social phobics.

Comorbidity rates for substance abuse are lower for all three anxiety disorders, ranging from about 10% to 36% in patient samples (29, 37, 61). Again, these rates are much lower in the ECA community survey (51).

Multiple studies have reported rates of comorbidity for the anxiety disorders and Axis II disorders, most commonly the anxious "Cluster C" type. Axis II comorbidity rates for panic/agoraphobia generally range from about 30% to 60% (18, 37), with more severe agoraphobia being associated with increased rates (18). Axis II comorbidity rates for GAD have been reported in the same 30–60% range (37, 53).

As can be seen from the above review, comorbidity is the rule and not the exception in the clinical picture of the anxiety disorders. Long-term treatment planning cannot be undertaken except on the assumption that the majority of patients suffering from a principal anxiety disorder diagnosis also suffer from another disorder, or will suffer from one in the near future. This clinical reality makes it important to maintain an index of suspicion concerning the development of other disorders, and to carefully reevaluate the patient at intervals. This is especially true if patients fail to respond fully to a course of drug therapy or if an initial good response is lost. In both instances, unsuspected comorbidity is frequently the culprit.

The high rates of comorbidity among anxiety and depressive disorders also raise serious questions about the true independence of these disorders. An alternative way of conceptualizing comorbidity, with depression for example, is to view the two disorders as varying manifestations of one underlying diathesis—that is, two phenotypic expressions of a common underlying "genotype." Such a view has been proposed by Tyrer et al. (65), who speak of a "general neurotic syndrome." Recent genetic analyses of women suffering from major depression and GAD (22) provide some confirmation for this perspective, with results suggesting that the genetic vulnerability for both disorders is largely shared.

Previous sections of this chapter have established that the anxiety disorders are frequently chronic conditions associated with (a) a high degree of comorbidity and (b) impairment in both psychosocial functioning and quality of life. Furthermore, evidence has been presented that acute (4–12 weeks) drug therapy frequently results in early return of symptoms (relapse) or subsequent episodes (recurrence) when treatment is discontinued. What is the evidence, though, that continuation therapy is effective in preventing relapse, or that maintenance therapy is effective in preventing recurrences?

Unfortunately, there is little controlled research addressing these issues for panic/agoraphobia or GAD, and none for social phobia. The evidence for the efficacy of continuation and/or maintenance therapy of panic disorder is currently being studied in an NIMH-funded multicenter trial that compares imipramine to a form of cognitive-behavioral therapy.

Schweizer et al. (58) have conducted an 8-month, placebo-controlled study of continuation therapy for panic disorder with alprazolam and imipramine that found sustained efficacy for both compounds with no dose escalation, suggesting an absence of tolerance to the therapeutic effect. When drug was discontinued after 8 months of maintenance therapy, at 3 weeks post discontinuation, approximately 25% of patients had experienced a return of their panic attacks. At 1 year follow-up, 30% of patients had a recurrence of their panic, and 49% had restarted or never stopped their medication (50). Of course, the selection bias that was likely operating in the choice of patients for the initial study, as well as the naturalistic nature of the follow-up, makes this only a suggestive pilot study.

Preliminary evidence for the efficacy of continuation therapy of GAD comes from two studies (43, 47). In both studies the benzodiazepine therapy achieved sustained remission of anxious symptomatology with no tolerance and no dose escalation over a 6-month period. In the second study, buspirone also achieved a comparable sustained remission of anxious symptoms, though the significantly higher attrition rate by patients treated with buspirone complicates the interpretation of results. In both studies, about 25% of the patients experienced a return of their initial anxiety within 4 weeks of drug discontinuation.

There is no good published evidence for the sustained efficacy of either antidepressants or the benzodiazepines in the continuation or maintenance treatment of social phobia. The only published reports found both tranylcypromine (69) and clonazepam (13) to sustain efficacy over an 11- to 12-month period of treatment.

The 1979 National Survey of Psychotherapeutic Drug Use (67) indicates that a small fraction of patients with anxiety syndromes take psychotherapeutic medications for a year or more (panic/agoraphobia, 25%; other phobias, 6%; general anxiety, 6%). On the other hand, experts in the pharmacotherapy of these disorders, aware that they are chronic, severe, and disabling, generally recommend vigorous, long-term treatment. Here, as is so often the case, clinicians have had to use their best judgment in addressing clinical problems that lie beyond the scope of existing empirical evidence. Whether long-term drug treatment represents effective continuation or maintenance therapy, or patients continue to take medication because it elicits only a partial response, is uncertain. Two of the present authors have reported HAM-A scores of 16 in anxious patients with a mean of 6 years of benzodiazepine therapy, suggesting that long-term drug therapy, at least in this population, may have been continued even though (and perhaps because!) it was less than effective. The public health interest urgently requires clinical research to address these long-term issues of efficacy.

The efficacy obtained from drug therapy must always be weighed carefully against the potential adverse effects of the drug. This is especially true when it comes to long-term or maintenance drug therapy. What is considered tolerable, and even safe, during acute treatment may be neither of the above during chronic administration. This section will focus most intensively on the safety of long-term administration of the benzodiazepines. Not only have they been the object of intense scrutiny and concern, both publicly and scientifically, over the past 10–15 years (see ref. 71 for an excellent review), but they continue to be widely prescribed, sometimes for long-term therapy. The most recent survey data indicate that 1.6% of the adult population in 1980 took benzodiazepine for 12 months or longer in the previous reporting year (32). Some decline in long-term usage appears to have occurred since then, but it is still estimated that approximately 1% of the adult population has currently or recently received long-term benzodiazepine therapy.

Discussion of the safety of the benzodiazepines during chronic use can be usefully reviewed under two headings: (i) the psychological and behavioral effects and (ii) the medical or physiological effects.

A variety of psychological and behavioral effects have been attributed to the benzodiazepines when chronically administered. These include: persistent attentional, psychomotor, cognitive, and memory-impairing effects; abuse liability; physical dependence and withdrawal; post-withdrawal craving; and effects on coping and stress response capabilities. A brief review of each of these potential safety concerns is necessary in order to obtain a more complete picture of the benefit–risk equation regarding long-term therapy.

Substantial tolerance appears to develop to the attentional and psychomotor effects of benzodiazepines, beginning after the first few weeks of acute administration. The actual clinical significance (e.g., effects on driving ability) of subtle residual impairments, or the potentiation of these impairments by low levels of alcohol consumption that otherwise would pose no problem, is uncertain. Results of laboratory assessments, including driving simulation, have had very variable and contradictory results (see refs. 1 and 71 for reviews). But there appears to be some persistent attentional and psychomotor impairment that may be relevant to the execution of complex real-world tasks.

Preliminary research suggests that cognitive tasks and (especially) short-term memory tasks continue to be impaired even after long-term (5–10 years) daily administration of benzodiazepines (25), apparently exhibiting less tolerance than psychomotor function. The existence of differential rates of tolerance development is an intriguing phenomenon, and may depend on subtle regional differences in the monomeric constituents of the benzodiazepine receptor. The amnestic effect of benzodiazepines, which appears to be relatively resistant to tolerance in humans, probably has a hippocampal substrate. It should be cautioned that statistically "significant" amnestic findings ascertained, for example, using meaningless word lists may have little generalizability to life situations. Furthermore, since moderate-to-severe anxiety frequently has been noted to impair performance, the net effect of benzodiazepine treatment may be an enhancement of performance. Nonetheless, the subtle effects on learning and motivation of chronic, mild benzodiazepine-induced amnesia have not been well-studied.

A final behavioral effect for which there is no good evidence that tolerance does, or does not, develop with long-term benzodiazepine administration is irritability and hostility.

Benzodiazepines have a potential for recreational abuse (1, 71). But actual recreational abuse appears to occur principally in persons who abuse other drugs (see ref. 71 for review).

The abuse liability of the benzodiazepines in drug and alcohol abusers appears to stem in large part from marked individual differences in the euphoriant effects of benzodiazepines. Drug addicts, alcoholics, and even the nonalcoholic sons of alcoholics appear to be much more susceptible to the euphorigenic properties of the benzodiazepines (11). In most other populations, even among anxious persons (14), benzodiazepines do not appear to have much of a euphoric effect. In fact, the reinforcing properties of the benzodiazepines appear to be relatively low compared to every other drug of abuse (71). The reinforcing property of a drug appears to be an important behavioral correlate of abuse liability that tends to be fairly consistent across primate species. Even when one corrects for availability, diazepam and alprazolam are more widely abused in at-risk populations. It is unknown whether this reported effect is due to methodological problems with the adjustment for availability or to the rapidity of onset of action (lipophilicity, etc), or to differences in the intrinsic efficacy of the benzodiazepine at the receptor.

Another indicator of abuse liability that has been identified is drug liking or preference in normal human subjects. Here, again, the benzodiazepines appear to elicit much less drug liking than do traditional drugs of abuse (14, 31).

An ineluctable consequence of chronic benzodiazepine therapy is physical dependence, along with the likelihood of developing a withdrawal syndrome upon drug discontinuation. The clinical picture of benzodiazepine withdrawal is not always easy to distinguish from anxiety, though the temporal pattern of onset is of some help. Further confounding the clinical picture is that the experience of benzodiazepine withdrawal may serve, in turn, to trigger a recrudescence of the patient's underlying anxiety. Common symptoms of the benzodiazepine withdrawal syndrome consist of increased anxious mood and nervousness, insomnia, restlessness, tension, irritability, lethargy, nausea, depression, hyperacusis, and tinnitus. There appear to be no pathognomonic symptoms that are unique to the withdrawal syndrome and that do not commonly occur in anxiety disorders. After abrupt discontinuation of a short half-life benzodiazepine (e.g., alprazolam or lorazepam), withdrawal symptoms begin to appear within 6–12 hr, with a peak severity generally at 2–4 days. Symptoms usually subside within 1–3 weeks. After abrupt discontinuation of a long half-life benzodiazepine (e.g., diazepam or clorazepate), withdrawal symptoms begin to appear within 24–36 hr, peak at 4–7 days, and subside within 2–4 weeks. The benzodiazepine withdrawal syndrome, unless complicated by other medical or psychiatric illness or by other drug or alcohol problems, can be managed in an outpatient setting.

Several factors have been identified that appear to contribute to the development of benzodiazepine withdrawal and that add to its severity. The first of these is duration of therapy. Rebound anxiety (to a level above pretreatment baseline) has been observed after as little as 4 weeks of benzodiazepine treatment (46). Such transient rebound reactions are thought to be the prodromal signs of a developing pattern of dependence and withdrawal. By 3–4 months, physical dependence is likely to have been established, and a withdrawal syndrome upon benzodiazepine discontinuation can clearly be observed in a significant number of patients. There is no good evidence that longer durations of maintenance therapy beyond 12 months contributes to greater dependence and withdrawal liability (49, 57).

A second factor that contributes to dependence and withdrawal is the daily dose of the drug. It is important to note, though, that dependence and withdrawal do not require aggressive dosing. Patients treated with as low as 5–10 mg of diazepam or its equivalent have been noted to experience withdrawal reactions upon drug discontinuation. The clinical implications of the dose and duration of treatment factors is that only the most short-term treatment is without risk of physical dependence and withdrawal. The medical–legal implication of this situation is that information on dependence and withdrawal effects should probably be a routine part of the educational introduction of all patients to the benefits and risks of benzodiazepine therapy.

Several other factors contribute to the severity of benzodiazepine withdrawal. These include the rate of taper (obviously abrupt discontinuation poses a greater problem than a gradual taper over 4–8 weeks) and the presence of Axis I and II disorders. Residual anxiety, panic, and depressive symptoms also predict greater difficulty with withdrawal.

One unresolved issue is whether the benzodiazepine withdrawal syndrome may persist, in some form and in some patients, for many months, or even years. A subset of patients subjectively reported ongoing symptoms and complaints (3), and a "post-withdrawal syndrome" has even been proposed (64). The pharmacologic mechanism of this proposed state of persistent withdrawal is unclear, though a putative "receptor shift" has been suggested. It is, of course, difficult to disentangle persistent symptoms of withdrawal from a recrudescence of anxiety. The fact that the symptoms differ from a patient's previous experience of anxiety is no guarantee that an iatrogenic state has been triggered, because symptoms of anxiety can change over time. It is also of interest that two of the authors (Rickels and Schweizer) have noted an absence of any reports of persistent withdrawal states in abstinent patients with a history of benzodiazepine dependence, but in whom benzodiazepines were not prescribed for anxiety. In most cases, such post-benzodiazepine discontinuation symptoms are most likely related to the patient's Axis I and Axis II disorders. Still, it is a topic of concern, and deserves further research, whether chronic administration of benzodiazepines may in some way permanently alter the benzodiazepine-GABAergic system.

It is an almost universal feature of drugs of abuse that they intermittently evoke craving in the user long after the withdrawal syndrome has subsided. Alcohol, tobacco, cocaine, amphetamines, opiates, and even coffee engender cravings in the months and years after drug discontinuation. Preliminary research (26) suggests that there is little or no craving associated with long-term dependence on benzodiazepines. Certainly former benzodiazepine-dependent patients do restart their benzodiazepine, but resumption of benzodiazepine use is almost exclusively caused by a recurrence of symptoms of anxiety. Naturalistic follow-up studies of chronic benzodiazepine users find a clear trend over time toward lower daily doses, and eventual discontinuation. The pharmacoepidemiological data (32) also indicate that the vast majority of regular daily benzodiazepine use continues less than a year (85%) and usually less than a month (67%).

The risk of adverse medical or physiological effects from chronic benzodiazepine therapy has not been systematically studied. Medical evaluations of over 400 benzodiazepine-dependent patients (mean duration of use: >5 years), which included blood chemistries, complete blood count with differential, urinalysis, and electrocardiogram, have yielded no obvious abnormalities (Rickels et al., personal communication). A question has been raised concerning whether chronic benzodiazepine treatment might be associated with increased ventricular–brain ratios on computerized tomography scans. Several studies have yielded conflicting results on this subject, but because a variety of other comorbid medical factors, particularly alcohol use, were not excluded, the issue remains unresolved (1).

Several other physiological effects appear to result from chronic benzodiazepine treatment. Preliminary data suggest that chronic treatment alters the functional status of one or more components of the benzodiazepine– GABAergic receptor complex (33). One of the hypothesized mechanisms for benzodiazepine withdrawal suggests that the symptoms can be explained by the "overdrive" of downstream noradrenergic, serotonergic, and cholinergic receptors that have been released from inhibitory control (that had been previously augmented by the exogenously administered benzodiazepine agonists).

Parenthetically, clinical anxiety may be due to a subsensitivity of the benzodiazepine receptors (52). If additional data establish that a functional deficit in benzodiazepine receptors is one of the pathological substrates of anxiety, then the rationale for maintenance therapy with benzodiazepines would be greatly strengthened.

Long-term safety assessments of the tricyclic antidepressants, and of the newer SSRI antidepressants, has long been a routine part of the drug development and FDA approval process. Most safety research on antidepressants has been conducted on depressed populations, but from a behavioral and medical standpoint one might tentatively extrapolate to a population of patients suffering from anxiety disorders. Safety and benefit–risk issues relating to antidepressants as a class are discussed elsewhere in this volume (see In Vivo Structural Brain Assessment, Methodological Issues in the Neuropathology of Mental Illness, and Psychiatric Molecular Genetics). Noyes et al. (36) have reported safety data on long-term use of tricyclic antidepressants in a population of panic patients followed-up naturalistically for 1–4 years. They found that 35% of these patients were unable to tolerate tricyclic antidepressants (TCAs) and discontinued them. Early discontinuations (in the first 6 weeks) were most commonly for overstimulation, orthostatic reactions, and allergic reactions. Late discontinuations were most commonly for weight gain and persistent anticholinergic effects. Forty percent of long-term patients reported weight gain, with the mean weight gained being 22 pounds.

Abrupt discontinuation of long-term antidepressants (TCAs, MAOIs, and short half-life SSRIs) has also been reported to result in a withdrawal reaction (15). The exact prevalence, or consequences, of such withdrawal reactions has not been systematically studied for panic disorder patients, though it appears to be correlated with the rapidity of the taper schedule. For TCAs, the withdrawal reaction may be related to cholinergic rebound, with common symptoms consisting of insomnia, nausea, headache, and tremor. For the MAOIs, vivid dreams associated with rapid eye movement (REM) rebound are also observed. The extent to which these withdrawal reactions trigger a recrudescence of panic is uncertain.

Buspirone appears to be safe in maintenance therapy (41, 47). Unlike the TCAs, no significant side effects were found to emerge with long-term therapy. In general, buspirone was well-tolerated during maintenance treatment. No withdrawal reaction has been reported to occur upon discontinuation.

Surveys of prescribing practices suggest that perhaps one-quarter to one-third of anxiolytic drug therapy consists of treatment with combinations of drugs. The addition of a second or third drug may be clinically indicated because of the presence of a comorbid condition—most commonly either another anxiety disorder or an affective illness. But virtually no controlled research exists concerning the effectiveness of long-term drug treatment for comorbid conditions with one, not to mention two, drugs.

Again, it is common clinical practice to combine anxiolytic drug therapy with some form of psychotherapy. Yet there are virtually no controlled studies that establish the clinical indications or additive efficacy of such treatment. Nor can it be confidently predicted that the addition of psychotherapy to a drug regimen, or vice versa, will result in enhanced efficacy. In fact, experienced cognitive therapy researchers (A. T. Beck, personal communication, May 1993) believe that the efficacy of cognitive therapy in panic disorder is reduced by the concomitant administration of benzodiazepines. The reason for this, if it is confirmed, is uncertain. Perhaps the cognitive effects of benzodiazepines hamper psychotherapy, or perhaps learning under the influence of a benzodiazepine does not carry over to the drug-free state (state-dependent learning).

The high chronicity and/or recurrence rate of the major anxiety disorders suggest that studies of the safety and efficacy of maintenance therapy should be integral to the assessment of any drug or psychotherapy. Acute symptom reduction is important to achieve, but is a hollow victory over the distress and disability associated with clinical anxiety if it cannot be sustained.

Research questions, largely unanswered, concerning maintenance treatment of GAD, social phobia, and panic disorder are legion and include the following: What constitutes an adequate course of maintenance drug therapy? What are the predictors (e.g., duration, severity, rate of recurrence, comorbidity) that suggest the need for further continuation therapy? What are optimal doses for both maintenance and continuation therapy? When, if ever, are intermittent medication strategies indicated as opposed to maintenance or continuation therapy? Does any further clinical improvement accrue from maintenance therapy compared to what is achieved by acute therapy? Is there any prophylactic benefit, in terms of long-term outcome, from continuation therapy? What is the safety of continuation drug therapy—including withdrawal, rebound, and other behavioral effects? Does tolerance develop during the course of maintenance or continuation drug therapy, or is efficacy maintained? What is the optimal way to discontinue maintenance or continuation drug therapy? How should emergent comorbidity be optimally managed? When are combination drug therapies indicated, and what are the safety and efficacy of such combination therapies? What is the comparative safety and long-term efficacy of drug versus psychotherapy? Can response predictors be identified for choosing drug versus psychotherapy? What are the clinical indications, safety, and efficacy of combined drug and psychotherapy?

All of these questions can, and should, be asked, not only for each anxiety diagnosis, but for each treatment modality: benzodiazepine, TCA, MAOI, SSRI, azapirones, cognitive therapy, dynamic therapy, and so on. The investigation of the long-term treatment and outcome of the anxiety disorders is a difficult area of research. It is afflicted with its own methodological problems that daunt even the most seasoned researcher who dares to enter this largely uncharted domain. These problems include: issues of sample size; patient retention over time; the influence of retention strategies on outcome; how to handle a host of potentially significant intervening variables (e.g., non-study medication usage, life events, etc.) that cannot be simply excluded as they can in acute treatment studies; how and when to randomize for maintenance therapy; how to handle emergent comorbidity; how to handle taper and discontinuation of study medication; how to undertake long-term follow-up post study; and how to determine appropriate outcome measures.

Preparation of this chapter was supported, in part, by USPHS Research Grant MHO-8957.

published 2000