Treatment-Resistant Depression

Michael E. Thase and A. John Rush

An essential characteristic of mood disorders is their relatively favorable prognosis. Nevertheless, only 60% to 70% of patients who can tolerate an antidepressant medication will respond to the drug of first choice, and 5% to 10% remain depressed despite multiple interventions (66, 104). The poorly responsive group has been variously described as resistant, refractory, or intractable. This chapter reviews the literature on the definition, assessment, and treatment of treatment-resistant depressions (TRD).

Table 1 summarizes common definitions relevant to the concept of treatment resistance. The definition of an adequate treatment trial of antidepressant medication has varied widely over the years, as have the corresponding definitions of treatment resistance. In actuality, TRD patients present with histories of varying degrees of treatment adequacy. A high proportion of cases referred to university settings specifically for evaluation and treatment of "refractory" depressions have not received even a single adequate antidepressant trial (9, 91).

There is no absolutely correct dosage for a specific antidepressant, since dosage requirements vary depending on factors such as age, weight, general health, concomitant medication usage, and tolerance of a particular medication. Confirmation of treatment adequacy by more objective means (e.g., serial plasma drug levels) is not the rule in clinical practice, and valid plasma level–response relationships are limited to only a subgroup of the tricyclic antidepressants (TCAs) and lithium salts. With respect to psychotherapy, adequacy of treatment may depend on the number of sessions, the expertise of the practitioner, the therapist's adherence to a particular form of therapy, and/or the interaction of the patient–therapist dyad. Electroconvulsive therapy (ECT) may be gauged by the total number of treatments, the use of bilateral electrode placement, and the verification of seizure time by electroencephalographic monitoring. Therefore, the terms "relative" and "absolute" treatment resistance may best describe lesser and greater degrees of certainty about the adequacy of a specific treatment trial (104).

Similarly substantial variability exists as to the definition of an acceptable treatment response (109). The most common response criteria in clinical trials are a rating of at least "much improved" on the Clinical Global Impressions (CGI) scale, a prespecified level of improvement on a depression symptom rating scale (e.g., >50% reduction in Hamilton Depression Rating Scale scores), a final absolute score on a symptom measure (e.g., a Beck Depression inventory score of {ewc MVIMG, MVIMAGE,!lesseq.bmp}9), or some combination of the above. Both the use of composite outcome criteria and documentation of persistent improvement (e.g., for 2 weeks or longer) may improve reliability and validity of classification (109).

At least a 50% reduction in depressive symptom severity generally corresponds to the clinician's global clinical impression of a moderate level of improvement (104). However, some patients meeting this commonly used response definition continue to have considerable residual symptomatology. Residual symptoms convey a higher risk of relapse during continuation treatment and likely contribute to suboptimal restoration of vocational or interpersonal functioning (104). Therefore, complete symptom remission is the desired outcome of acute treatment. The term remission describes a response in which a formerly depressed person's level of residual symptomatology is essentially indistinguishable from someone who has never been depressed. With respect to standardized scales, a score of 6 or less on the 17-item Hamilton Rating Scale for Depression is often used to define a remission (109).

The concept of relative treatment resistance (104) refers to a patient who has received at least an average dose of a specific class of antidepressant for a minimally adequate period of time. For example, a relative TCA resistant depression would describe a patient who failed to respond to a typically adequate dose of imipramine, such as 200 mg/day or its equivalent, for at least 4 weeks (66, 104). Similarly, relative resistance to a selective serotonin reuptake inhibitor (SRI) could be described as failure to benefit from a comparable trial of fluoxetine (20 mg/day).

Relative resistance recognizes that some patients would respond if they were treated with higher dosages and/or longer treatment periods. By contrast, absolute resistance refers to a patient failing to respond to a maximal, nontoxic dose of a given antidepressant, with confirmed compliance over an extended treatment period (e.g., an 8-week trial of imipramine at 300 mg per day and therapeutic blood levels or an 8-week trial of sertraline at 200 mg per day). Unfortunately, drug plasma levels are only of value for imipramine, desipramine, and nortriptyline (23).

The words resistance, refractory, and intractable are sometimes used interchangeably in the literature (66, 104). Such definitional imprecision probably does the field little good. A more narrow application of the term treatment-resistant depression (as it pertains to failure to respond to an adequate trial of a specific antidepressant treatment) allows for a more precise description of each patient's past treatment history. For example, patients can be staged according to the number and classes of antidepressants that have failed to produce a response. Staging typically moves from more common (e.g., TCAs and SRIs) to less common [e.g., monoamine oxidase inhibitors (MAOIs) or ECT] treatments (see, for example, Table 2).

Staging of the treatment responsiveness of patients previously treated by other practitioners can be difficult, since both patient recollection and medical records often provide insufficient documentation of previous treatment trials. Several investigators have developed interviews or standardized rating scales to try to improve this process (29, 64, 89). However, the reliability of such measures has not been documented, and the major obstacle continues to be the source of the information not the means to elicit and record the data.

The psychobiological heterogeneity of major depressive disorder is well recognized and may be mirrored by heterogeneity of antidepressant response. To some extent, such heterogeneity can be attributed to sociodemographic factors, such as patient age and gender (104). Elderly patients, for example, may be somewhat less treatment responsive than those at midlife. Conversely, younger women may benefit less from TCAs than men or women treated with MAOIs (17, 104). Individuals with fewer interpersonal or economic resources also may be less responsive to treatment (104). Perhaps these poorer outcomes reflect a higher level of objective stress, poorer social supports, and/or a greater risk of noncompliance.

Other correlates of response include symptomatic and diagnostic variables. One of the best studied symptomatic correlates is global severity (10, 104). For example, the probability of placebo response appears to decrease as initial symptom severity increases. Although pharmacotherapy response rates may also decline in relation to more marked levels of severity, a significant drug-placebo difference remains even in the most markedly severe cases (104).

The broad grouping of major depressive disorders includes a number of clinical subtypes that show greater or lesser degrees of responsivity to different classes of antidepressants. For example, optimum response (particularly with respect to the drug-placebo difference) to TCAs appears to occur in unipolar (nonbipolar, nonpsychotic) major depressions of moderate severity (31, 104). Although less well established, the proportional difference between active drug and placebo (PBO) may be highest when patients manifest classic melancholic features (i.e., psychomotor retardation or early morning awakening) (104). Some evidence exists indicating poorer response to TCAs in the following subforms of major depressive disorder: atypical depressions (including patients manifesting reversed vegetative signs and/or anxiety features such as phobia, panic attacks, or obsessions), bipolar depressions, psychotic depressions, and depressions associated with significant Axis I, Axis II, or Axis III comorbidity (104). In fact, patients with severe, concomitant Axis I, Axis II, or Axis III illnesses are typically excluded from contemporary efficacy trials, which greatly limits the generalizability of their findings to clinical practice. We believe that PBO-controlled trials of more complicated patients should be included within phase III and phase IV efficacy studies, as should trials of patients with a confirmed history of nonresponse to a standard TCA or SRI.

Controlled studies of the efficacy of alternative therapies for TRD are difficult to conduct for both pragmatic and ethical reasons. Consequently, controlled studies of TRD account for less than 1% of publications on the treatment of depression since 1959 (109).

Efforts to identify and sustain a large enrollment of patients in a study of treatment(s) for TRD are administratively and financially daunting. If, for example, cell sizes of 30 patients per condition are necessary to detect a clinically meaningful difference between two types of treatment for TRD, the investigator must recruit and treat as many as 300 patients in order to obtain 60 with a documented failure on the initial treatment (e.g., assuming an 80% completion rate and a 40% treatment nonresponse rate). Formation of multicenter collaborative study groups or the creation of regional data bases to pool the experience of a number of clinical investigators might help to remedy these logistical problems.

Other difficulties in designing studies of TRD result from the need to include either a standard comparator treatment and a placebo-expectancy control group. Although ECT may be thought of as the standard of efficacy for TRD (40), its use is largely limited to inpatients, and biased sampling may result if too many TRD patients decline to be in a randomized clinical trial utilizing ECT as one option. As discussed in subsequent sections of this review, augmentation of an ineffective agent with either lithium or thyroid or alternate treatment with an MAOI could be considered as possible comparators for outpatient studies of TRD.

Unlike uncomplicated new cases of depression, which typically have a 25% to 40% placebo response rate (23), most patients with TRD are relatively unresponsive to placebo (i.e., rates of 0% to 10%) (109). Thus, placebo control may be considered unnecessary in comparative studies of TRD, particularly if treatment resistance is established prospectively. The elimination of the placebo treatment condition might enhance the desirability of a study to eligible patients, facilitating research recruitment, and, as a result, improving generalizability of results. However, because response rates to active treatments may be as low as 25% to 35% in randomized controlled trials (RCT) of TRD (66, 109), use of a placebo control provides the best opportunity to determine the statistical significance of a novel treatment in the context of such modest outcomes.

Alternatives to this dilemma include randomization strategies such as "play the winner" or the use of high-versus-low doses of a novel treatment (109). Another option is for patients to be randomly assigned to either the novel strategy or continued treatment with the ineffective agent currently received. The latter design, which provides a more ethically acceptable variation of the waiting list control design for TRD (109), controls for the passage of time and does not deprive patients of receiving some form of ongoing treatment. Staged criteria for level of treatment resistance afford another possibility to obviate placebo control. For example, for patients with a confirmed history of nonresponse to a TCA or SRI, the novel treatment strategy may be contrasted against a new trial of an alternate member of a previously ineffective drug class (66, 109).

The simplest strategy for nonresponse is to extend the initially ineffective treatment trial for another 2 to 4 weeks (109). Advantages of longer medication trials have been summarized by Quitkin et al. (53). This strategy capitalizes on the natural history of episodic depressions to remit over time and counteracts the tendency of some to discontinue the antidepressant prematurely. It also helps distinguish an enduring "true" antidepressant response from a more transient placebo response. Specifically, placebo responders have a greater likelihood of relapse between weeks 6 and 12 than patients who have responded to an active antidepressants (53).

The value of extending an ineffective, yet optimized, antidepressant beyond the sixth week of treatment has not been established for most agents. Indeed, in controlled trials comparing standard TCAs with various forms of psychotherapy, most of the variance associated with antidepressant response occurs during the first 4 to 6 weeks of treatment (23). In the Pittsburgh group's study of recurrent depression, a group of slower responders was identified who typically required 10 to 16 weeks of treatment to remit (33). This group, about one-third of the sample, was characterized by partial response between weeks 4 and 8 with a considerable degree of symptomatic fluctuations. These patients were characterized by higher levels of personality pathology and life stress, and somewhat lower levels of neurobiologic disturbance (33).

In sum, empirical evidence supports an extension of an optimized medication trial from the previous standard of 4 weeks up to 6 weeks. Past week 6, however, there is little empirical basis for this strategy, unless patients have shown at least a partial response or are receiving concomitant psychotherapy. An exception to this suggestion may be treatment with fluoxetine, which has been associated with continued evolution of treatment response for at least 8 weeks of treatment (94).

Underdosage of the antidepressant is a common cause of treatment failure, although psychiatrists may be less likely to underdose than 10 or 20 years ago. An older literature suggests that routine prescription of maximal dosages of TCAs or MAOIs is associated with a greater likelihood of response than more modest dosages. Case reports similarly suggest response to megadosages of TCAs, SRIs, or MAOIs in some individuals unresponsive to typical maximum dosages (109). It is likely that the strategy to use higher dosages of antidepressants will prove more fruitful for agents with linear dose–response pharmacokinetics. For example, approximately 5% to 10% of patients will have subtherapeutic TCA plasma levels despite compliance with conventionally maximal dosages (e.g., 300 mg/day of imipramine or 200 mg/day of nortriptyline) (109). In such cases, a supranormal dosage is necessary to achieve a blood level within the therapeutic range.

Treatment with nortriptyline has an apparent therapeutic window for blood levels. Although not as extensively documented as may be desired, the therapeutic-window phenomenon has been illustrated in several patients who, while nonresponsive at higher blood levels (e.g., >200 mg/ml), responded when blood levels were reduced to within the proposed therapeutic range (i.e., 50 to 150 ng/ml) (53). A therapeutic window has not yet been clearly shown for the SRIs, though anecdotal reports suggest such a possibility (12). The SRIs pose additional challenges to pharmacotherapists trying to implement rational treatment plans based on the dose–response behavior of this class of drugs. For example, initial randomization to either low dosages (e.g., 20 mg/day of fluoxetine or paroxetine or 50 mg/day of sertraline) or high dosages (e.g., 60 mg/day of fluoxetine, 50 mg/day of paroxetine, or 200 mg/day of sertraline) appear to yield comparable responses over 4 to 6 weeks of pharmacotherapy (1, 26, 116). In a study of patients not responsive to fluoxetine at 20 mg per day for 3 weeks (94), patients were randomly assigned to either an additional 8 weeks of continued therapy at 20 mg/day or an increased dosage of fluoxetine (60 mg/day). Both groups had similar outcomes at week 8, failing to identify any advantage for the higher dose. Further research on the efficacy of higher dosages of SRIs (after failure to respond to lower doses) is essential.

The dose–response relationships for trazodone and bupropion also are not well established. Both agents were initially tested in relatively high dosages (i.e., up to 600 mg/day) when compared to the dosages now typically prescribed. Much evidence of the efficacy of these drugs in severe depressive states was derived from studies at higher doses. However, concern about dose-dependent side effects now largely limits their prescription to dosages in the 200–450 mg/day range. Evidence of the relationship of plasma level to clinical response has not been established for either drug. In the absence of controlled data, a case for an 8-week, two-phase trial can be made for both trazodone and bupropion, in which 4 weeks of treatment at low-to-moderate doses is followed, when necessary, by 4 weeks of treatment at higher-to-maximal dosages.

A time-honored dictum in medicine is that a combination of two drugs should not be used if one drug will suffice. Accordingly, the most common pharmacological approach to TRD over the past 30 years has been to discontinue the ineffective agent and substitute an alternate antidepressant (109). Prior to 1980, clinicians in the United States were limited to substitutions within the first generation of antidepressants. Given the unfortunate reluctance to prescribe MAOIs that characterized the era, most psychiatrists would first opt to switch from one TCA to another. For example, a secondary amine TCA, such as desipramine, often would be prescribed when a relatively dissimilar tertiary amine TCA, such as amitriptyline, was ineffective. The introduction of a large number of novel antidepressant compounds over the past 15 years has broadened therapeutic options substantially.

The evidence tables used in the following sections were originally prepared for use by the Agency for Health Care Policy and Research (AHCPR) for development of guidelines for treatment of depression in primary care (23).

Despite the apparent frequency of substitution of one TCA for another in clinical practice, only two controlled studies of this strategy have been published (see Table 3). Both studies document TCA response rates of only 10% to 30% in patients with a past history of TCA nonresponse. Preliminary evidence from an ongoing trial employing a plasma level-guided course of nortriptyline similarly suggests a 30% response in patients with a prior history of TCA failure (Nierenberg, personal communication).

Such low response rates are particularly unimpressive when compared to the outcomes reported in studies of patients treated by switching to alternate drug classes (see below). Moreover, the theoretical rationale that originally guided this practice, namely, that depressions might be subdivided on the basis of the primacy of noradrenergic or serotonergic dysfunction and matched to an appropriate TCA (e.g., amitriptyline for serotonergic depressions and desipramine for noradrenergic depressions), has not been supported by the available data. This is perhaps in part because the more serotonergic tertiary amine TCAs, amitriptyline and imipramine, are readily converted in vivo to noradrenergically active metabolites, nortriptyline and desipramine.

The tertiary amine TCA, clomipramine (CMI) is the most potent reuptake inhibitor of serotonin among this class of antidepressants, and CMI's active metabolite, desmethylclomipramine, is also a potent noradrenergic reuptake blocker. Generally, CMI is not widely used as an antidepressant in the United States, in part because of the decision of its manufacturer to obtain approval from the Food and Drug Administration (FDA) as an antiobsessional agent, but not as an antidepressant. Nevertheless, CMI is an effective primary antidepressant and, in addition to antiobsessional and antidepressant effects, CMI has broad anxiolytic and antipanic coverage. Also, CMI is a potent suppressor of REM sleep and it can be safely administered parenterally (110). Thus, CMI might represent a logical first choice in a study or treatment sequence in which TCA nonresponse is confirmed prospectively. However, it has not been established that the CMI response rate in patients with tricyclic TRD would exceed the 30% ceiling observed in controlled studies utilizing other TCAs.

The first generation TCAs and the second-generation heterocyclic antidepressants (HCAs) (e.g., trazodone, maprotiline, amoxapine, nomifensine, and bupropion) differ in both chemical structures and side-effect profiles. Therefore, it is reasonable to expect that a second-generation antidepressant may prove effective in some patients who either do not respond to or cannot tolerate adequate or maximal doses of TCA. The psychiatric literature from the early 1980s is replete with case reports of patients who responded to newer antidepressants after failure to respond to one or more TCAs (109).

As summarized in Table 3, the efficacy of bupropion treatment of tricyclic TRD has been examined by Stern et al. (100) in a report including both a double-blind, placebo-controlled inpatient study (n = 30) and an open-label outpatient trial of 56 patients intolerant or nonresponsive to TCAs. The findings of both controlled and open-label protocols indicate that bupropion is an effective treatment of tricyclic TRD, although response rates were not reported. A 56% response rate to trazodone was observed by Cole et al. (14) in an open-label study of a diverse group of 25 TCA nonresponders. However, only about half of these patients had clearly robust responses. Nomifensine, an approved HCA withdrawn from the market nearly 10 years ago because of toxicity concerns, has been studied by two groups. Nolen et al. (69) found only a 10% response rate in TRD patients resistant to both TCAs and SRIs. Schmauss et al. (92) similarly found intravenous infusions of nomifensine to be relatively ineffective in TRD (10%), although a slightly higher response rate (30%) was observed with oral administration. Although fondly remembered by some, the absence of nomifensine does not appear to be a marked loss for most TRD patients.

No controlled data have examined amoxapine or maprotiline in TCA failures. Alprazolam, a potent triazolobenzodiazepine anxiolytic with provisional efficacy in anxious depressed patients, also has not been studied in TCA failures. With respect to other HCAs not available in the United States, Nolen et al. (68) found that the relatively selective noradrenergic reuptake inhibitor oxaprotiline had statistically significant symptomatic effects in 33 TRD patients. However, the response rate to oxaprotiline was only 27% (9/33), placing its effectiveness in tricyclic TRD in a range comparable to currently available TCAs.

In summary, the studies grouped on Table 3 indicate that a 20% to 50% response rate may be expected when a TCA nonresponder is crossed-over to a second generation HCA. However, this conclusion rests on only a modest number of studies, of which the largest effects were found with bupropion and trazodone.

Several small (ns {ewc MVIMG, MVIMAGE,!lesseq.bmp} 10) double-blind, crossover studies of the prototypic SRI, zimelidine, showed some degree of efficacy (i.e., 25% to 75% response rates) in patients who had failed to respond to trials of desipramine or maprotiline (109). Following the introduction of fluoxetine to the United States in 1988, a spate of published case reports emerged attesting to this SRI's effectiveness for TCA nonresponders (109). Studies of SRI treatment of TCA failures are summarized on Table 3. The utility of fluoxetine in patients with a history of TCA nonresponse subsequently has been confirmed in two outpatient studies, with response rates of 43% and 51% observed in these trials. Peselow et al. (74) similarly reported that paroxetine was significantly more effective than placebo in IMI nonresponders, with a 50% response rate to the SRI. Most recently, Gagiano et al. (34) conducted a 6-week, open-label study of paroxetine in 28 patients with a history of TCA nonresponse, including 15 patients who had failed to respond to amitriptyline (150 mg/day) in a controlled clinical trial. Eighteen patients (64%) responded to paroxetine treatment. No controlled data are yet available for sertraline or nefazodone treatment of tricyclic TRD.

Fluvoxamine, like clomipramine, has been approved by the FDA for treatment of obsessive–compulsive disorder; its manufacturer apparently will not seek an indication for major depression even though this drug is widely used for this purpose in Canada and Europe. Three inpatient studies of fluvoxamine in tricyclic TRD have yielded disappointing results. Fluvoxamine had response rates of 17% (1 of 6) and 14% (3 of 21) in two small inpatient studies of patients with a history of TCA nonresponse. Moreover, Nolen et al. (69) found almost no evidence of efficacy in a larger study of TRD inpatients: only 2 of 56 patients (4%) responded to a 4 week trial. Outpatient trials of fluvoxamine have yielded substantially more promising results. Delgado et al. (20) reported a 71% (5 of 7) response rate in an open-label study of outpatient desipramine nonresponders. White et al. (117) reported a 75% response rate (9 of 12) in a double-blind, crossover study of desipramine failures. The pooled outpatient response rate to fluvoxamine across outpatient studies (74%; 14 of 19) is both clinically and significantly greater than observed in inpatients (7%; 6 of 83) (p < 0.0001). Given the magnitude of this difference, it is important to note that all of the evidence pertaining to the effectiveness of other SRIs are crossover treatments derived from outpatient studies. With respect to investigational SRIs, Faravelli et al. (28) reported that viqualine was superior to placebo (5/10 versus 0/10) in a small study of outpatients with TRD.

In summary, the studies listed on Table 3 indicate that SRI treatment of tricyclic TRD typically yields 30% to 70% outpatient response rates, with fluvoxamine appearing to be an ineffective strategy for hospitalized TRD cases.

American psychiatrists are now much more likely to initiate outpatient treatment with an SRI than a TCA. Therefore, it is important to establish the efficacy of TCA in SRI-treated TRD. No studies were identified of TCA treatment of fluoxetine, sertraline, fluvoxamine, or citalopram nonresponders. Several small crossover studies of desipramine or maprotiline in patients resistant to zimelidine suggest a TCA response rate of approximately 50%. Peselow et al. (74) studied imipramine in a double-blind trial of 15 paroxetine nonresponders, in which 11 (73%) responded to 6 weeks of IMI therapy. Although not yet extensively studied, it seems reasonable to anticipate that the TCAs will prove to be a strategy of first choice for SRI nonresponders. Comparable data for bupropion and venlafaxine treatment of SRI failures are needed.

The proliferation of different SRIs begs the question of their interchangeability. Should patients nonresponsive to one SRI receive a trial of another or should they be routinely switched to another class of agents? As of yet, there are no controlled studies to address this important question. Clinical experience suggests that intolerance to one SRI does not necessarily convey intolerance to the whole class. A large prospective study by Brown and Harrison (11) found that 91% (85 of 93) patients intolerant to fluoxetine were able to complete an adequate trial of sertraline (50 to 200 mg/day), of which 69 (76%) responded. There are, as of yet, no controlled data to address the equally important question about the effectiveness of an alternate SRI when another member of this class is not effective.

Two uncontrolled studies on the addition of fluoxetine (20 to 40 mg/day) to an adequate, but ineffective, dose of a TCA reported improvement rates on the order of 80% (39, 72). Although the failure of fluoxetine monotherapy was not established in these open-label studies prior to the use of the combined strategy, the magnitude of response reported certainly surpasses those observed in the crossover studies of the SRIs reviewed earlier. Studies of the effectiveness of sertraline or paroxetine in combination with a TCA have yet to be published. Controlled studies are needed to confirm the effectiveness of this now-popular strategy.

It is not clear if the apparently robust effects of SRI–TCA cotherapy represent pharmacodynamic synergy or, more simply, a pharmacokinetic interaction. A study contrasting TCA cotherapy with sertraline, which has a lesser effect on TCA metabolism, versus fluoxetine or paroxetine might help to elucidate the mechanism as pharmacodynamic or pharmacokinetic, especially if TCA blood levels are carefully monitored.

Price et al. (50) studied the efficacy of an alternate serotonergic cotherapy strategy by adding the direct agonist fenfluramine (40 to 120 mg/day) to ongoing desipramine therapy in 15 patients with tricyclic TRD. In contrast to the positive reports regarding the use of the fluoxetine–TCA combination, fenfluramine cotherapy was virtually useless in this small study (1/15; 8% response).

One of the oldest augmentation strategies is the addition of small doses of thyroid hormone [e.g., 25 to 50 mg of L-triiodothyronine (T3)] to the ineffective antidepressant agent. The theoretical underpinnings for this practice have been reviewed extensively (78, 96). Thyroid augmentation can be traced to several early reports in which simultaneous initiation of TCA treatment and concomitant thyroid hormone was found to reduce the time to response in depressed women (see ref. 96). Of note, thyroid supplementation has not been demonstrated convincingly to speed response in depressed men (78). Suggested mechanisms include potentiation of effects on noradrenergic receptor sensitivity, increased efficiency of nonadrenergic neurotransmission, and correction of subtle thyroid abnormalities (78, 96).

Studies of thyroid augmentation in TRD are summarized in Table 4. Early uncontrolled clinical series reported efficacy rates approaching 80% (5, 27, 71, 111). The value of T3 (25 to 50 mg/day) subsequently was documented by Goodwin et al. (39) in a placebo-controlled, within-subject study of 12 severely depressed inpatients, and by Joffe and Singer (46) in a study comparing T3 and T4 augmentation in outpatient TCA nonresponders. In the latter study, response to T3 (53%; 9/17) was significantly greater than T4 (19%; 4/21).

Interpretation of the studies of Goodwin et al. (39) and Joffe and Singer (46) is limited by the lack of a randomized, parallel control group receiving continued antidepressant treatment but not thyroid supplementation. Several contemporaneous studies were not generally supportive of the utility of the thyroid augmentation strategy. Targum et al. (101), for example, found that the effectiveness of thyroid augmentation (either T3 or T4) was limited to a small subgroup of patients with exaggerated thyroid-stimulating-hormone responses to thyrotropin-releasing hormone, an indicator of early thyroid dysfunction. Two other groups found T3 augmentation to be no more effective than continued TCA treatment (37, 105). Although none of these negative reports can be considered definitive (e.g., two were open-label and the third had relatively small cell sizes), they do raise questions about the generalizability of the effectiveness of thyroid augmentation across clinical settings and/or patient populations. Most recently, Joffe et al. (47) published a randomized, placebo-controlled, parallel groups study contrasting T3 and lithium (Li) augmentation. Thyroid augmentation was significantly more effective than placebo and was comparable to lithium augmentation in this study.

Despite some variability (and, possibly, regionality) in outcomes, the studies summarized in Table 4 indicate that augmentation of TCA antidepressants with T3 in doses of 25 to 50 mg/day appears to be effective in 25% to 60% of TCA-treated TRD cases, with very little risk of associated toxicity. The effectiveness of T3 augmentation in SRI, venlafaxine, and bupropion nonresponders remains to be established. Of note, Stern et al. (63) recently found T3 treatment (50 mg per day) to have a significant accelerative effect on speed of ECT response in a small, but well-controlled trial. Although it is not clear what proportion of these cases suffered from TRD, this promising finding warrants further attention.

The use of lithium salts in dosages of 600 to 1200 mg/day (or blood levels of 0.4 to 0.8 mmol) to augment antidepressant response has also been extensively reviewed (52, 79). As summarized in Table 5, lithium augmentation is the best-studied outpatient treatment strategy for TRD.

The utility of the lithium augmentation was heralded by a number of reports in the 1970s, suggesting that the concomitant use of lithium and antidepressants or antipsychotic agents resulted in responses in patients who had otherwise been resistant to treatment (109). In 1981, de Montigny and associates (21) reported a rapid and dramatic response to lithium augmentation in eight TRD patients. These patients had not benefitted from at least 3 weeks of treatment with a variety of antidepressants, yet all eight responded within 48 hr of beginning lithium treatment. The effectiveness of lithium augmentation was subsequently supported by a large number of enthusiastic case reports and small series (see ref. 109). The efficacy of lithium augmentation has been further underscored by the results of six placebo-controlled trials of TRD and four open-label series (see Table 5). Several small series and case reports have suggested that the utility of lithium augmentation extends to combined TCA–neuroleptic treatment of psychotic forms of TRD (see ref. 109). Of note, both Garbutt et al. (30) and Thase et al. (106) have described small groups of tricyclic TRD patients who responded to lithium augmentation after an initial unsuccessful trial of T3 augmentation.

Most investigators have not observed a large number of the dramatic 48-hr responses initially reported by de Montigny and colleagues (21). Thase et al. (106) reported a bimodal distribution of lithium augmentation responses in their 6-week study. One subgroup of patients responded within the first 2 weeks of therapy, whereas a second subgroup required 4 to 6 weeks of treatment. This observation could suggest two modes of action: an acute synergistic effect (such as that described by DeMontigny and associates) and a more slowly emerging primary antidepressant effect.

Variable results have been reported when lithium is used in combination with the more serotonergic TCA clomipramine (109). In one of the more positive reports, the combination of clomipramine, lithium carbonate, and L-tryptophan (i.e., the "New Castle Cocktail") was particularly effective in seven markedly resistant cases (42). At present, the evidence pertaining to the effectiveness of lithium augmentation of other second generation HCAs and SRIs is largely limited to a number of case reports (109). Delgado et al. (20) found a 50% response to lithium augmentation in an open-label study of 18 fluvoxamine nonresponders. In a more recent study, Fontaine et al. (32) found that lithium augmentation of fluoxetine or desipramine was equally effective. However, lithium augmentation of fluoxetine was associated with both significantly more side effects and a greater number of transient responses or relapses than lithium augmentation of desipramine. Controlled trials evaluating the efficacy and side-effects of lithium augmentation of the SRIs should be considered a priority.

Only two randomized prospective trials have compared lithium augmentation against an alternate active treatment. Dinan and Barry (25) reported that a group of 15 TRD patients treated with lithium augmentation responded significantly more rapidly than 15 patients who were randomly assigned to treatment with ECT. By the end of the 3-week trial, however, the lithium augmentation and ECT groups had comparable levels of improvement. As noted previously, Joffe et al. (47) found lithium and T3 augmentation strategies to be equally effective in their placebo-controlled study.

In summary, the studies included on Table 5 indicate that lithium augmentation of an ineffective TCA or HCA yields variable responses, ranging from as low as 20% up to 100% in reported series. A response rate of 50% to 65% is suggested by published data drawn from studies permitting dosage adjustments and at least 4 weeks of augmentation treatment.

Reviews of older data suggest that antipsychotic medications have a modest efficacy (i.e., 20% to 30%) in TRD (61, 87). It is likely that neuroleptics help in the treatment of severe nonpsychotic depression with respect to anxiolytic and sedative effects, as well as more specific control of psychomotor agitation. Some phenothiazine neuroleptics, such as perphenazine, also may potentiate TCA response via competitive inhibition of the antidepressant's metabolism. Although many clinicians continue to use neuroleptic augmentation in severe, nonpsychotic TRD syndromes, empirical data to specifically support this practice are sparse. Furthermore, both significant short-term extrapyramidal side effects and the long-term risk of tardive dyskinesia associated with these agents certainly dampens enthusiasm for their systematic application. By contrast, the concomitant use of neuroleptics with TCAs is clearly indicated in delusional (psychotic) forms of TRD (61), yielding response rates that approach those for ECT.

Beyond augmentation of a failed SRI trial with a TCA, lithium, or T3, some anecdotal experience using postsynaptic serotonergic antagonists, such a buspirone (45), or mixed agents like trazodone (65), is accumulating. These strategies have not yet been subjected to randomized, placebo controlled investigation in TRD.

The use of stimulant agents, such as dextroamphetamine and methylphenidate, in TRD has a long history but little empirical support (4). Some patients refractory to MAOI and tricyclic combinations may be treated successfully with the addition of methylphenidate or Damphetamine (30, 32). Safety concerns often dictate that patients treated with this rather heroic combination be started as inpatients. Unfortunately, no controlled studies of stimulant augmentation therapy have been conducted. Moreover, the abuse potential of these substances certainly places their role in TRD at a lower priority level.

The use of reserpine in TRD has been reviewed by Zohar et al. (119). This strategy has been hypothesized to induce changes in postsynaptic receptor sensitivity in response to a reserpine-induced depletion of brain monoamines. It should also be recalled that reserpine has modest antipsychotic potency, providing a second possible mechanism of action. Several poorly controlled early studies reported outcomes on the order of 70% to 90% in patients treated with repeated intramuscular injections of 5 to 10 mg of reserpine (see ref. 119). Given such dramatic findings, it is curious that reserpine strategies have received so little subsequent attention. Perhaps this is because two recent controlled clinical trials have failed to support the value of this approach for TRD patients (3, 83).

Oppenheim et al. (72) have reviewed the use of estrogen supplementation strategies in women suffering from TRD. However, the value of estrogen supplementation in TRD has not been confirmed in a single double-blind, placebo-controlled trial.

The rationale for use of b-adrenergic agonists in TRD has been reviewed by Zohar et al. (118) and Charney et al. (13) have discussed the rationale for use of the a2 antagonist yohimbine as an augmentation treatment. Both strategies are intended to speed down-regulation of cortical b-receptors, although no convincing evidence exists to support their effectiveness in TRD.

The use of various of monoamine precursors has received considerably more study in TRD (see ref. 109). The literature is rather mixed with respect to whether two common serotonin precursors, L-tryptophan (5-HT) and 5-hydroxytryptamine (5-HTP), are useful as either primary antidepressants or as augmentors (18, 48). Walinder et al. (113) found no evidence of 5-HT enhancement of the efficacy of the relatively specific SRI zimelidine. Furthermore, Steiner and Fontaine (98) reported severe idiosyncratic reactions in a series of five patients receiving fluoxetine combined with 5-HT.

The potential use of alterations in the sleep–wake cycle and sleep deprivation in TRD has been reviewed by several groups (e.g., refs. 51 and 114). Case vignettes and results in small series of patients suggest the utility of these methods in TRD, though randomized controlled trials are typically not available.

The MAOIs have been used as a second- or third-line strategy for TRD for over 30 years. A number of reviews of the use of MAOIs in TRD have been published (e.g., ref. 24).

A number of open-label reports of MAOI treatment of TRD have been published (see Table 6). In a series of comparative studies, Nolen et al. (67, 69, 70) found tranylcypromine (TCP) to be significantly more effective than sleep deprivation, 5-HPT, and nomifensine; 50% of their highly refractory patients responded to TCP. In an open-label, crossover study of 42 recurrent unipolar depressives who had been vigorously treated with both IMI (mean dose 257 mg/day) and interpersonal psychotherapy, Thase et al. (107) found MAOI response to be both statistically and clinically significant (57% response rate). Moreover, MAOI treatment outcome was comparable to that observed in an earlier cohort treated with lithium augmentation and superior to thyroid augmentation. Most recently, Nolen et al. (70) compared responses to tranylcypromine (n = 17) with the novel MAO-type A selective agent brofaromine (n = 22) in hospitalized TRD patients. Response rates to the two MAOIs were identical (59%), although the novel MAOI was significantly better tolerated.

Several groups have conducted comparative studies of MAOI efficacy in TRD using double-blind, crossover methodology (see Table 6). As noted above, Nolen et al. (67, 69, 70) found both tranylcypromine (TCP) and brofaromine to be effective in patients resistant to either nortriptyline or maprotiline. Thase et al. (107) found TCP to be an effective treatment of anergic bipolar depressions resistant to IMI. Similarly, McGrath et al. (58) found phenelzine to be significantly effective in IMI-resistant nonbipolar patients with atypical depression (as defined by Columbia criteria).

As summarized in Table 6, the spectrum of MAOI efficacy in TRD appears to extend from adolescent to elderly patients. Unconventionally high doses of MAOIs also may be effective in carefully selected, side-effect-free patients who fail to respond to conventional doses of these agents (2, 40).

It is unclear if the MAOIs are particularly useful in TRD because the structural dissimilarity between MAOIs and TCAs reflects distinctly different mechanisms of action or, more simply, because the two families of antidepressants have substantially different side-effect profiles. This interesting line of research requires verification. As reviewed earlier, subgroups of TRD patients more responsive to MAOIs may include atypical, anergic bipolar, and comorbid anxious/phobic subgroups (17, 104). For example, Thase et al. (102) found a MAOI response in TRD of only 33% in patients with typical major depressions, whereas nearly 80% of patients with reversed vegetative features responded to MAOI therapy.

Given such apparent selectivity for synergic or atypical syndromal presentations, it remains to be seen if other nonsedating compounds such as the SRIs, venlafaxine, or bupropion can be used in a similar fashion (38, 86). The finding of Nolen et al. (69) that nomifensine apparently lacks efficacy would suggest that the relative efficacy of MAOIs in TRD goes beyond a nonsedating clinical pharmacology. Conversely, the effectiveness of the MAOIs, which have potent effects on serotonergic neurotransmission, has not yet been confirmed in patients who have failed therapy with a SRI. Nevertheless, the MAOIs are probably the treatments of choice for Stage 3 TRD (i.e., patients resistant to both SRIs and TCAs, as well as their augmentation).

A number of open-label case reports in small groups have described positive responses to lithium or T3 augmentation when patients had not benefitted from treatment with an MAOI alone (see ref. 109). These strategies have not, however, been tested prospectively. Among other possible augmentation strategies for use with MAOIs, an older literature indicates that response to MAOIs may be enhanced by the addition of L-tryptophan (5-HT) (109). However, some individuals have developed serotonin syndrome when these agents are used in tandem (75). Similarly, SRIs are not to be used in combination with MAOIs because of several deaths reported with this combination suggestive of malignant hyperthermia (5).

The combined use of TCAs and MAOIs violates an explicit prohibition against their concurrent use in the Physician's Desk Reference. Nevertheless, a number of case reports and open-label studies dating to the 1960s, suggest that these agents may be synergistically effective in TRD (109). The general safety of combined TCA and MAOI treatment has been demonstrated in several controlled trials of nonresistant depressed patients. Combined therapy appears well-tolerated when moderate-to-low doses of both agents are used and the MAOI is added to an established dose of a TCA. However, these studies did not find evidence that the MAOI–TCA combination was more effective than either agent used alone in adequate dosages. Only one clinical trial has specifically dealt with the efficacy of combined MAOI and TCA treatment of TRD (16). In this trial, the combination of amitriptyline and phenelzine was clearly less effective than ECT. The combination of a TCA and MAOI thus is probably best reserved for after failure to respond to monotherapies, augmentation trials, and ECT.

The literature on the use of anticonvulsants in TRD has been reviewed by Kahn (50) and Post and Uhde (76). The value of anticonvulsants, such as carbamazepine and valproate as treatments of bipolar affective disorder has been recently established. Several series and case studies suggest that a subset of unipolar TRD patients, ranging from 20% to 40%, may respond to acute antidepressant treatment with carbamazepine (15, 50, 76). Kramlinger and Post (54) found additional improvement in 8 of 15 patients when lithium augmentation was added to an ineffective course of carbamazepine. Pending further study, the anticonvulsants seem particularly indicated for use in bipolar TRD syndromes or in clinically labile variants of unipolar disorder.

The oldest available treatment of TRD is still the most consistently effective (89, 109), and response rates of 50% to 70% are consistently observed (see Table 7). Although the ECT response rate in TRD is significantly lower than typically seen in nonresistant cases (84), at the worst the response rate reported for ECT is equivalent to response rates to MAOI treatment or lithium augmentation, and superior to all other therapeutic options for TRD (109). Electroconvulsive therapy response also may be enhanced by switching from unilateral to bilateral treatment modes and monitoring duration of seizure time to prevent missed treatments (89). Thus, ECT remains the treatment of choice for the most severe, incapacitating forms of TRD.

Although ECT is clearly an effective treatment of TRD, it has also recently been shown that relapse rates are significantly higher in TRD patients after a successful course of therapy (90). Research is necessary to establish the efficacy of alternate methods to prevent relapse following successful ECT, including maintenance ECT and combination pharmacotherapy strategies.

Patients who fail to respond to ECT (proposed Stage 5 TRD), represent some of the most challenging cases of TRD. Assuming that the course of ECT has been optimized (i.e., at least 12 total treatments, including at least 6 bilateral treatments with confirmation of the adequacy of seizure duration), it would seem prudent to allow a brief medication-free washout and proceed with one of the major strategies that have not yet been tried. Anecdotal clinical experience suggests that some patients are more responsive to antidepressant agents after an unsuccessful course of ECT, perhaps because of treatment-induced changes in postsynaptic receptor sensitivity (95). Thus, failure on ECT does not forbode nonresponsiveness to all somatic therapies.

The value of various forms of psychotherapy in the management of patients with TRD has been reviewed elsewhere (see ref. 103). Traditionally, psychotherapy has been considered useful in the management of TRD primarily as an adjunct to help patients maintain morale and optimism. However, data to support these suggested indications are sparse, and those studies that are available are limited to the newer forms of psychotherapy, such as cognitive behavior therapy (CBT). Among several small groups of TRD patients, CBT response rates of 25% to 40% have been reported (see ref. 103). Cognitive behavior therapy thus may have about the same probability of response in TRD outpatients as retreatment with a similar antidepressant agent. The magnitude of this estimated effect is likely to be smaller than those observed for lithium augmentation, MAOIs, and ECT. Some evidence suggests that TRD patients benefit from more intensive models of therapy, for example, those offering frequent sessions and an admixture of individual, group, and milieu modalities. Such regimens are most practicable in partial-hospital or inpatient settings. Inpatient treatment with intensive CBT (in lieu of alternate somatic modalities) would appear to be best suited for patients who are poor candidates for further medication trials or for those who refuse ECT (103). A preliminary study by Miller et al. (60) suggests that CBT also might may be usefully combined with pharmacotherapy in TRD. In this regard, provision of psychotherapy may reduce attrition rates from pharmacological treatment and strengthen therapeutic alliances during the long-term management of chronic mood disorders (103).

The use of psychosurgery in TRD has been reviewed by Bridges (8). Although a number of early reports described dramatic success following stereotactic leukotomy (i.e., sustained improvement rates of 50% to 80%), more recent studies suggest that sustained remission rates of only about 25% to 50% can be expected (55, 77). Complications include epilepsy and irreversible personality changes (8). Psychosurgery, therefore, remains the last line of somatic treatment for patients with TRD.

Whereas mood disorders have generally been viewed as episodic and of good prognosis, a large subset of this population (45% to 50%) can be expected to either be intolerant to or fail to respond to an initial medication trial (27, 104). Evidence to date indicates that a second monotherapy will effectively treat about 40% to 50% of those who have failed with the initial treatment, especially if the second drug has a pharmacological profile distinct from the initial medication. The remaining 25% of mood-disordered patients are candidates for one or more augmentation strategies, followed by treatment with a MAOI. Electroconvulsive therapy has a place in the treatment plan for patients failing multiple monotherapies or augmented treatment packages. The need for randomized controlled studies utilizing innovative designs to identify the preferred therapies for those patients with varying degrees of treatment resistance is clear. Even now, however, present evidence argues for carefully conducted sequenced medication trials in patients who do not respond satisfactorily to the initial treatment.

The review was supported in part by NIMH grants MH-41115 (to A. J. Rush), MH-30915 (University of Pittsburgh Mental Health Clinical Research Center), and MH-41884 (to M. E. Thase). The authors appreciate the secretarial support of Fast Word, Inc. of Dallas, Lisa Stupar, and David Savage, and the administrative support of Kenneth Z. Altshuler, Stanton Sharp Professor and Chairman of the Department of Psychiatry of the University of Texas Southwestern Medical Center, and David J. Kupfer, Chairman of the Department of Psychiatry of the University of Pittsburgh Medical Center.

published 2000