The efficacy of antidepressants in the treatment of depression and other related affective disorders has been demonstrated in randomised controlled trials (RCTs) (Reference Song, Freemantle and SheldonSong et al, 1993; Reference Anderson and TomensonAnderson & Tomenson, 1994; Reference Hotopf, Lewis and NormandHotopf et al, 1996; Reference Geddes, Freemantle and MasonGeddes et al, 1999; Reference AndersonAnderson, 2000). Taking an antidepressant drug at an effective dose for an adequate period reduces the number and severity of acute symptoms (Reference Anderson and TomensonAnderson & Tomenson, 1994; Reference Montejo, Gilaberte and FombellidaMontejo et al, 1998), improves the patient's ability to function (Reference Mintz, Mintz and ArrudaMintz et al, 1992; Reference Sturm and WellsSturm & Wells, 1995) and minimises the risk of relapse or recurrence (Reference Prien and KupferPrien & Kupfer, 1986; Reference Maj, Veltro and PirozziMaj et al, 1992; British Association for Psychopharmacology, 1993). In major depression, pooled data from meta-analyses of clinical trials have failed to show consistently that any one antidepressant or group of antidepressants has superior efficacy over any other (Anon., 1993; Reference Song, Freemantle and SheldonSong et al, 1993; Reference Anderson and TomensonAnderson & Tomenson, 1994; Reference Hotopf, Lewis and NormandHotopf et al, 1996; Reference Geddes, Freemantle and MasonGeddes et al, 1999; Reference AndersonAnderson, 2000). Although one meta-analysis found tricyclic antidepressants (TCAs) to be more efficacious than selective serotonin reuptake inhibitors (SSRIs) in hospital in-patients (Reference AndersonAnderson, 1998), another meta-analysis using different methodology failed to find any difference (Reference Geddes, Freemantle and MasonGeddes et al, 1999).

However, the conditions in which clinical trials are conducted may be very different from the conditions of clinical practice. Clinical trials typically take place in highly selected populations treated under controlled and optimal conditions, with treatment determined by a clear study protocol. This type of study design is quite deliberate — to reduce the potential for bias and to ensure that any differences found can be attributed to differences in the properties of the compounds being investigated.

In contrast, treatment conditions in clinical practice are far removed from those of clinical trials, with no patient selection, no randomisation, and with variations in the behaviour of prescribers and patients and in the systems of health care in which treatment is delivered (Reference Simon, Wagner and Von KorffSimon et al, 1995). Data suggest that patient compliance and adherence to medications in clinical practice may also differ from that in RCTs (Reference DemyttenaereDemyttenaere, 1997).

While the efficacy findings of RCTs are certainly important factors in antidepressant selection, the efficacy of a compound is no more than a measure of its therapeutic potential under standardised and optimal conditions. A more useful concept is that of real-world efficacy or ‘effectiveness’: the use of the compound in ways that will deliver the desired outcomes of treatment. Given the complexities of clinical practice, an antidepressant's efficacy as demonstrated by RCTs is likely to overestimate its effectiveness in clinical practice settings. Differences in effectiveness between antidepressant classes and individual compounds may help to guide antidepressant selection when efficacy is similar.

TREATMENT GUIDELINES

The findings of efficacy from clinical trials, and the conditions that need to be met to ensure effectiveness in clinical practice, have influenced the content of both national and international guidelines for the management of depression (Reference Donoghue and TaylorDonoghue & Taylor, 2000). Generally, guidelines recommend that to be effective antidepressants need to be used at a known effective dose, and to reduce the likelihood of relapse, treatment should be maintained at the same dose for a minimum period, usually 4-6 months following resolution of acute symptoms (Reference Donoghue and TaylorDonoghue & Taylor, 2000). In clinical practice there are many factors that may influence whether these conditions are met, including the pharmacological properties of the drug (half-life, dosing regimen); tolerability; patients' beliefs and behaviour (compliance); providers' beliefs and behaviour (experience, preferences); health care system characteristics (financing of health care benefits and ease of access to treatment); and societal factors (stigma of treatment, discrimination).

ADEQUACY OF DOSE

ADHERENCE TO DOSING REGIMEN

Naturalistic data suggest that between 30% and 60% of patients do not take their medication as prescribed (Reference CramerCramer, 1995; Reference DemyttenaereDemyttenaere, 1997). A 9-week RCT found that only 42.5% of patients took their antidepressant medication correctly (right dose, right time of day) 80% of the time (Reference Demyttenaere, Van Ganse and GregoireDemyttenaere et al, 1998); in a non-controlled clinical setting, adherence might be expected to be even worse. Another study found that between 15% and 25% of patients had a gap of 2 weeks or more between their antidepressant prescriptions, which raises obvious questions about adherence to their medication (Reference Hylan, Dunn and TepnerHylan et al, 1998b ). Many factors contribute to non-compliance with antidepressants, including the health care system, intrinsic pharmacological characteristics of the antidepressant (including adverse effects), and patient and prescriber behaviour (Reference DemyttenaereDemyttenaere, 1998).

A general perception exists that patients are more compliant with SSRIs than with TCAs. However, definitive evidence supporting this perception is lacking, although this may be attributable to the difficulty in studying issues of compliance (Reference DemyttenaereDemyttenaere, 1997). Simple methods of assessing compliance, such as pill counts and interrogation, are often inaccurate and generally overestimate adherence. More accurate methods of compliance assessment, such as electronic monitoring of medication bottle openings, are not simple to employ (Reference DemyttenaereDemyttenaere, 1997).

In a 9-week evaluation of compliance, assessed by use of a microelectromechanical system (MEMS) to monitor patients taking amitriptyline or fluoxetine (Reference Demyttenaere, Van Ganse and GregoireDemyttenaere et al, 1998), both patient groups were found to have poor compliance in general, with more than half the patients having treatment adherence of less than 50%. No significant differences in treatment adherence were seen between groups. Interestingly, therapeutic effect was not correlated with compliance.

Another study has evaluated compliance in out-patients prescribed dothiepin or fluoxetine (Reference Thompson, Peveler and StephensonThompson et al, 2000) using these methods: pill count, patient report and MEMS monitoring. Similar compliance rates were found using pill count (76% fluoxetine; 64% dothiepin) and patient report (79% fluoxetine; 80% dothiepin) with both methods, suggesting that patients were generally compliant with their dosing regimen. However, when assessed using the MEMS, compliance was found to be less than that assessed by either pill count or patient report. The authors suggested that long windows (7 days or more) of total non-compliance over the course of the treatment period would be more damaging to treatment response than brief episodes spread out over the course of treatment, and defined a compliance index as a measure of this. Using the MEMS data to calculate the compliance index, adjusted mean compliance rates were significantly greater for fluoxetine-treated patients compared with dothiepin-treated patients (70.0% fluoxetine; 55.8% dothiepin;P=0.01). Additionally, treatment response for both medications was significantly correlated with the compliance index (P<0.001), but not with the 80% MEMS compliance measure. The discrepancy between the crude compliance measures (pill count and patient report) and the more rigorous method (MEMS) highlights some of the difficulties in studying this complex phenomenon.

One potential consequence of non-adherence to the prescribed regimen is the possibility of discontinuation symptoms, which have been described for both TCAs and SSRIs, with symptoms sometimes reported to occur after as little as one missed dose (Reference Charney, Heninger and SternbergCharney et al, 1982; Reference Dilsaver and GredenDilsaver & Greden, 1984; Reference Dilsaver, Greden and SniderDilsaver et al, 1987; Reference Schatzberg, Haddad and KaplanSchatzberg et al, 1997; Reference Parker and BlennerhassettParker & Blennerhassett, 1998). Comparative studies of individual SSRIs show that they have different propensities for discontinuation symptoms (Reference Coupland, Bell and PotokarCoupland et al, 1996; Reference Lejoyeux, Ades and MouradLejoyeux et al, 1996; Reference Price, Waller and WoodPrice et al, 1996), with patients taking sertraline or paroxetine more likely to experience adverse events and a re-emergence of depressive symptoms on discontinuation than patients taking fluoxetine (Reference Rosenbaum, Fava and HoogRosenbaum et al, 1998).

DURATION OF THERAPY

In clinical practice, only a small proportion of patients with depression experience adequate duration of therapy (Reference Katon, von Korff and LinKaton et al, 1992; Reference MacDonald, McMahon and ReidMacDonald et al, 1996; Reference Lepine, Gastpar and MendlewiczLepine et al, 1997; Reference DonoghueDonoghue, 1998b ; Reference Dunn, Donoghue and OzminkowskiDunn et al, 1999), which is defined as 4-6 months beyond acute symptom resolution (World Health Organization Mental Health Collaborating Centers, 1989; Reference Paykel and PriestPaykel & Priest, 1992; Agency for Health Care Policy Research, 1993; British Association for Psychopharmacology, 1993; Reference Reimherr, Amsterdam and QuitkinReimherr et al, 1998). For example, in a population-based study from Scotland of over 400 000 patients, only 32% of patients prescribed an antidepressant received more than 90 days of treatment (Reference MacDonald, McMahon and ReidMacDonald et al, 1996). This duration of therapy is often required for acute symptom resolution, and does not meet the criteria for appropriate continuation therapy.

In primary care there is some limited evidence that TCA-treated patients discontinue treatment more rapidly than SSRI-treated patients, with one study reporting that during a year's treatment, patients who began therapy on the TCA dothiepin received fewer prescriptions for the initially prescribed antidepressant than patients who began therapy on the SSRI fluoxetine (Reference Treglia, Neslusan and DunnTreglia et al, 1999). In a retrospective Swedish study, patients who began treatment on an SSRI were more likely to have a longer initial therapy length (and possibly a reduced risk of recurrence) than patients who began therapy on a TCA. Subsequently, TCA-treated patients were more than twice as likely to have a recurrent episode of antidepressant treatment (at least 9 months between the end of the initial episode and restarting an antidepressant) than patients who began therapy on an SSRI (Reference Isacsson, Boethius and HenrikssonIsacsson et al, 1999).

Naturalistic studies of SSRIs have found that patients started on fluoxetine are more likely to complete a minimum course of therapy than patients started on other antidepressants (Reference Katon, von Korff and LinKaton et al, 1992; Sclar et al, Reference Sclar, Robison and Skaer1994, Reference Sclar, Robison and Skaer1995; Reference Simon, Von Korff and HeiligensteinSimon et al, 1996; Reference Croghan, Lair and EngelhartCroghan et al, 1997; Reference Montejo, Gilaberte and FombellidaMontejo et al, 1998; Reference Dunn, Donoghue and OzminkowskiDunn et al, 1999; Reference Hylan, Crown and MeneadesHylan et al, 1999a ; Reference McCombs, Nichol and StimmelMcCombs et al, 1999). In a large naturalistic study, patients initially prescribed sertraline were found to drop out of treatment at a significantly faster rate than patients prescribed fluoxetine (sertraline v. fluoxetine P<0.001) or paroxetine (sertraline v. paroxetine P<0.001) (Reference DonoghueDonoghue, 1998b ).

LONGITUDINAL STUDIES OF TREATMENT PATTERN

DISCUSSION: WHY DO THESE OBSERVED DIFFERENCES OCCUR?

Longitudinal studies of antidepressant use patterns confirm the findings of previous cross-sectional studies. Differences in use patterns are clearly demarcated by the class of antidepressant prescribed: patients who begin therapy on a TCA generally receive sub-therapeutic doses and discontinue the medication rapidly. In contrast, patients who begin therapy on an SSRI are more likely to have an adequate dose, although the duration of treatment is still shorter than that recommended in treatment guidelines. Within the SSRIs, differences also exist, although to a lesser degree. These differences are likely to affect the effectiveness of antidepressant treatment in the clinical setting.

It is possible that a single influence, a range of influences, or an interaction between them could cause these differences to occur. They could be due to pharmacological differences between the compounds. This seems plausible, as TCAs as a group have many similarities, as do the SSRIs. Controlled trials have shown that both groups of compounds have similar efficacy, so pharmacological differences are likely to be manifested as differences in tolerability, which could then be reflected in the prescribing patterns. Additionally, prescribing patterns are likely to be influenced by both the patient's behaviour and expectations of treatment, and the physician's behaviour and experience with treatments. Co-prescription of other medicines, concomitant physical or psychiatric illnesses, use of other health services and disease severity are also likely to influence prescribing patterns. Indeed, in the large longitudinal database studies described earlier (Dunn et al, Reference Dunn, Donoghue and Ozminkowski1998, Reference Dunn, Donoghue and Ozminkowski1999), the diagnostic criteria were not defined and disease severity was not assessed, both factors that might have influenced the outcomes observed. It has been suggested that many patients in primary care treated with TCAs at low doses actually suffer from mild, mixed anxiety-depression disorders and can be treated effectively with lower antidepressant doses than patients with major depression, although evidence to support this is lacking (Reference Thompson and ThompsonThompson & Thompson, 1989a ; Reference AliAli, 1998; Reference DonoghueDonoghue, 1998a ). Similarly, a dose of dothiepin of 75 mg per day was found to be effective in preventing recurrence of depression in elderly patients (Old Age Depression Interest Group, 1993), although doses prescribed for the elderly in clinical practice in primary care were found to be much lower than this (Reference Donoghue, Katona and TyleeDonoghue et al, 1998). None the less, differences in use patterns, especially between TCAs and SSRIs, appear to be consistent across national boundaries, among patient populations who are likely to have very different health beliefs and expectations, in doctors who experience widely different training, and in systems of health care that deliver services in very different ways. The only constant influence appears to be biological: the differences in prescribing patterns occur as a result of pharmacological differences between the compounds.

The tolerability profile of an antidepressant can be expected to influence both adherence to dosing and overall duration of treatment (Reference Kuzel, De Wester and RichardsonKuzel et al, 1996; Reference Demyttenaere, Van Ganse and GregoireDemyttenaere et al, 1998). The SSRIs are generally considered to have tolerability advantages over TCAs, as they have a more benign adverse effect profile (Reference Benfield and WardBenfield & Ward, 1986; Reference Benfield, Heel and LewisBenfield et al, 1986; Reference Dechant and ClissoldDechant & Clissold, 1991; Reference Feighner, Gardner and JohnstonFeighner et al, 1991; Reference EdwardsEdwards, 1992; Reference Grimsley and JannGrimsley & Jann, 1992; Reference Murdoch and McTavishMurdoch & McTavish, 1992; Reference Edwards, Inman and WiltonEdwards et al, 1994; Reference Montgomery and KasperMontgomery & Kasper, 1998). In addition, the adverse effects associated with TCAs typically do not decrease in severity with continued therapy (Reference Reimherr, Byerley and WardReimherr et al, 1988), whereas the side-effects of SSRIs generally tend to be transient (Reference Reimherr, Byerley and WardReimherr et al, 1988; Reference Zajecka, Amsterdam and QuitkinZajecka et al, 1999) and more easily managed, for example by altering the time of administration (morning or evening) or taking the drug with food (Reference Rickels and SchweizerRickels & Schweizer, 1990; Reference NemeroffNemeroff, 1994; Reference Kuzel, De Wester and RichardsonKuzel et al, 1996). Of course, some patients on SSRIs do not experience symptom relief with continued therapy, and some side-effects associated with SSRIs (such as sexual dysfunction) are not expected to be short-lived.

Meta-analyses of clinical trials have confirmed that more patients discontinue treatment because of adverse events on TCAs than on SSRIs (Reference Montgomery, Henry and McDonaldMontgomery et al, 1994; Reference Anderson and TomensonAnderson & Tomenson, 1995; Reference Montgomery and KasperMontgomery & Kasper, 1995; Reference Geddes, Freemantle and MasonGeddes et al, 1999), although the differences in drop-out rates are modest. Other analyses have reported a higher incidence of adverse events with TCAs relative to fluoxetine, with significantly higher discontinuation rates for TCAs (Reference Pande and SaylerPande & Sayler, 1993), and have found more adverse events reported with TCAs than with the SSRI citalopram (Reference Bech and CialdellaBech & Cialdella, 1992). Further evidence can be found in a study that accounted for both frequency and severity of side-effects, in which the side-effect burden of amitriptyline was found to be substantially greater than that of sertraline (Reference Reimherr, Chouinard and CohnReimherr et al, 1990), and in a controlled study where the rate of discontinuation due to adverse events was three times higher for the TCAs desipramine and imipramine than for fluoxetine (Reference Simon, Von Korff and HeiligensteinSimon et al, 1996).

The possibility that differences in prescribing patterns occur as a result of pharmacological differences is given additional weight by the logistic regression analyses conducted by Dunn et al (Reference Dunn, Donoghue and Ozminkowski1998, Reference Dunn, Donoghue and Ozminkowski1999). This technique allows the potentially confounding influences of patient characteristics, region of the country, diagnosis of depression, concomitant physical and psychiatric illness, co-prescription of other medicines and the use of other health resources to be controlled for, with the results as reported above. It may be that there are still more influences at work of which we are as yet unaware, but on the evidence currently available, the most significant predictor of the subsequent pattern of treatment (which will, in its turn, determine the outcome in terms of response, recovery or likelihood of relapse) is the antidepressant with which treatment is initiated. The evidence clearly shows the advantage of initiating treatment with an SSRI.

Differences between the SSRIs are more difficult to explain, as pharmacological differences between them are small and none has been shown consistently to be better tolerated (Reference LeonardLeonard, 1993; Reference FinleyFinley, 1994). Clinical trials and pooled data from meta-analyses suggest that the SSRIs have more similarities than differences: similar efficacy, similar side-effects, similar tolerability and similar toxicity in overdose (Reference LeonardLeonard, 1993; Reference Song, Freemantle and SheldonSong et al, 1993; Reference Anderson and TomensonAnderson & Tomenson, 1994; Reference FinleyFinley, 1994; Reference Henry, Alexander and SenerHenry et al, 1995). However, despite these similarities, differences have been reported in their adverse effect profiles. Paroxetine has been associated with the highest incidence of antimuscarinic side-effects because of its affinity for cholinergic receptors (Reference GagianoGagiano, 1993; Reference DeVaneDeVane, 1995; Reference PreskornPreskorn, 1995; Reference NelsonNelson, 1997), and citalopram has been associated with a higher degree of sedation because of its degree of binding to H₁ receptors (Reference FinleyFinley, 1994). Fluoxetine has been reported to produce an increase in the frequency of anxiety or anxiety-related symptoms including nervousness and agitation (Reference DeVaneDeVane, 1995; Reference Fisher, Kent and BryantFisher et al, 1995; Reference PreskornPreskorn, 1995; Reference Price, Waller and WoodPrice et al, 1996; Reference NelsonNelson, 1997; Reference MacKay, Dunn and MartinMacKay et al, 1999), although other studies found fluoxetine to be similar to other SSRIs (Reference TignolTignol, 1993; Reference Patris, Bouchard and BougerolPatris et al, 1996; Reference Fava, Rosenbaum and HoogFava et al, 1998). Weight loss and anorexia have been associated with fluoxetine (Reference TignolTignol, 1993; Reference PreskornPreskorn, 1995; Reference NelsonNelson, 1997; Reference Michelson, Amsterdam and QuitkinMichelson et al, 1999), while weight gain has been associated with paroxetine (Reference Fava, Rosenbaum and HoogFava et al, 1998) and citalopram (Reference Bouwer and HarveyBouwer & Harvey, 1996). Headache has been associated with fluoxetine (Reference DeVaneDeVane, 1995; Reference PreskornPreskorn, 1995; Reference NelsonNelson, 1997; Reference Ontiveros and Garcia-BarrigaOntiveros & Garcia-Barriga, 1997). Based on data from clinical practice, the incidence of adverse events has been reported to be highest with fluvoxamine (Reference MacKay, Dunn and WiltonMacKay et al, 1997; Reference Arias, Padin and GilaberteArias et al, 1998). Gastrointestinal complaints, including nausea and vomiting, have been reported with fluvoxamine (Reference Price, Waller and WoodPrice et al, 1996; Reference Arias, Padin and GilaberteArias et al, 1998), paroxetine (Reference MacKay, Dunn and MartinMacKayet al, 1999) and sertraline (Reference DeVaneDeVane, 1995; Reference Fisher, Kent and BryantFisher et al, 1995). Tremor has been reported with paroxetine (Reference Price, Waller and WoodPrice et al, 1996; MacKay et al, Reference MacKay, Dunn and Wilton1997, Reference MacKay, Dunn and Martin1999), fluvoxamine (Reference MacKay, Dunn and WiltonMacKay et al, 1997) and sertraline (Reference Arias, Padin and GilaberteArias et al, 1998). Sexual dysfunction has been observed with all SSRIs, although reported to occur more frequently with paroxetine (Reference Price, Waller and WoodPrice et al, 1996; Reference MacKay, Dunn and WiltonMacKay et al, 1997; Reference NelsonNelson, 1997; Reference Fava, Rosenbaum and HoogFava et al, 1998; Reference Rosen, Lane and MenzaRosen et al, 1999) and sertraline (Reference DeVaneDeVane, 1995; Reference Fisher, Kent and BryantFisher et al, 1995; Reference NelsonNelson, 1997).

There is, however, one property that separates fluoxetine from the other SSRIs. The pharmacokinetics of fluoxetine, with the very long half-life of its active metabolite norfluoxetine, suggest that until steady-state plasma levels are achieved (which may take up to 5 weeks), patients taking 20 mg daily undergo very gradual incremental increase in plasma concentration. The incidence and discomfort of adverse effects is likely to be experienced more acutely in situations where treatment is initiated at a high dose, or where doses are increased rapidly. Thus, the long half-life of fluoxetine/norfluoxetine may confer greater tolerability, at least in the important early stages of treatment where the dropout rate is highest, which in turn results in more patients achieving a minimum period of stable therapy. The corollary to this, however, is that although adverse effects may be experienced less acutely, when they occur they will persist for longer than similar effects experienced with drugs with shorter half-lives. Similarly, problems may occur with drug interactions and on switching from fluoxetine to another antidepressant, when a long wash-out period may be required before another agent can be prescribed.

Because non-compliance and non-adherence to dosage in clinical practice is common, therapeutic coverage that is maintained despite missing doses may be beneficial. This benefit may be further enhanced if the potential for adverse effects when patients miss doses or abruptly discontinue therapy is reduced.

The data presented here show the disparity between the findings of controlled trials and those of observational studies. Controlled trials show no important differences in efficacy, and only modest tolerability advantages for the SSRIs compared with TCAs. Observational studies show that SSRIs appear to have important clinical advantages over TCAs, but the findings may be subject to bias. This creates a dilemma for practitioners over which results should take precedence. Although there seems to be a compelling case to abandon control of bias (controlled studies) in favour of generalisability (observational studies), a careful consideration of all the available evidence suggests that both sources of data are needed to make fully informed clinical decisions. Data from RCTs are essential to ensure that interventions for which efficacy has not been demonstrated are not applied in clinical practice, and observational studies provide information indicating which of a number of interventions of proven efficacy is most likely to deliver the desired outcome in a practice rather than a research setting.

CONCLUSION

Common behaviours in clinical practice include sub-therapeutic dosing, missing doses during therapy and inadequate lengths of therapy. These practices decrease the likelihood of acute symptoms improving and increase the likelihood of relapse or recurrent depressive episodes, demonstrating that there are significant opportunities for improvement in antidepressant prescribing. One possible opportunity is for individual prescribers to consistently select antidepressants that naturally maximise the potential for real-world efficacy.

Despite the proven efficacy of TCAs in controlled trials, patients treated with these antidepressants appear to have little hope of receiving adequate therapy. The findings of small naturalistic studies over many years — that patients prescribed TCAs are treated at doses that are not effective for the treatment of their depression — have been confirmed by large epidemiological studies which have revealed the pervasiveness of this practice, which seems to be independent of nationality or system of health care. More recent longitudinal analyses find that patients beginning treatment with a TCA start and remain on low doses, and still discontinue treatment prematurely.

In contrast, patients beginning treatment with an SSRI are at least seven times more likely to receive an adequate course of antidepressant therapy. Even so, a large proportion of patients drop out of treatment prematurely, leaving considerable scope for improvement.

These findings are restricted to the TCAs and SSRIs most commonly prescribed for the treatment of depression in primary care, and come largely from studies performed in the UK. On the basis of these data, SSRIs must be regarded as the gold standard for the treatment of depression in primary care. As newer antidepressants become used more widely, further studies will be required to evaluate the outcomes obtained from their use, and their potential to improve the management of depression.