Acute phase ADM trials: efficacy and limitations

Response rates (symptom reduction of 50% or more from baseline) in aADM arms exceed those in PLA arms, typically by 10–15%. The largest recent meta-analysis of aAD efficacy by Cipriani et al. included 522 trials, published and unpublished, comprising 1 16 477 participants and found that ADs had better response rates than placebo, with ORs ranging between 1.37 and 2.13 and an averaged SMD = 0.30, similar to NNT = 8 (Meadows et al., Reference Meadows, Prodan, Patten, Shawyer, Francis and Enticott2019), comparable with other recent meta-analyses (Cipriani et al., Reference Cipriani, Furukawa, Salanti, Chaimani, Atkinson, Ogawa and Geddes2018). Although treatment settings were often unclear, most trials did not take place in primary care settings.

Cipriani et al.'s meta-analysis may even underestimate antidepressants efficacy as post-hoc analyses have found consistent superiority and dose-dependency for selective serotonin reuptake inhibitors (SSRIs) relative to PLA when using depressed mood as the major outcome (Hieronymus, Emilsson, Nilsson, & Eriksson, Reference Hieronymus, Emilsson, Nilsson and Eriksson2016a; Hieronymus, Nilsson, & Eriksson, Reference Hieronymus, Nilsson and Eriksson2016b) or the six prototypical depression items from the HAMD [depressed mood, guilt feeling, loss of interests, psychomotor retardation, psychic anxiety, and tiredness (Bech, Reference Bech2010)]. The dose-dependency figures suggest that inclusion of trials using suboptimal doses in Cipriani et al.'s meta-analyses may have led to an underestimation of aADM efficacy. Likewise, the use of more narrow depression symptoms might have yielded higher efficacies, although caution is needed given the post-hoc nature of these findings. Because it is not fully clear whether findings from specialty settings can be generalized to primary care as only a minority of cases are referred to specialty settings, Linde et al. meta-analyzed 66 randomized controlled trials (RCTs) with 15.161 patients treated in primary care (Linde et al., Reference Linde, Kriston and Ruecker2015). ADs were found to be significantly superior to PLA, with estimated ORs between 1.69 and 2.03, rather similar to the Cipriano results. There were no significant differences between drug classes. Compared to general medicine drugs (e.g. ACE inhibitors and statins for prevention of cardiovascular events and stroke) ADM are not in general less efficacious (Leucht et al., Reference Leucht, Hierl, Kissling, Dold and Davis2012), although the clinical significance of a given effect size is context-dependent and may differ between psychiatry and general medicine.

Such data should settle the debate, at least for the acute phase treatment. However, a few issues still produce uncertainties. First, the authors caution that the ‘certainty of evidence was moderate to very low’ with ‘46 (9%) of 522 trials rated as high risk of bias, 380 (73%) as moderate, and 96 (18%) as low’ (p. 1), leaving space for speculation about bias. Second, the analysis did not address full remission efficacy, only response, i.e. 50% or more improvement relative to baseline. That is, patients were better but not necessarily well. Third, the modest SMD, corresponding to a few HAMD points, has been argued to be of limited clinical significance (Jakobsen et al., Reference Jakobsen, Katakam, Schou, Hellmuth, Stallknecht and Leth-Moller2017; Kirsch, Reference Kirsch2014; Moncrieff & Kirsch, Reference Moncrieff and Kirsch2015; Turner et al., Reference Turner, Matthews, Linardatos, Tell and Rosenthal2008). Fourth, it remains unclear how large the risk of breaking the blind is in SSRI trials because they have no ‘active’ placebo. Side effects of TCAs can be mimicked and a meta-analysis comparing TCA and active placebos, though limited by methodological problems found no significant difference (Moncrieff, Reference Moncrieff2003). These limitations keep providing fuel for critics.

Prolonged acute phase trials: 6–8 months outcomes

Some acute phase placebo-controlled trials have prolonged follow-up assessments of response and remission up to 8 months post-baseline randomization. Deshauer et al.'s meta-analysis included six studies that randomized a total of 1299 patients to either ADM or PLA and followed both groups for 6–8 months (Deshauer et al., Reference Deshauer, Moher, Fergusson, Moher, Sampson and Grimshaw2008). Response rates averaged 61% in the ADM and 48% in the PLA arm, a difference of 13%; rates for remission, assessed in only four trials, were 45% and 38%, a difference of 7%, yielding a non-significant trend (p < 0.1). It is important to note that at least half of the studies (Davidson et al., Reference Davidson, Gadde, Fairbank, Krishnan, Califf and Binanay2002; Detke et al., Reference Detke, Wiltse, Mallinckrodt, McNamara, Demitrack and Bitter2004; Murray et al., Reference Murray, von Arbin, Bartfai, Berggren, Landtblom and Lundmark2005) included in Deshauer et al.'s meta-analysis excluded patients from follow-up if they did not respond to the acute phase blinded treatment, even if their short-term data were used in the intention-to-treat analysis. It is unclear whether similar exclusions occurred in the other three studies. Differences in treatment dropouts and study completers were not significant between the arms but many participants did not complete the final outcome assessment, ranging from 23% to 73% with mean 52% (Deshauer et al., Reference Deshauer, Moher, Fergusson, Moher, Sampson and Grimshaw2008). The studies evidenced moderate risk of bias and most failed to report key methodologic issues, for instance on what happened to those who did not improve during the acute phase. Hence it is unclear how valid and generalizable the results of these prolonged aADM trials are. Compared to remission rates in the 8–12 week trials, Deshauer et al.'s findings suggest that the placebo group tends to ‘catch up’, possibly due to the natural history of depression.

Placebo effects and spontaneous remission; net and gross efficacy

Response rates (⩾50% improvement) in PLA arms of RCTs of depression are large, often between 35% and 40% in the PLA arm (Furukawa et al., Reference Furukawa, Cipriani, Atkinson, Leucht, Ogawa, Takeshima and Salanti2016; Levkovitz, Tedeschini, & Papakostas, Reference Levkovitz, Tedeschini and Papakostas2011; Rutherford & Roose, Reference Rutherford and Roose2013). The response in PLA arms is due to multiple factors: (i) spontaneous remission referring to the phenomenon that some people get better in the absence of any treatment as part of the natural course of their depression; (ii) regression to the mean, in which individuals who are far from the mean at the baseline measurement tend to be closer to the mean at the next measurement, in as far as their extreme baseline scores were inflated by measurement error; and (iii) non-specific treatment effects, which consist of improvement due to being treated by a professional including explanation, attention, support, positive expectations, amongst others. It is possible that the placebo response in ADM trials is larger than with real-world patients because patients in regulatory trials are not fully representative of real-world patients (Rush et al., Reference Rush, Trivedi, Wisniewski, Nierenberg, Stewart and Warden2006; Van der Lem, van der Wee, Van Veen, & Zitman, Reference Van der Lem, van der Wee, Van Veen and Zitman2012). The large placebo effect indicates that many respond not because of the specific ADM treatment, but because of other effects. This makes it difficult for trials to establish the excess value of specific treatments and provides critics with ammunition to minimize the benefits of ADMs by focusing on net efficacy and advocates to maximize benefits by focusing on gross efficacy.

Continuation and maintenance ADM trials: efficacy and limitations

The most recent meta-analysis evaluated 72 cADM and 37 mADM trials with 14.450 and 7253 participants, respectively (Sim, Lau, Sim, Sum, & Baldessarini, Reference Sim, Lau, Sim, Sum and Baldessarini2016). In the 72 cADM trials, lasting 33 weeks on average, the pooled advantage of cADM over discontinuation to placebo in terms of the relative response rate (RR) was 1.90, CI 1.73–2.08, NNT 4.4; in the 37 mADM trials lasting 27 months on average, it was 2.03, CI 1.80–2.28; NNT = 3.8; with minor differences among drug types (Sim et al., Reference Sim, Lau, Sim, Sum and Baldessarini2016). The authors did not distinguish between the two designs that have been used to investigate the efficacy of continuation and maintenance treatment with ADMs (cmADMs): the discontinuation and the extension design (Fig. 1).

Fig. 1. Illustration of designs.

In the discontinuation (or placebo-substitution) design, patients who respond to aADM are randomized to cmADM v. PLA. Meta-analyses document substantially higher relapse/recurrence rates during the post-randomization 6–12 months in placebo-substitution arms (~42%) compared to the cmAD arms (~22%) (Geddes et al., Reference Geddes, Carney, Davies, Furukawa, Kupfer, Frank and Goodwin2003; Glue, Donovan, Kolluri, & Emir, Reference Glue, Donovan, Kolluri and Emir2010; Hansen et al., Reference Hansen, Gaynes, Thieda, Gartlehner, Deveaugh-Geiss, Krebs and Lohr2008; Kaymaz, van Os, Loonen, & Nolen, Reference Kaymaz, van Os, Loonen and Nolen2008). The major problem of the discontinuation design, however, is that AD withdrawal symptoms may masquerade as relapse/recurrence (Fava, Gatti, Belaise, Guidi, & Offidani, Reference Fava, Gatti, Belaise, Guidi and Offidani2015; Greenhouse & Meyer, Reference Greenhouse and Meyer1991). Its bias potential is unclear because of uncertainty about the risk period of withdrawal phenomena (Borges et al., Reference Borges, Chen, Laughren, Temple, Patel, David and Khin2014; El-Mallakh, Waltrip, & Peters, Reference El-Mallakh, Waltrip and Peters1999; Kaymaz et al., Reference Kaymaz, van Os, Loonen and Nolen2008). Excess relapse/recurrence rates in the placebo-substitution arms compared to the cmADM arms are strongest in the first month after discontinuation and gradually drop to approximately zero in the next 3–6 months (Borges et al., Reference Borges, Chen, Laughren, Temple, Patel, David and Khin2014; El-Mallakh et al., Reference El-Mallakh, Waltrip and Peters1999; Kaymaz et al., Reference Kaymaz, van Os, Loonen and Nolen2008).

Extension trials start as regular double-blind, placebo-controlled aADM trials that continue blinded treatment in responders to ADMs (~45%) and responders to PLA (~32%) for another 5–12 months (Zimmerman, Posternak, & Ruggero, Reference Zimmerman, Posternak and Ruggero2007). In five double-blind extension trials including 901 participants and lasting on average 35 weeks, relapse/recurrence rates in those who initially remitted averaged 25% in the PLA arm v. 8% in the ADM arm, showing a strong benefit for cADM relative to cPLA (RR = 2.0) (Zimmerman et al., Reference Zimmerman, Posternak and Ruggero2007). The extension design lacks the withdrawal issues of discontinuation trials, but still has the limitation that acute phase non-responders are not included and followed. This is a serious limitation as about half do not remit on acute phase treatment (Khin, Chen, Yang, Yang, & Laughren, Reference Khin, Chen, Yang, Yang and Laughren2011; Levkovitz et al., Reference Levkovitz, Tedeschini and Papakostas2011; Pigott, Leventhal, Alter, & Boren, Reference Pigott, Leventhal, Alter and Boren2010; Rush et al., Reference Rush, Trivedi, Wisniewski, Nierenberg, Stewart and Warden2006). Treatment completion rates were often <50%.

In conclusion, compared to discontinuation trials, extension trials provide stronger evidence that cmADM in responders to aADM treatment reduce relapse/recurrence risk relative to cmPLA in aPLA responders because aADM responders were not switched to PLA, excluding confounding effects of ADM withdrawal. However, extension trials are relatively rare and patients on cmPLA, although they initially responded, may slowly realize that they are using placebos, with possibly negative psychological consequences.

Naturalistic long-term outcome studies

Many naturalistic outcome studies (i.e. non-randomized observational cohort or longitudinal studies) have charted the long-term course and outcome of depression (⩾2 years). Reviews typically report worse long-term outcomes for depressed patients who are managed in specialty settings compared to those managed in primary care (e.g. Brodaty, Luscombe, Peisah, Anstey, & Andrews, Reference Brodaty, Luscombe, Peisah, Anstey and Andrews2001; Hardeveld, Spijker, De Graaf, Nolen, & Beekman, Reference Hardeveld, Spijker, De Graaf, Nolen and Beekman2010; Piccinelli & Wilkinson, Reference Piccinelli and Wilkinson1994; Steinert, Hofmann, Kruse, & Leichsenring, Reference Steinert, Hofmann, Kruse and Leichsenring2014; van Weel-Baumgarten, Schers, van den Bosch, van den Hoogen, & Zitman, Reference van Weel-Baumgarten, Schers, van den Bosch, van den Hoogen and Zitman2000). Piccinelli and Wilkinson, reviewing 50 outcome studies, reported weighted averages of 43% sustained recovery and 15% persistent depression at 1-year follow-up, and 24% sustained recovery and 12% persistent depression at >10-year follow-up. Other reviews report recurrence rates of up to 85% in 15 years for depression treated in specialty settings (Hardeveld et al., Reference Hardeveld, Spijker, De Graaf, Nolen and Beekman2010; Mulder & Frampton, Reference Mulder and Frampton2014). In contrast, two major reviews of adult depression in general practice and the community (20 studies of almost 6000 participants, most with 3–7 years follow-ups) found that although 10–17% had a chronic course, up to 85% recovered for some time and that 35–60% experienced stable recovery (Steinert et al., Reference Steinert, Hofmann, Kruse and Leichsenring2014; van Weel-Baumgarten et al., Reference van Weel-Baumgarten, Schers, van den Bosch, van den Hoogen and Zitman2000). Regarding outcome of non-treated community and general practice cases, multiple studies suggest that many depressive episodes (80–50%) are self-limiting as they remit within 3–12 months (Goldberg, Privett, Ustun, Simon, & Linden, Reference Goldberg, Privett, Ustun, Simon and Linden1998; Regier et al., Reference Regier, Kaelber, Rae, Farmer, Knauper, Kessler and Norquist1998; Sareen et al., Reference Sareen, Henriksen, Stein, Afifi, Lix and Enns2013; Spijker et al., Reference Spijker, de Graaf, Bijl, Beekman, Ormel and Nolen2002; Wang et al., Reference Wang, Henriksen, ten Have, de Graaf, Stein, Enns and Sareen2017; Whiteford et al., Reference Whiteford, Harris, McKeon, Baxter, Pennell, Barendregt and Wang2013). For instance, Whiteford's (2013) meta-analysis of 19 studies of untreated depression in general practice settings estimated that 23% remit within 3 months, 32% within 6 months, and 53% within 12 months.

Observational studies typically report similar or worse outcomes for treated v. untreated depressive episodes in primary care and community samples (Boerema et al., Reference Boerema, ten Have, Kleiboer, de Graaf, Nuyen, Cuijpers and Beekman2017; Goldberg et al., Reference Goldberg, Privett, Ustun, Simon and Linden1998; Hughes & Cohen, Reference Hughes and Cohen2009; Wang et al., Reference Wang, Henriksen, ten Have, de Graaf, Stein, Enns and Sareen2017). Hughes and Cohen's review compared AD-treated cases in 12 naturalistic cohorts of treated patients (n = 3901 at final follow-up) and three non-ADM-treated samples of 1160 patients. Most participants were white females with one inpatient stay. Frequency, duration, and severity of episodes varied substantially. Outcomes were unrelated to treatment status, and not better in ADM-treated cohorts relative to non-ADM-treated samples. Heterogeneity of study designs and outcome definitions were large and hampered statistical analysis. Recent naturalistic long-term outcome studies comparing ADM-treated and untreated individuals with diagnosed depression consistently corroborate the findings of Hughes and Cohen's review (Bockting, Hollon, Jarrett, Kuyken, & Dobson, Reference Bockting, Hollon, Jarrett, Kuyken and Dobson2015; Hengartner, Angst, & Roessler, Reference Hengartner, Angst and Roessler2018; Nuijen, ten Have, Tuithof, van Dorsselaer, & van Bon-Martens, Reference Nuijen, ten Have, Tuithof, van Dorsselaer and van Bon-Martens2014; Verduijn et al., Reference Verduijn, Verhoeven, Milaneschi, Schoevers, van Hemert, Beekman and Penninx2017; Vittengl, Reference Vittengl2017; Wang et al., Reference Wang, Henriksen, ten Have, de Graaf, Stein, Enns and Sareen2017). Statistical adjustment for confounders (e.g. educational level, marital status, baseline severity, family history) did not attenuate the association of ADM treatment with outcome (Hengartner et al., Reference Hengartner, Angst and Roessler2018; Vittengl, Reference Vittengl2017). However, statistical adjustment cannot adjust for unmeasured differences between ADM v. non-ADM (known as confounding by indication) and it is very likely that untreated cases with established major depressive disorder are less vulnerable and more resilient than treated cases (e.g. Goldberg et al., Reference Goldberg, Privett, Ustun, Simon and Linden1998; Ormel, Oldehinkel, Brilman, & van den Brink, Reference Ormel, Oldehinkel, Brilman and van den Brink1993).

Efficacy and limitations of psychological treatments

Critics of ADMs often advocate psychological treatments, especially cognitive behavioral therapy (CBT), interpersonal therapy (IPT), mindfulness-based cognitive therapy (MBCT), and problem-solving therapy (PST). But how efficacious are these treatments and do they offer a serious alternative to cmADM or are they vulnerable to similar criticisms as the ADM? Compared to PLA, care-as-usual (CAU), and waiting-list controls, meta-analyses indicate that short-term efficacy of psychological treatments is modest in magnitude and comparable to aADM (e.g. Barth et al., Reference Barth, Munder, Gerger, Nueesch, Trelle and Znoj2013; Cuijpers et al., Reference Cuijpers, Karyotaki, Weitz, Andersson, Hollon and van Straten2014; Cuijpers, Cristea, Karyotaki, Reijnders, & Huibers, Reference Cuijpers, Cristea, Karyotaki, Reijnders and Huibers2016). For instance, Barth's meta-analysis identified 198 studies, including 15 118 adult patients with depression. All major types of psychotherapeutic interventions were superior to waitlist control condition with moderate-to-large effects (range d = 0.62–0.92). Studies with larger samples and higher quality (blinded observers, self-report measures, adequate concealed allocation, etc.) reported substantially smaller effects of about d = 0.30. Cuijpers et al.'s meta-analysis of CBT for MDD, the most studied psychological treatment, including 63 studies and approximately 4000 participants, reported a pooled effect size of 0.75 (95% CI 0.64–0.87; NNT = 3.9). High-quality studies using CAU as a control had significantly smaller effect sizes (0.43, NNT 7.3) and so did trials using PLA as the control (0.55, NNT = 5.5). Direct comparisons between ADM and major psychological treatments indicate that they are about equally efficacious in the short-term treatment of major depression (Cuijpers et al., Reference Cuijpers, Berking, Andersson, Quigley, Kleiboer and Dobson2013a, Reference Cuijpers, Hollon, van Straten, Bockting, Berking and Andersson2013b), although an earlier meta-analysis found a slight excess benefit for SSRIs (Cuijpers, van Straten, van Oppen, & Andersson, Reference Cuijpers, van Straten, van Oppen and Andersson2008). Dropout rates are typically smaller in psychological treatments (Cuijpers et al., Reference Cuijpers, van Straten, van Oppen and Andersson2008). Similar to ADM trials, accumulating biases have also resulted in overestimation of the short-term efficacy of psychological treatments^, (Cuijpers et al., Reference Cuijpers, Cristea, Karyotaki, Reijnders and Huibers2016; Driessen, Hollon, Bockting, Cuijpers, & Turner, Reference Driessen, Hollon, Bockting, Cuijpers and Turner2015).

Three meta-analyses examined relapse/recurrence in responders to psychological treatments (largely CBT, BA, and IPT) (Cuijpers et al., Reference Cuijpers, Berking, Andersson, Quigley, Kleiboer and Dobson2013a; Sim et al., Reference Sim, Lau, Sim, Sum and Baldessarini2016; Vittengl, Clark, Dunn, & Jarrett, Reference Vittengl, Clark, Dunn and Jarrett2007). Vittengl et al.'s meta-analysis of 28 studies including 1880 adults found substantial levels of symptom return after discontinuation of aCBT in CBT responders (29% within 1 year and 54% within 2 years). These rates appeared comparable to those associated with other depression-specific psychotherapies but lower than those associated with aADM. Seven trials compared aCBT and aADM. Averaging separately by treatment type yielded relapse/recurrence rates of 39% for aCBT and 61% for aADM over an average of 68 weeks.

Cuijpers et al.'s (Reference Cuijpers, Hollon, van Straten, Bockting, Berking and Andersson2013b) meta-analysis examined the long-term (6–18 months) effects of aCBT v. aADM with and without cmADM (nine studies, 506 patients) (Cuijpers et al., Reference Cuijpers, Hollon, van Straten, Bockting, Berking and Andersson2013b). Short-term outcomes were comparable although dropout was lower in CBT. Patients who received acute phase CBT were significantly less likely to relapse after discontinuation than patients who were withdrawn from ADM (OR = 2.61, 95% CI 1.58–4.31, NNT = 5). There was a non-significant trend favoring prior CBT over acADM (five studies) (p < 0.1; OR = 1.62, 95% CI 0.97–2.72; NNT = 10) (p < 0.1; OR = 1.62, 95% CI 0.97–2.72; NNT = 10); suggesting that aCBT might be even more efficacious than continuation ADM in preventing relapse. Finally, Sim et al.'s meta-analysis included 22 psychosocial treatment trials [CBT, IPT, MBCT, psychoeducation that followed 1969 (mostly) remitted patients with recurrent depression across 24 months (Sim et al., Reference Sim, Lau, Sim, Sum and Baldessarini2016)]. Treatments were slightly more effective than controls in preventing relapse/recurrence: pooled RR = 1.39 (1.13–1.70); substantially less effective than in Vittengl et al.'s meta-analysis.

Some trials have recently examined whether preventive variants of CBT/MBCT have better long-term outcomes than cmADM. These trials randomized aADM responders into two or three arms: continuing ADM v. preventive CBT/MBCT with or without continuing ADM (Bockting et al., Reference Bockting, Klein, Elgersma, van Rijsbergen, Slofstra and Ormel2018; Huijbers et al., Reference Huijbers, Spinhoven, Spijker, Ruhe, van Schaik and van Oppen2016; Kuyken et al., Reference Kuyken, Hayes, Barrett, Byng, Dalgleish, Kessler and Byford2015; Segal et al., Reference Segal, Bieling, Young, MacQueen, Cooke, Martin and Levitan2010). A recent meta-analysis of four preventive MBCT trials (637 participants, 266 relapses during the 60-week follow-up period) found a statistically significant advantage for MBCT compared to cmADM, suggesting that MBCT appears efficacious as a treatment for relapse prevention for those with recurrent depression in (partial) remission (Kuyken et al., Reference Kuyken, Warren, Taylor, Whalley, Crane and Bondolfi2016). Additionally, in a recent trial not included in that review, cmADM was not superior to preventive CT administered while tapering off ADM [relapse/recurrence risks over 15–24 months (60% v. 63%) (Bockting et al., Reference Bockting, Klein, Elgersma, van Rijsbergen, Slofstra and Ormel2018)]. However, tapering was often unsuccessful as many patients continued or resumed cmADM, which may have decreased symptom return in the taper PCT/MBCT arm. On the other hand, misclassified withdrawal symptoms may have increased symptom return in this arm.

Placebo-controlled trials with relevant long-term outcomes

Logistically extremely difficult but in our opinion highly desirable are randomized placebo-controlled long-term outcome studies of ADM and psychotherapy. Long-term means at least 1 year follow-up, preferably 2 years. If sufficiently powered, an additional arm of combined treatment (COM) could be added. An example of a top priority trial is randomization of currently ADM-free depressed individuals, stratified by history of ADM treatment (never v. ever), to ADM v. CBT v. PLA. Acute phase non-responders, irrespective of treatment, will be offered the treatment they prefer, but kept in the study and administered the very same follow-up assessments, thereby addressing the major deficiency of previous trials. Alternatively, non-responders could be offered the other treatment modality in addition to the allotted treatment (COM) and provide the same for the PLA non-responders. Primary outcomes should be psychopathology, role function, adverse events, and acceptability. Machine learning can be used to generate selection algorithms that identify the optimal intervention for each individual who does not remit on COM and use those algorithms in the patient by mechanism interactions to improve the power of the tests of mediation (Cohen & DeRubeis, Reference Cohen and DeRubeis2018). If putative effect moderators are included at baseline, much can be learned about interactions between treatment modality and effect moderators. Finally, it is important to perform cost-effectiveness analyses as well because what drives shifts in practice are often shifts in reimbursement and what drives shifts in reimbursement is evidence of cost-effectiveness. As an anonymous program officer at NIMH once said: ‘You cannot herd cats but you can move their food.’

Multiple problems threaten the feasibility and validity of the proposed study: dropouts, protocol violations, imbalance between the arms at the entry of the continuation phase, complex logistics, and ethical concerns. Since 2000, it has been difficult to keep patients on PLA for more than a few months because it is considered unethical to withhold effective treatment. However, the modest excess efficacy of ADM and CBT relative to PLA implies that the majority of patients who respond to specific treatments do so for other reasons; that is, due to non-specific treatment effects and spontaneous remission, and not a ‘specific’ ADM or CBT effect. Furthermore, ethical concerns are manageable using patient-level stopping rules. The modest net efficacy of ADM or CBT (specificity) reduces the threat of imbalance as well, as many will respond to PLA, and balance can be managed with propensity analyses. Logistical problems necessitate the cooperation of many primary care providers from large catchment areas.

Inventive analyses of observational databases

The unsurpassed advantage of randomization is that the resulting groups are comparable (apart from differences due to chance). This allows the unambiguous attribution of differences in outcomes between groups to differences in treatment. Observational, non-randomized research can never rule out that the groups differ in characteristics (confounders) that affect both treatment and outcome. Notorious ‘confounders’ are severity and history. However, statisticians have recently developed methods that allow, under certain conditions and with reasonable certainty, causal statements about treatment effects based on observational data, even if not all potential ‘confounders’ are measured. The methods are instrumental variable analysis (Martens, Pestman, de Boer, Belitser, & Klungel, Reference Martens, Pestman, de Boer, Belitser and Klungel2006) and propensity score analysis (Austin, Reference Austin2011). Potentially relevant observational databases can be generated from routine outcome measurement (ROM), patient cohort data, and longitudinal population-based studies, provided they assess long-term outcome.

Cohort studies including effect moderators

Predicting who will benefit from which treatment is difficult. As a result, personalized treatment (‘precision medicine’) is not yet possible with depression although there are interesting developments in this direction (Cohen & DeRubeis, Reference Cohen and DeRubeis2018; DeRubeis et al., Reference DeRubeis, Cohen, Forand, Fournier, Gelfand and Lorenzo-Luaces2014; Fournier et al., Reference Fournier, DeRubeis, Shelton, Hollon, Amsterdam and Gallop2009; Kessler, Reference Kessler2018). Cohort studies of patients with the same diagnosis can provide insight if putative effect moderators are included at baseline. If treatments become more precise and personalized, the relevance of the ADM controversy will decline. Relatively easy to measure potential effect moderators are clinical characteristics, personality, biomarkers to be determined from bodily fluids including DNA, social network, and support characteristics.

Psychological help as a first-line treatment

An approach that may also reduce the relevance of the controversy as well is to offer short-term psychological treatment as the first-line treatment step for all patients with depression unless they refuse or cannot receive psychological treatment or have characteristics shown to predict a better response to ADM. ADM then remains reserved for patients who do not want, get, or respond to this first step. It is of great importance which of the non-responders to psychological treatment improves on ADM and whether psychological treatment can subsequently contribute to the phasing out of their ADM use. Severely ill patients could be started on combination of ADM and psychotherapy. Some guidelines actually do advise psychological treatment for adolescents with depression as the first-line treatment step, and other guidelines, such as the UK NICE guideline for depression and the Dutch multidisciplinary guideline for depression, also recommend psychological treatment as the default first step for adults with mild-to-moderate depression(Clark, Reference Clark2018; NICE, 2011; Spijker et al., Reference Spijker, Meeuwissen, Aalbers, van Avendonk, van Bon and Huson2019).