Major depressive disorder (MDD) is a common mental disorder entwined with impaired functioning in academic, occupational, and social domains (Kupferberg, Bicks, & Hasler, Reference Kupferberg, Bicks and Hasler2016) and compromised physical health (Zainal & Newman, Reference Zainal and Newman2022) and quality of life (Hohls, Konig, & Hajek, Reference Hohls, Konig and Hajek2023). Economically, MDD imposes a hefty financial burden on government budgets due to substantial healthcare utilization costs, increased absenteeism and presenteeism, and reduced work productivity (Greenberg et al., Reference Greenberg, Fournier, Sisitsky, Simes, Berman, Koenigsberg and Kessler2021). One of the most devastating consequences of MDD is mortality due to a suicide attempt and other related causes (Chiu et al., Reference Chiu, Liu, Li, Tsai, Chen and Kuo2023). Determining effective treatments for MDD to reduce the risk for or prevent suicide attempts and other serious psychiatric adverse events is thus essential.
The first-line treatments for MDD include psychotherapy-only, especially cognitive-behavioral and related therapies, antidepressant medication (ADM)-only, and combined treatment (GBD 2017 Disease and Injury Incidence and Prevalence Collaborators, 2018). Meta-analyses of randomized controlled trials (RCTs) that examined MDD treatment response or remission rates as primary outcomes of interest showed that combined treatment conferred a substantially stronger aggregate effect than ADM-only, psychotherapy-only had a better aggregate effect than ADM-only, and the aggregate effect between combined treatment and psychotherapy-only was not different (Cuijpers et al., Reference Cuijpers, Miguel, Harrer, Plessen, Ciharova, Ebert and Karyotaki2023; Furukawa et al., Reference Furukawa, Shinohara, Sahker, Karyotaki, Miguel, Ciharova and Cuijpers2021; Karyotaki et al., Reference Karyotaki, Smit, Holdt Henningsen, Huibers, Robays, de Beurs and Cuijpers2016). Collectively, such findings might translate to the possibility that MDD treatments that comprise psychotherapy, especially those that embody cognitive-behavioral and related theories, would be associated with reduced risk of suicide attempts and associated serious psychiatric adverse events.
Only six meta-analyses have determined if psychotherapy or ADM monotherapies mitigated the risk for suicide attempts and related serious psychiatric adverse events. Five recent meta-analyses showed that depressed patients randomized to psychotherapy-only, such as cognitive-behavioral therapies (CBT; Calati & Courtet, Reference Calati and Courtet2016; Gotzsche & Gotzsche, Reference Gotzsche and Gotzsche2017), dialectical behavioral therapy (Kothgassner, Robinson, Goreis, Ougrin, & Plener, Reference Kothgassner, Robinson, Goreis, Ougrin and Plener2020), psychodynamic-oriented therapies (Briggs et al., Reference Briggs, Netuveli, Gould, Gkaravella, Gluckman, Kangogyere and Lindner2019), and suicide-focused psychotherapies (Riblet, Shiner, Young-Xu, & Watts, Reference Riblet, Shiner, Young-Xu and Watts2017), were less likely to engage in suicide attempt, self-harm behaviors, and encounter other serious psychiatric adverse events compared to their counterparts assigned to treatment-as-usual or waitlist. However, another recent meta-analysis showed that the probability for depressed patients randomized to receive ADM to experience any of these serious psychiatric adverse events was not different from those who received placebos (Braun, Bschor, Franklin, & Baethge, Reference Braun, Bschor, Franklin and Baethge2016). Further, a major limitation of these meta-analyses was the exclusion of RCTs that examined the comparative effects of psychotherapy-only, ADM-only, and combined treatment.
Accordingly, the present exploratory meta-analysis evaluated the comparative effects of psychotherapy-only, ADM-only, and combined treatment on suicide attempts and other serious psychiatric adverse events. Such serious psychiatric adverse events included psychiatric emergency department (ED) visits, psychiatric hospitalization, and/or suicide deaths among depressed patients. Knowing what treatment option works best to decrease the likelihood of suicide attempts and other serious psychiatric adverse events at the aggregate level could aid with successful suicide prevention efforts for depressed patients.
Method
Procedures and eligibility criteria
The present meta-analysis protocol was preregistered on the Open Science Framework (OSF; https://osf.io/c7z8d) and the International Prospective Register of Systematic Reviews (PROSPERO; CRD42023443819). The conventional approach instructed by Cochrane was applied for the analyses, and the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines were followed in reporting the methods and results (Fig. 1; Moher et al., Reference Moher, Shamseer, Clarke, Ghersi, Liberati, Petticrew and Group2015; Page et al., Reference Page, McKenzie, Bossuyt, Boutron, Hoffmann, Mulrow and Moher2021). The inclusion criteria included clinical trials that reported on number of patients with suicide attempts and/or other serious psychiatric adverse events (i.e. psychiatric ED visit, psychiatric hospitalization, or suicide death) at post-treatment in the intervention and control arms. Baseline randomization to psychotherapy-only, ADM-only, or combined treatment and recruitment of patients with a primary diagnosis of a depressive disorder was required. Following best practices, relevant RCTs that reported zero suicide attempts and serious psychiatric adverse events in the treatment and control arms were included, given that this is a more conservative method and enhances the generalizability of the results (Friedrich, Adhikari, & Beyene, Reference Friedrich, Adhikari and Beyene2007). Exclusion criteria included studies that used waitlist, treatment-as-usual, placebo, or another active control (e.g. ketamine) as a comparator, non-RCTs, and RCTs that did not recruit patients with a primary diagnosis of a depressive disorder. Although challenges in utilizing RCTs to examine rare outcomes such as suicide attempts and other serious psychiatric adverse events were acknowledged, RCTs can provide essential information on the comparative risk of such serious psychiatric adverse events when evaluating two or more treatment options. Nonetheless, the present meta-analysis centered on RCTs given that it is the gold standard design to determine comparative effectiveness (or efficacy; Rosen, Manor, Engelhard, & Zucker, Reference Rosen, Manor, Engelhard and Zucker2006), and an adequate number of RCTs (i.e. 34 eligible RCTs; Fig. 1) have examined how psychotherapy-only, ADM-only, and combined treatment might reduce the risk of suicide attempts and other serious psychiatric adverse events. Table 1 summarizes the 34 eligible RCTs in the present meta-analysis and relays information about the number of suicide attempts, psychiatric ED visits, psychiatric hospitalization, and/or suicide deaths extracted from each RCT.
PRISMA flow diagram. PRISMA, Preferred Reporting Items for Systematic Reviews and Meta-Analyses; RCT, randomized controlled trial; ADM, antidepressant medication; PSY, psychotherapy; SA, suicide attempt.
Summary of RCTs that examined comparative average treatment effect (ATE) of antidepressant medication (ADM) and/or psychotherapy (PSY) predicting suicide attempt (SAs), psychiatric emergency department (ED) visit, psychiatric hospitalization, and/or suicide death
ATE, average treatment effect (or efficacy); ADM, antidepressant medication; CBT, individual cognitive behavioral therapy; CI, confidence interval; CT, cognitive therapy; ED, emergency department; GCBT, group CBT; MDD, major depressive disorder; OR, odds ratio; PCP, primary care physician; PSY, psychotherapy; RCT, randomized controlled trial; SA, suicide attempt; SP, supportive psychotherapy; SSRI, selective serotonin reuptake inhibitor.
aOnline Supplementary Table S1 details blow-by-blow how each study reported the rates of suicide attempt(s) and/or other serious psychiatric adverse event(s) (psychiatric ED visit, psychiatric hospitalization, or suicide death) in each arm.
b These RCTs were not included in the current meta-analysis as the exact rates of SA and/or psychiatric hospitalization in each intervention and comparator arm were not reported.
Study selection
Cochrane Library, Google Scholar, PubMed, and ScienceDirect were searched from the inception of each database until 30 June 2023, to identify published articles evaluating our research question. The following search terms and Boolean terms were used: (‘suicide attempt’) AND (‘antidepressant’) AND (‘psychotherapy’) AND (‘trial’) AND (‘depression’) AND (‘serious adverse events’). A highly sensitive search approach that prioritized the detection of RCTs in these online databases was used (Riblet et al., Reference Riblet, Shiner, Young-Xu and Watts2017). Online Supplementary Table S1 details blow-by-blow how each study measured and reported the rates of suicide attempt(s) and/or other serious psychiatric adverse event(s) in each arm.
Intervention arms
Concerning psychotherapy type, the most common face-to-face psychotherapy was individual CBT (n = 11, 32.4%), followed by individual, interpersonal psychotherapy (n = 5, 14.7%), group CBT (n = 1, 2.9%), and psychotherapy without a specified theoretical orientation (n = 1, 2.9%). The remaining 16 studies (47.1%) did not use psychotherapy-only as a comparator. Regarding ADM type, most studies did not specify the ADM type (n = 15, 44.1%) followed by selective serotonin reuptake inhibitors (i.e. citalopram, fluoxetine, imipramine, sertraline; n = 12, 35.3%), and venlafaxine (a serotonin–norepinephrine reuptake inhibitor; n = 1, 2.9%). The remaining six studies (17.6%) did not use ADM-only as a comparator.
Data analyses
Using the Peto approach (Bradburn, Deeks, Berlin, & Russell Localio, Reference Bradburn, Deeks, Berlin and Russell Localio2007; Lane, Reference Lane2013), summary odds ratios (ORs) with 95% confidence intervals (CIs) and p values were computed. Standard meta-analytic techniques (e.g. calculating risk ratio) are not recommended to determine rare outcomes such as suicide attempts and other serious psychiatric adverse events (i.e. psychiatric ED visits, psychiatric hospitalization, and/or suicide death; Bradburn et al., Reference Bradburn, Deeks, Berlin and Russell Localio2007; Lane, Reference Lane2013). The Peto approach is an appropriate and robust alternative option for meta-analyzing data when event rates are less than 1% (Bradburn et al., Reference Bradburn, Deeks, Berlin and Russell Localio2007; Lane, Reference Lane2013). No continuity correction for treatment arms with zero negative outcome events was applied since prior research showed that it would bias estimates with the Peto approach (Friedrich et al., Reference Friedrich, Adhikari and Beyene2007; Sweeting, Sutton, & Lambert, Reference Sweeting, Sutton and Lambert2004).
The rmeta R package was used to conduct these random-effects meta-analyses for rare events and to generate forest plots. The conventional definition of an OR was used to interpret the present findings (i.e. likelihood of the occurrence of suicide attempts and other serious psychiatric adverse events v. likelihood of the non-occurrence of suicide attempts and other serious psychiatric adverse events in the intervention arm v. comparator arm). An OR value < 1 indicated that the relative odds of suicide attempts and other serious psychiatric adverse events were lower in the treatment arm v. comparator arm, and the opposite was true with an OR value > 1. The OR was considered to be statistically significant if the 95% CIs did not cross 1 (Szumilas, Reference Szumilas2010). Standard thresholds of Cochrane's Q p value < 0.10 and I 2 > 50% were used to determine between-study heterogeneity (Deeks, Higgins, & Altman, Reference Deeks, Higgins, Altman, J. P. T., J, J, M., T., M. J. and V. A.2019).
Contour-enhanced funnel plots were created to evaluate publication bias to determine if any funnel plot asymmetry indicated substantial publication bias (Peters, Sutton, Jones, Abrams, & Rushton, Reference Peters, Sutton, Jones, Abrams and Rushton2008). Moreover, Harbord's modified assessment of small study effects was used to test for publication bias for binary outcomes (Harbord, Egger, & Sterne, Reference Harbord, Egger and Sterne2006; Peters, Sutton, Jones, Abrams, & Rushton, Reference Peters, Sutton, Jones, Abrams and Rushton2006) with the meta R package (Balduzzi, Rucker, & Schwarzer, Reference Balduzzi, Rucker and Schwarzer2019). This approach utilized a modified linear regression analysis to detect significant funnel plot asymmetry that would denote publication bias (i.e. with a p < 0.05 value; Harbord et al., Reference Harbord, Egger and Sterne2006). The Harbord approach is stronger than other standard tests of publication bias under conditions of few events per RCT (Harbord et al., Reference Harbord, Egger and Sterne2006). Sensitivity analyses were also conducted with Duval and Tweedie trim-and-fill method (Duval & Tweedie, Reference Duval and Tweedie2000) to determine if the pattern of outcomes from the primary analyses changes after adjusting for any funnel plot asymmetry caused by publication biases. In addition, to assess study quality, the Cochrane Risk of Bias Tool was used (Higgins et al., Reference Higgins, Altman, Gøtzsche, Jüni, Moher, Oxman and Sterne2011) with the robvis R package to generate risk-of-bias plots (McGuinness & Higgins, Reference McGuinness and Higgins2020).
The heterogeneity of treatment effects was explored in the present meta-analysis. Specifically, we conducted subgroup analyses of the comparative treatment effects by age group (adults or youths), study duration (1.5–5, 6–12, or 13–60 months), and treatment setting (outpatient v. inpatient). In the case of continuous effect modifiers (moderators), our meta-regression analyses utilized a restricted maximum likelihood model, incorporating the Knapp–Hartung method (Borenstein, Hedges, Higgins, & Rothstein, Reference Borenstein, Hedges, Higgins and Rothstein2009). Specifically, we determined how mean age, percentage of females, prior suicide attempts, patients with recurrent v. first-onset MDD, and any psychiatric comorbidities might modify the average treatment effects.
Results
Study characteristics
Thirty-one out of 34 studies reported baseline sociodemographic variables. The sample age averaged 32.06 years (standard deviation [s.d.] = 14.57, range = 13.80–58.43), and percentage of females was 61.6% across those studies. Regarding study duration, 16 studies (47.1%) reported on 3–5-month follow-up outcomes, another 11 studies (32.4%) on 6–12-month follow-up outcomes, and the remaining seven studies (20.6%) on 13–60-month follow-up outcomes. Concerning RCT design, four were four-arm trials (11.8%), eight were three-arm trials (23.5%), and 22 were two-arm trials (64.7%). Most RCTs occurred within outpatient settings (n = 30, 88.2%) compared to inpatient locations (n = 4, 11.8%). The following variables were regarded as various indices of suicide risk (indexed as percentages): prior suicide attempts (number of studies [k] = 34, observed mean [M] = 17.72, standard deviation [s.d.] = 28.06, range = 0–100), recurrent v. first-onset MDD (k = 32, M = 61.60, s.d. = 39.87, range = 0–100), and any psychiatric comorbidities (k = 20, M = 59.32, s.d. = 29.93, range = 5.2–100).
Combined treatment v. psychotherapy-only
Seventeen RCTs determined the comparative treatment effects of combined treatment v. psychotherapy-only on suicide attempts and other serious psychiatric adverse events. Forty-seven out of 1273 (3.7%) in the combined treatment arm and 24 out of 1240 (1.9%) in the psychotherapy-only arm had suicide attempts and other serious psychiatric adverse events, and this difference was statistically significant (OR 1.96 [1.20–3.20], p = 0.012). There was no significant between-study heterogeneity (Q[df = 11] = 9.29, p = 0.595; I 2 = 0.0% [0.0–58.3]; τ 2 = 0 [0.00–1.56], p = 0.60) when comparing combined treatment to psychotherapy-only (Fig. 2). Subgroup analysis revealed that the average treatment effect of lower suicide attempts and other serious psychiatric adverse events in psychotherapy-only over combined treatment was statistically significant in youths (OR 1.86 [1.07–3.23]) but not in adults (OR 1.95 [0.52–7.35]; Table 2 part A). However, study duration (Table 2 part A), mean age, percentage of females, prior suicide attempts, patients with recurrent MDD, and psychiatric comorbidities (Table 3 part A) were non-significant effect modifiers. All studies within this set of combined treatment v. psychotherapy-only comparisons occurred in outpatient treatment settings.
Forest plot for average treatment effect comparing combined treatment v. psychotherapy-onlya. ADM, antidepressant medication; PSY, psychotherapy; OR, Peto odds ratio; CI, confidence interval; I 2, an index of between-study heterogeneity expressed in a percentage, such that higher values indicated more between-study heterogeneity; τ 2, tau-squared statistic, such that a statistically significant τ 2 indicated substantial between-study variance of the underlying distribution of true effect sizes. aEvents referred to the number of patient(s) (if any) within a specific treatment arm that had a negative treatment outcome (i.e. suicide attempt [SA], psychiatric emergency department [ED] visit, psychiatric hospitalization, and/or suicide death). The random-effects model only pertained to studies that examined SA as an outcome (Bockting et al., Reference Bockting, Klein, Elgersma, van Rijsbergen, Slofstra, Ormel and Burger2018; Cornelius et al., Reference Cornelius, Bukstein, Wood, Kirisci, Douaihy and Clark2009; Davey et al., Reference Davey, Chanen, Hetrick, Cotton, Ratheesh, Amminger and Berk2019; Deas et al., Reference Deas, Randall, Roberts and Anton2000; Hollon et al., Reference Hollon, DeRubeis, Evans, Wiemer, Garvey, Grove and Tuason1992; Iftene et al., Reference Iftene, Predescu, Stefan and David2015; Khazanov et al., Reference Khazanov, Xu, Hollon, DeRubeis and Thase2021; Kuyken et al., Reference Kuyken, Byford, Taylor, Watkins, Holden, White and Teasdale2008; Kuyken et al., Reference Kuyken, Hayes, Barrett, Byng, Dalgleish, Kessler and Byford2015; Lespérance et al., Reference Lespérance, Frasure-Smith, Koszycki, Laliberté, van Zyl, Baker and Guertin2007; March et al., Reference March, Silva, Petrycki, Curry, Wells, Fairbank and Severe2004; March et al., Reference March, Silva, Petrycki, Curry, Wells, Fairbank and Severe2007; Melvin et al., Reference Melvin, Tonge, King, Heyne, Gordon and Klimkeit2006; Vitiello et al., Reference Vitiello, Silva, Rohde, Kratochvil, Kennard, Reinecke and March2009; Vitriol et al., Reference Vitriol, Ballesteros, Florenzano, Weil and Benadof2009). There were too few studies that examined psychiatric ED visit (Riggs et al., Reference Riggs, Mikulich-Gilbertson, Davies, Lohman, Klein and Stover2007), psychiatric hospitalization (Browne et al., Reference Browne, Steiner, Roberts, Gafni, Byrne, Dunn and Kraemer2002; Hollon et al., Reference Hollon, DeRubeis, Evans, Wiemer, Garvey, Grove and Tuason1992; Huijbers et al., Reference Huijbers, Spinhoven, Spijker, Ruhe, van Schaik, van Oppen and Speckens2015), and suicide death (Bockting et al., Reference Bockting, Klein, Elgersma, van Rijsbergen, Slofstra, Ormel and Burger2018; Hollon et al., Reference Hollon, DeRubeis, Evans, Wiemer, Garvey, Grove and Tuason1992; Kuyken et al., Reference Kuyken, Hayes, Barrett, Byng, Dalgleish, Kessler and Byford2015; Vitiello et al., Reference Vitiello, Silva, Rohde, Kratochvil, Kennard, Reinecke and March2009) to conduct a random-effects model meta-analysis separately by those adverse outcomes.
Subgroup analysis
k, Number of studies; OR, Peto odds ratio; CI, lower and upper limits of the 95% confidence interval; Q, Cochrane's heterogeneity index denoting within-subgroup heterogeneity (Q WITHIN) or between-subgroup heterogeneity (Q BETWEEN); I 2, index of between-study heterogeneity; ADM, antidepressant medication.
* Significant estimates at the p < 0.05 α level (two-tailed).
Meta-regression analysis
β, meta-regression estimate; CI, lower and upper limits of the 95% confidence interval; p, p value associated with β; %, percentage of; ADM, antidepressant medication; MDD, major depressive disorder; MH, mental health.
* Significant estimates at the p < 0.05 α level (two-tailed).
Combined treatment v. ADM-only
Twenty-three RCTs examined the comparative treatment effects of combined treatment v. ADM-only. In total, 133 out of 2213 (6.0%) in the combined treatment arm and 162 out of 1855 (8.7%) in the ADM-only arm had suicide attempts and other serious psychiatric adverse events, and this difference was statistically significant (OR 0.74 [0.56–0.96], p = 0.029). There was no significant between-study heterogeneity (Q[df = 18] = 15.96, p = 0.595; I 2 = 0.0% [0.0–48.9]; τ 2 = 0.04 [0.00–0.36], p = 0.60) when comparing combined treatment to ADM-only (Fig. 3). Subgroup analysis revealed that the average treatment effect of lower suicide attempts and other serious psychiatric adverse events in combined treatment over ADM-only was statistically significant in adults (OR 0.66 [0.45–0.96]) but not in youths (OR 0.80 [0.51–1.25]), and during follow-ups of 1.5–5 months (OR 0.59 [0.37–0.93]) but not 6–12 months (OR 0.71 [0.44–1.18]) and 13–60 months (OR 0.90 [0.45–1.79]; Table 2 part B). This average treatment effect was significantly stronger as the percentage of females increased (β = 0.030 [0.004–0.055], p = 0.026) and psychiatric comorbidities increased (β = 0.015 [0.006–0.025], p = 0.007; Table 3 part B). However, mean age, percentage of prior suicide attempts, and patients with recurrent MDD were non-significant effect modifiers. There were insufficient studies to determine whether the treatment setting (outpatient v. inpatient) might be an effect modifier.
Forest plot for average treatment effect comparing combined treatment v. ADM-onlya. ADM, antidepressant medication; PSY, psychotherapy; OR, Peto odds ratio; CI, confidence interval; I 2, an index of between-study heterogeneity expressed in a percentage, such that higher values indicated more between-study heterogeneity; τ 2, tau-squared statistic, such that a statistically significant τ 2 indicated substantial between-study variance of the underlying distribution of true effect sizes. aEvents referred to number of patient(s) (if any) within a specific treatment arm that had a negative treatment outcome (i.e. suicide attempt, psychiatric emergency department [ED] visits, psychiatric hospitalization, and/or suicide death). The random-effects model only pertained to studies that examined SA as an outcome (Bockting et al., Reference Bockting, Klein, Elgersma, van Rijsbergen, Slofstra, Ormel and Burger2018; Brent et al., Reference Brent, Emslie, Clarke, Asarnow, Spirito, Ritz and Keller2009; Goodyer et al., Reference Goodyer, Dubicka, Wilkinson, Kelvin, Roberts, Byford and Harrington2007; Hollon et al., Reference Hollon, DeRubeis, Evans, Wiemer, Garvey, Grove and Tuason1992; Iftene et al., Reference Iftene, Predescu, Stefan and David2015; Kennard et al., Reference Kennard, Emslie, Mayes, Nakonezny, Jones, Foxwell and King2014; Khazanov et al., Reference Khazanov, Xu, Hollon, DeRubeis and Thase2021; Kocsis et al., Reference Kocsis, Thase, Trivedi, Shelton, Kornstein, Nemeroff and Keller2007; Lespérance et al., Reference Lespérance, Frasure-Smith, Koszycki, Laliberté, van Zyl, Baker and Guertin2007; March et al., Reference March, Silva, Petrycki, Curry, Wells, Fairbank and Severe2004; March et al., Reference March, Silva, Petrycki, Curry, Wells, Fairbank and Severe2007; Melvin et al., Reference Melvin, Tonge, King, Heyne, Gordon and Klimkeit2006; Michel et al., Reference Michel, Gysin-Maillart, Breit, Walther and Pavlidou2021; Schramm et al., Reference Schramm, van Calker, Dykierek, Lieb, Kech, Zobel and Berger2007; Vitiello et al., Reference Vitiello, Silva, Rohde, Kratochvil, Kennard, Reinecke and March2009; Wei et al., Reference Wei, Liu, Bi, Li, Hou, Tan and Liu2013; Wilkinson et al., Reference Wilkinson, Kelvin, Roberts, Dubicka and Goodyer2011). There were too few studies that examined psychiatric hospitalization (Browne et al., Reference Browne, Steiner, Roberts, Gafni, Byrne, Dunn and Kraemer2002; Hollon et al., Reference Hollon, DeRubeis, Evans, Wiemer, Garvey, Grove and Tuason1992; Kennard et al., Reference Kennard, Emslie, Mayes, Nakonezny, Jones, Foxwell and King2014; Mandoki et al., Reference Mandoki, Tapia, Tapia, Sumner and Parker1997; Nakagawa et al., Reference Nakagawa, Mitsuda, Sado, Abe, Fujisawa, Kikuchi and Ono2017; Schramm et al., Reference Schramm, van Calker, Dykierek, Lieb, Kech, Zobel and Berger2007) and suicide death (Bockting et al., Reference Bockting, Klein, Elgersma, van Rijsbergen, Slofstra, Ormel and Burger2018; Hollon et al., Reference Hollon, DeRubeis, Evans, Wiemer, Garvey, Grove and Tuason1992; Nakagawa et al., Reference Nakagawa, Mitsuda, Sado, Abe, Fujisawa, Kikuchi and Ono2017; Vitiello et al., Reference Vitiello, Silva, Rohde, Kratochvil, Kennard, Reinecke and March2009; Watanabe et al., Reference Watanabe, Furukawa, Shimodera, Morokuma, Katsuki, Fujita and Perlis2011) to conduct a random-effects model meta-analysis separately by those adverse outcomes. We are unaware of any studies that examined the effect of combined treatment v. ADM-only on psychiatric ED visits.
Psychotherapy-only v. ADM-only
Thirteen RCTs evaluated the comparative treatment effects of psychotherapy-only v. ADM-only. Thirty-one out of 1030 (3.0%) in the psychotherapy-only arm, and 67 out of 1192 (5.6%) in the ADM-only arm had suicide attempts and other serious psychiatric adverse events. This difference was statistically significant (OR 0.45, 95% CI [0.30–0.67], p = 0.001). There was no significant between-study heterogeneity (Q[df = 10] = 7.28, p = 0.699; I 2 = 0.0% [0.0–60.2]; τ 2 = 0 [0.00–1.50], p = 0.700) when comparing psychotherapy-only to ADM-only (Fig. 4). Subgroup analysis revealed that the average treatment effect of lower suicide attempts and other serious psychiatric adverse events in psychotherapy-only over ADM-only was statistically significant in youths (OR 0.39 [0.22–0.71]) but not adults (OR 0.54 [0.25–1.18]; Table 2 part C). However, study duration (Table 2 part C), mean age, percentage of females, prior suicide attempts, patients with recurrent MDD, and psychiatric comorbidities (Table 3 part C) were non-significant effect modifiers. All studies within this set of psychotherapy-only v. ADM-only comparisons occurred in outpatient settings.
Forest plot for average treatment effect comparing psychotherapy-only v. ADM-onlya. PSY, psychotherapy; ADM, antidepressant medication; OR, Peto odds ratio; CI, confidence interval; I 2, an index of between-study heterogeneity expressed in a percentage, such that higher values indicated more between-study heterogeneity; τ 2, tau-squared statistic, such that a statistically significant τ 2 indicated substantial between-study variance of the underlying distribution of true effect sizes. aEvents referred to number of patient(s) (if any) within a specific treatment arm that had a negative treatment outcome (i.e. suicide attempt, psychiatric emergency department [ED] visit, psychiatric hospitalization, and/or suicide death). The random-effects model only pertained to studies that examined SA as an outcome (Bockting et al., Reference Bockting, Klein, Elgersma, van Rijsbergen, Slofstra, Ormel and Burger2018; Dunlop et al., Reference Dunlop, Polychroniou, Rakofsky, Nemeroff, Craighead and Mayberg2019; Hollon et al., Reference Hollon, DeRubeis, Evans, Wiemer, Garvey, Grove and Tuason1992; Iftene et al., Reference Iftene, Predescu, Stefan and David2015; Lespérance et al., Reference Lespérance, Frasure-Smith, Koszycki, Laliberté, van Zyl, Baker and Guertin2007; March et al., Reference March, Silva, Petrycki, Curry, Wells, Fairbank and Severe2004; March et al., Reference March, Silva, Petrycki, Curry, Wells, Fairbank and Severe2007; Melvin et al., Reference Melvin, Tonge, King, Heyne, Gordon and Klimkeit2006; O'Hara et al., Reference O'Hara, Pearlstein, Stuart, Long, Mills and Zlotnick2019; Rucci et al., Reference Rucci, Frank, Scocco, Calugi, Miniati, Fagiolini and Cassano2011; Vitiello et al., Reference Vitiello, Silva, Rohde, Kratochvil, Kennard, Reinecke and March2009). Too few studies examined psychiatric hospitalization (Browne et al., Reference Browne, Steiner, Roberts, Gafni, Byrne, Dunn and Kraemer2002; Hollon et al., Reference Hollon, DeRubeis, Evans, Wiemer, Garvey, Grove and Tuason1992; O'Hara et al., Reference O'Hara, Pearlstein, Stuart, Long, Mills and Zlotnick2019), and suicide death (Hollon et al., Reference Hollon, DeRubeis, Evans, Wiemer, Garvey, Grove and Tuason1992; Vitiello et al., Reference Vitiello, Silva, Rohde, Kratochvil, Kennard, Reinecke and March2009) to conduct a random-effects model meta-analysis separately by those adverse outcomes. We are unaware of any studies that examined the effect of psychotherapy-only v. ADM-only on psychiatric ED visits.
Publication bias and study quality
Harbord's modified test of small study effects suggested that there was no significant publication bias for the analyses that examined combined treatment v. psychotherapy-only (t = −0.09, degrees of freedom [df] = 10, p = 0.933), combined treatment v. ADM-only (t = −1.76, df = 17, p = 0.096), and psychotherapy-only v. ADM-only (t = 0.14, df = 9, p = 0.88; online Supplementary Fig. S1). Trim-and-fill analyses showed that the comparative treatment effects on decreasing the risk for suicide attempts and other serious psychiatric adverse events were significantly better with psychotherapy-only over combined treatment (OR 1.96 [1.20–3.20], p = 0.012; Q[df = 11] = 9.29, p = 0.595), and psychotherapy-only over ADM-only (OR 0.45 [0.30–0.67], p = 0.001; Q[df = 10] = 7.28, p = 0.699), but not combined treatment v. ADM-only (OR 0.76 [0.58–1.00], p = 0.052; Q[df = 21] = 19.03, p = 0.583). In addition, the risk-of-bias plots indicated that most study publication biases were due to missing outcome data (Fig. 5).
Risk-of-bias plots of the comparative effects of ADM-only, psychotherapy-only, and combined treatment. ADM, antidepressant medication; l, low risk of bias; s, some concerns of bias; h, high risk of bias.
Discussion
The present novel meta-analysis consistently observed that psychotherapy monotherapy was superior to combined treatment (1.9% v. 3.7%) and ADM-only (3.0% v. 5.6%) in reducing the probability of suicide attempts and other serious psychiatric adverse events (i.e. psychiatric ED visits, psychiatric hospitalization, and/or suicide death) for MDD patients. The primary meta-analyses indicated that combined treatment was substantially better than ADM-only in decreasing the probability of suicide attempts and other serious psychiatric adverse events (6.0% v. 8.7%). However, subsequent trim-and-fill sensitivity analyses suggested that this difference was statistically non-significant. Potential accounts for these outcomes were considered in hypothesis-generating instead of confirmatory ways to catalyze more research in this area.
Why was psychotherapy monotherapy better than combined treatment and ADM alone in the entire sample? Notably, these findings concurred with patterns that emerged when MDD treatment remission or response was the dependent variable of interest in other published meta-analyses of RCTs (Cuijpers et al., Reference Cuijpers, Miguel, Harrer, Plessen, Ciharova, Ebert and Karyotaki2023; Furukawa et al., Reference Furukawa, Shinohara, Sahker, Karyotaki, Miguel, Ciharova and Cuijpers2021; Karyotaki et al., Reference Karyotaki, Smit, Holdt Henningsen, Huibers, Robays, de Beurs and Cuijpers2016). These results also aligned with evidence of a medium aggregate effect size on depression symptom reduction in support of psychotherapy combined with treatment-as-usual (mostly pharmacotherapy) compared to psychotherapy alone (van Bronswijk, Moopen, Beijers, Ruhe, & Peeters, Reference van Bronswijk, Moopen, Beijers, Ruhe and Peeters2019). Plausibly, psychotherapy (alone or with ADM), but not ADM-only, equipped depressed patients of various suicide risk levels with essential skills to self-monitor, detect early incipient cues (e.g. feelings of loneliness, suicidal thoughts) of suicide attempts, and intervene effectively by recalling and focusing on reasons for living (e.g. community, family, hobbies, religion; Bakhiyi, Calati, Guillaume, & Courtet, Reference Bakhiyi, Calati, Guillaume and Courtet2016) and harnessing other mood-uplifting strategies (e.g. eliciting social support, engaging in valued activities; Bhar & Brown, Reference Bhar and Brown2012).
Given these findings, future studies should conduct sophisticated instrumental variable complier average causal effects analyses (Hesser, Reference Hesser2020) to determine if differential engagement might mediate the comparative efficacy of psychotherapy-only v. combined treatment and ADM alone on suicide attempts and other serious psychiatric adverse events. Exploring other purported mechanisms of change of efficacy of psychotherapy-only v. combined treatment and ADM alone, such as reduced hopelessness (Celano et al., Reference Celano, Beale, Mastromauro, Stewart, Millstein, Auerbach and Huffman2017) and self-focused negative repetitive thinking (Chesin et al., Reference Chesin, Benjamin-Phillips, Keilp, Fertuck, Brodsky and Stanley2016), would also advance clinical psychological and related sciences. The growth of effective digitally delivered suicide-focused psychotherapies (Franz et al., Reference Franz, Mou, Kessler, Stubbing, Jaroszewski, Ray and Nock2022; Torous et al., Reference Torous, Larsen, Depp, Cosco, Barnett, Nock and Firth2018) provides ripe opportunities on these lines of investigation in the near future.
Another noteworthy finding was that the comparative effects of psychotherapy monotherapy over combined treatment and ADM-only in reducing the risk for serious psychiatric adverse events were substantial for children and adolescents instead of adults. These findings aligned with and extended warnings by drug regulators about the risk of ADM to heighten suicide risk, particularly among youths (Sharma, Guski, Freund, & Gotzsche, Reference Sharma, Guski, Freund and Gotzsche2016). Some investigations posit that the maturing brain may exhibit heightened susceptibility to unfavorable drug responses (Andersen & Navalta, Reference Andersen and Navalta2004, Reference Andersen and Navalta2011; Murphy et al., Reference Murphy, Capitao, Giles, Cowen, Stringaris and Harmer2021; Safer & Zito, Reference Safer and Zito2006), with the underlying mechanisms still shrouded in mystery. Alternatively, the linkage with suicidality may stem from the limited effectiveness of ADMs in clinical trials among suicidal children and adolescents, as implied by RCTs, rather than the ADM use directly amplifying suicidality and risks of other serious psychiatric adverse events (Cipriani, Barbui, & Geddes, Reference Cipriani, Barbui and Geddes2005; Dragioti et al., Reference Dragioti, Solmi, Favaro, Fusar-Poli, Dazzan, Thompson and Evangelou2019). Together, findings suggest that understanding the enduring advantages and potential hazards of ADM (alone or combined with psychotherapy) for depressive disorders among youths is paramount.
Conversely, the observation that combined treatment worked better than ADM alone for adults, rather than youths, in reducing the probability of suicide attempts and other serious psychiatric adverse events is consistent with clinical guidelines that recommend combined treatment for adult patients with high-risk severe v. mild-moderate depression (McQuaid et al., Reference McQuaid, Buelt, Capaldi, Fuller, Issa, Lang and Williams2022). These findings provide indications of ADMs (particularly when combined with psychotherapy) playing a protective role against suicidality in depressed adults (Barbui, Esposito, & Cipriani, Reference Barbui, Esposito and Cipriani2009; Craighead & Dunlop, Reference Craighead and Dunlop2014). However, our observed significant average treatment effect between combined treatment and ADM-only did not remain in the trim-and-fill sensitivity analysis, suggesting that publication bias driven by studies with small sample sizes might have influenced the results. Further, numerous studies indicating an elevated suicide risk associated with ADMs will likely stay unpublished. Overall, these results concurred with placebo-controlled RCTs (Braun et al., Reference Braun, Bschor, Franklin and Baethge2016; Fergusson et al., Reference Fergusson, Doucette, Glass, Shapiro, Healy, Hebert and Hutton2005) and observational studies (Hengartner et al., Reference Hengartner, Amendola, Kaminski, Kindler, Bschor and Ploderl2021) that consistently documented a notable elevation in suicide risk among depressed adult patients treated with ADM-only.
Additionally, the meta-regression analyses showed that higher percentage of females was associated with increased risk of suicide attempts and other serious psychiatric adverse events in response to ADM alone over combined treatment. Clinically, gender might be an essential consideration since prior meta-analyses showed that, on average, women responded better to psychotherapy (alone or combined with ADM) v. treatment-as-usual than men regarding reduction in depression symptoms (Braun, Gregor, & Tran, Reference Braun, Gregor and Tran2013) and suicide risk (Calati & Courtet, Reference Calati and Courtet2016). Relatedly, growing evidence underscores the heightened risk of repeated suicide attempts among female patients, contrasting with males who frequently choose more lethal means (Corcoran, Keeley, O'Sullivan, & Perry, Reference Corcoran, Keeley, O'Sullivan and Perry2004; Pandey et al., Reference Pandey, Bolstad, Lien, Walby, Myhre and Bramness2022; World Health Organization, 2021). The factors contributing to gender-based variations in treatment effectiveness remain presently uncertain. Nevertheless, there is a paucity of comprehensive investigation into the impact of gender on treatment efficacy on serious psychiatric adverse events, highlighting the necessity for further research in this domain.
Psychiatric comorbidities were also an effect modifier, with higher comorbidities associated with heightened risk of serious psychiatric adverse events in response to ADM-only over combined treatment. These outcomes, coupled with recent meta-analytic evidence on the protective effects of psychotherapies over pharmacotherapies in high-risk subgroups (de la Torre-Luque et al., Reference de la Torre-Luque, Pemau, Ayad-Ahmed, Borges, Fernandez-Sevillano and Garrido-Torres2023; Inagaki et al., Reference Inagaki, Kawashima, Kawanishi, Yonemoto, Sugimoto, Furuno and Yamada2015), suggest that psychotherapy should be prioritized for such patients. Collectively, the number of psychiatric comorbidities might be a salient risk marker concerning the odds of experiencing serious adverse events during and after treatment for depression.
Simultaneously, treatment timing might matter, as we found that combined treatment was more potent than ADM alone in decreasing the likelihood of serious psychiatric adverse events, particularly in studies with follow-up durations of 1.5–5 months rather than 6–60-month follow-ups. This finding extended a prior meta-analysis (Calati & Courtet, Reference Calati and Courtet2016), which showed that treatment duration was not an effect modifier of the effect of combined treatment v. treatment-as-usual. The present meta-analysis did not find evidence for a higher risk of subsequent suicide attempts the more extended the follow-up duration. A growing body of evidence indicates a rising prevalence of suicide attempts during the initial 6 months following a basal episode (Birtwistle, Kelley, House, & Owens, Reference Birtwistle, Kelley, House and Owens2017; Cully et al., Reference Cully, Corcoran, Leahy, Griffin, Dillon, Cassidy and Arensman2019; Liu, Lunde, Jia, & Qin, Reference Liu, Lunde, Jia and Qin2020) and that this timeframe might be the most optimal time to intervene to decrease the risk of suicide attempts and other serious psychiatric adverse events (Andreoli, Burnand, Frambati, Manning, & Frances, Reference Andreoli, Burnand, Frambati, Manning and Frances2021). Together, these findings might imply the enduring preservation of short-term combined treatment benefits in decreasing suicide risk and improving psychosocial functioning across extended periods.
The current meta-analysis had some limitations. First, the sample size of the current meta-analysis was modest, given the rare nature of the outcome of interest. Such efforts are essential to developing precision treatment models that could guide clinicians to optimize treatment selection for MDD patients (Cohen & DeRubeis, Reference Cohen and DeRubeis2018; Kessler & Luedtke, Reference Kessler and Luedtke2021). Second, more than half of the RCTs included herein did not report on their approaches to handling missing data in their population-level intention-to-treat analyses, i.e. a limitation that future similar RCTs should remedy. Third, Peto ORs could generate biased outcomes in the presence of notable differences in sample sizes of the intervention and comparator arms (Lane, Reference Lane2013). Except for the RCTs by Hollon et al. (Reference Hollon, DeRubeis, Evans, Wiemer, Garvey, Grove and Tuason1992) and Kocsis et al. (Reference Kocsis, Thase, Trivedi, Shelton, Kornstein, Nemeroff and Keller2007), the remaining 32 RCTs had well-balanced sample sizes between the intervention and comparator arms. Fourth, while most pharmacological studies included adverse event data, there was considerable variability in the extent of information provided and heterogeneity in the methods employed for adverse event assessment and analysis (Meister et al., Reference Meister, von Wolff, Mohr, Nestoriuc, Harter, Holzel and Kriston2016). In contrast, psychotherapy studies exhibited a notable dearth in the reporting of adverse event information. In order to enhance the comprehensive reporting of adverse events (including serious ones like suicide attempts examined herein) in published research on psychiatric disorders, researchers must prioritize the meticulous examination of potential harms. Fifth, the sample size was too small to determine if ADM type, such as the prescription of paroxetine in adults (Aursnes, Tvete, Gaasemyr, & Natvig, Reference Aursnes, Tvete, Gaasemyr and Natvig2006) and venlafaxine in youths (Cipriani et al., Reference Cipriani, Zhou, Del Giovane, Hetrick, Qin, Whittington and Xie2016), might be an effect modifier. Relatedly, there was insufficient data to determine how treatment setting (outpatient v. inpatient) might be an effect modifier. Future well-powered studies should examine this since the period following discharge from psychiatric hospitalization entails a heightened suicide risk (Kessler et al., Reference Kessler, Bauer, Bishop, Bossarte, Castro, Demler and Weinstock2023). Despite these limitations, fundamental strengths of the current meta-analysis include its novelty and ability to answer an essential question of how three first-line treatment options for MDD (i.e. psychotherapy and/or ADM) could reduce the risk of suicide attempts and other serious psychiatric adverse events.
To conclude, the present meta-analysis consistently found that psychotherapy monotherapy had stronger aggregate effects than combined treatment (1.9% v. 3.7%) and ADM-only (3.0% v. 5.6%) in decreasing the probability of suicide attempt, psychiatric ED visit, psychiatric hospitalization, and/or suicide death for MDD patients. The observation that combined treatment had better aggregate effect compared to ADM alone to reduce the likelihood of suicide attempts and other psychiatric events (6.0% v. 8.7%) did not consistently surpass the statistical significance threshold. On average, psychotherapies, especially those that integrate cognitive-behavioral and related theories, appear to be the best first-line intervention option to mitigate the risk of suicide attempts and other serious psychiatric adverse events in depressed youth and adult populations. Therapists should be cognizant of the absence of conclusive evidence, based on RCTs, regarding the preventive efficacy of ADM monotherapy in mitigating the risk of suicide attempts and other serious psychiatric adverse events, in contrast to prominent assertations (Dragioti et al., Reference Dragioti, Solmi, Favaro, Fusar-Poli, Dazzan, Thompson and Evangelou2019). Further, our data underscore the importance of discouraging the routine prescription of ADMs (alone or combined with psychotherapy) for pediatric and adolescent populations, except perhaps for sertraline. Meta-analytic data of placebo-controlled RCTs suggested no increased suicide risk and superior comparative efficacy regarding remission rates with sertraline prescribed to depressed youths (Cipriani et al., Reference Cipriani, Zhou, Del Giovane, Hetrick, Qin, Whittington and Xie2016). The benefits of utilizing ADMs in adults and older populations, especially in conjunction with psychotherapy, may outweigh manageable safety concerns, considering their effectiveness in addressing a range of psychiatric disorders, including MDD (Calati & Courtet, Reference Calati and Courtet2016; Craighead & Dunlop, Reference Craighead and Dunlop2014). Future similar comparative effectiveness research investigating these rare events as the dependent variable of interest could also build on the current meta-analysis by using large, less difficult-to-obtain prospective observational datasets to evaluate aggregate effects (i.e. trial emulation studies; Hernan & Robins, Reference Hernan and Robins2016) as a precursor to developing individualized treatment rules for optimizing treatment allocation using RCT designs (Kessler & Luedtke, Reference Kessler and Luedtke2021).
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0033291723003306
Competing interests
None.
Open access
