Description of Included Trials

In all, 20 primary RCTs (reported in 25 articles) compared SGAs with a psychological treatment and provided data for KQ 1a.^{85–90,95–103,108,116,118,119,121,129,131–134} Trials are grouped according to the type of psychotherapy compared with the SGA. They are listed within this chapter’s tables first by subtype of psychotherapy (if applicable) and then alphabetically by SGA. We found no trials eligible for KQ 1a that compared an SGA with behavior therapy or behavior modification or with humanistic therapies.

Five trials^{86,88,100,103,121} were conducted in primary care settings; the remainder took place in mental health care locations. Most trials were funded by the government; seven trials^{85,88–90,100,108,116} received at least partial funding from the pharmaceutical industry. Six trials^{88,95–97,99,100,129} took place solely in the United States; other countries included Brazil,¹⁰¹ Canada,^90,98,108 England,¹²¹ Finland,⁸⁶ Germany,¹⁰² Iran,^118,119 Italy,¹⁰³ Romania,⁸⁷ and The Netherlands.^89,116 One trial was conducted in both the United States and Italy.⁸⁵

Generally, patients were between 18 and 65 years of age; most trials reported a mean age between 35 and 45 years. In all trials, the majority of patients were female. One trial enrolled only women.¹⁰⁰ In the few trials that reported race or ethnicity, three^88,96,100 included more than 33 percent nonwhite patients. All trials reported mean baseline depressive severity of at least a moderate degree; most trials reported mean baseline HAM-D-17 scores between 16 (moderate depression) and 23 (severe). The total daily dose of each SGA medication was within the usual ranges prescribed for adults.

Second-Generation Antidepressants Compared With Behavior Therapies/Behavior Modification Therapies

We found no eligible trials that compared an SGA with behavior therapy/behavior modification.

Second-Generation Antidepressants Compared With Cognitive Behavioral Therapy

Table 11 describes the 11 included trials (13 publications) of an SGA compared with a CBT (grouped by therapy subtype and in alphabetical order by first author). Six trials employed CBT,^{90,95,98,100,102,108} four used cognitive therapy (CT),^87,97,99,119 and one each used problem solving therapy (PST)¹²¹ and rational emotive behavior therapy (REBT).^87,133 Trial counts exceed 11 because one trial had both cognitive therapy (CT) and REBT arms.⁸⁷ All but one trial compared SGA monotherapy with CBT alone; Lam and colleagues compared SGA monotherapy with SGA plus CBT.¹⁰⁸ Two trials included an additional comparison of SGA monotherapy with a combination of SGA and CBT.^119,121 Treatment duration ranged from 8 weeks to 1 year; some trials also reported followup results once patients were off- treatment.

Table 11

Second-generation antidepressants versus cognitive behavioral therapy: Trial characteristics, main outcomes, and risk of bias ratings.

One trial was rated overall low risk of bias,¹⁰⁸ five were rated medium risk,^{87,97,99,100,121,129} and five trials were rated high.^{90,95,98,102,119} Reasons for high risk of bias ratings included high attrition without proper handling of missing data, high differential attrition between treatment arms, potentially meaningful differences in baseline characteristics between treatment groups, potential reporting bias, and little or no information on randomization and allocation procedures. In two cases, we applied a second risk of bias rating for specific outcomes: one medium-risk trial⁹⁹ was rated high for change in HAM-D score, and one overall high-risk trial¹⁰² was rated medium for remission and response because we could use data from the full sample for those outcomes.¹⁰² Full risk of bias assessments for included trials are found in Appendix D.

Second-Generation Antidepressant Versus Cognitive Behavioral Therapy: Monotherapy Comparisons

We conducted random-effects meta-analyses of trials rated low or medium risk of bias for three outcomes: (1) remission (three trials [four comparisons];^{87,97,121,129} 432 patients), (2) response (five trials [six comparisons];^{87,97,99,102,121,129} 660 patients), and (3) change in HAM-D-17 score (three trials [four comparisons];^87,100,121 427 patients). We also performed sensitivity analyses for those outcomes including additional trials rated high risk of bias.^90,95,98

For remission, we included results measured between 12 and 16 weeks; all trials compared an SGA with CBT, and all trials defined remission based on a HAM-D-17 score of either less than 7⁸⁷ or less than or equal to 7.^97,99,121 One trial⁹⁷ also required a score less than or equal to 10 on the Beck Depression Inventory for remission. Patients treated with SGAs had numerically lower but not significantly different remission rates than patients on CBT (40.7 percent versus 47.9 percent; relative risk [RR], 0.98; 95% CI, 0.73 to 1.32; Figure 7). We found similar results when we stratified by subtype of CBT (CT versus PST versus REBT). Our sensitivity analysis included one additional SSRI trial,⁹⁵ a trial of an SNRI (venlafaxine),⁹⁰ and a trial that allowed patients to receive either an SSRI or an SNRI.⁹⁸ Our sensitivity analysis yielded a similar, nonsignificant difference.

Figure 7

SGA versus cognitive behavioral therapy: Remission. CBT = cognitive behavioral therapy; CI = confidence interval; SGA = second-generation antidepressant

For response, we included results measured between 8 and 16 weeks. Trials defined response as a 50 percent or greater reduction in HAM-D-17 score from baseline. Treatment effects were similar for SGAs and CBT (44.2 percent versus 45.5 percent; RR, 0.91; 95% CI, 0.77 to 1.07; Figure 8). We found similar results when we stratified by subtype of CBT and by time point (<12 weeks versus 12 to 16 weeks). The sensitivity analysis including three high risk of bias studies^90,95,98 yielded a similarly nonstatistically significant difference in response between SGAs and CBT.

Figure 8

SGA versus cognitive behavioral therapy: Response. CBT = cognitive behavioral therapy; CI = confidence interval; SGA = second-generation antidepressant

Our weighted mean difference analysis of the three trials that reported change in HAM-D-17 scores at 8 weeks or longer found no statistically significant difference between SGAs and CBT (WMD, −0.38; 95% CI, −2.87 to 2.10; Figure 9), although heterogeneity was somewhat high (I² = 44.3 percent). Potential sources of heterogeneity include variation between CBT subtypes (included trials used CT,⁸⁷ PST,¹²¹ and REBT⁸⁷) and type of provider delivering the psychotherapy (general practitioner¹²¹ versus psychologists or psychiatrists^87,100). We performed a sensitivity analysis that included the study that reported HAM-D-17 results at 4 weeks¹⁰⁰; doing so changed the direction of effect but not to a clinically or statistically significant degree (WMD, 0.57; 95% CI, −1.86 to 3.00). In addition, the heterogeneity increased to 62.9 percent. Further sensitivity analyses were not possible owing to too few trials. Adding the high risk of bias trials to the model yielded no difference in comparative effectiveness.

Figure 9

SGA versus cognitive behavioral therapy: Change in HAM-D-17. CBT = cognitive behavioral therapy; CI = confidence interval; HAM-D = Hamilton Depressive Scale; SGA = second-generation antidepressant

Two trials, both rated medium risk of bias, reported response, remission, or change in HAM-D-17 score at time points beyond 16 weeks. In one,⁸⁷ patients receiving either REBT or CT reported higher rates of remission and response at 6 months than patients taking fluoxetine, although neither difference was statistically significant. At 6 months, patients receiving REBT or CT reported significantly lower HAM-D-17 scores than the patients taking fluoxetine. In the trial that compared either fluvoxamine or paroxetine with PST,¹²¹ rate of remission at 1 year was higher in the PST arms, although rate of response at 1 year was higher in the SGA arm. In that trial, patients’ HAM-D-17 scores continued to decline, with 1-year scores being lower in the PST arms than the SGA arm. Again, these differences failed to reach statistical significance.

With respect to other health outcomes, three trials reported relapse rates during off-treatment followup.^87,97,98 Two trials defined relapse as symptom levels meeting criteria for MDD; the third⁹⁷ defined relapse as either a HAM-D-17 score of 14 or greater or a psychiatric status rating of 5 or greater during the first year of followup. During the followup period of that trial, patients who had initially received CT did not receive any treatment, and patients who had received SGA were randomized to continue SGA or be withdrawn to pill placebo.

In one medium risk of bias trial,⁸⁷ 10.6 percent of patients treated with fluoxetine relapsed within 6 months, compared with 2.1 percent and 6.1 percent of patients treated with REBT and CT, respectively. In the other medium risk of bias trial,⁹⁷ the rates of relapse were 39 percent for prior CT, 53 percent for patients who were on SGA and continued to receive it during followup, and 59 percent for patients who received SGAs during acute phase but were withdrawn to placebo during followup. Prior CT was significantly different from followup placebo (p=0.02). In the trial rated high risk of bias,⁹⁸ 47 percent of remitted patients treated with an SGA and 39 percent of remitted patients treated with CBT relapsed within 18 months.

Finally, one of the medium risk of bias trials⁹⁷ reported recurrence during the second year of followup, defined as either a HAM-D-17 score of 14 or greater or a psychiatric status rating of 5 or greater among those who did not relapse during year 1 of followup. The rates of recurrence during year 2 were 24 percent for prior CT and 52 percent for patients who were on SGAs during the acute phase. Owing largely to small numbers of patients in each group (17 in each group), the difference was not statistically significant, and results should be interpreted with caution. The single trial that reported measures of functional capacity used the Social Adjustment Scale;¹²¹ SGA and PST did not differ at end of treatment or at 40-week off-treatment followup.

Second-Generation Antidepressants Compared With Humanistic Therapies

We found no eligible trials that compared an SGA with humanistic therapies.

Second-Generation Antidepressants Compared With Integrative Therapies

The only type of integrative therapy used in the included studies was interpersonal psychotherapy (IPT). Table 12 describes the four included trials (five publications) of an SGA compared with IPT.^{85,88,89,103,131} One trial also included a combination SGA+IPT arm.⁸⁹ Two trials took place outside the United States;^89,103 two were conducted in outpatient primary care clinics.^88,103 Three of the four trials received a combination of industry and government funding.^85,88,89,131

Table 12

Second-generation antidepressants versus interpersonal psychotherapy: Trial characteristics, main outcomes, and risk of bias ratings.

Patients ranged between 18 and 66 years of age, and the samples comprised at least 72 percent females. Trial enrollment ranged from 60 to 318 patients. Treatment duration ranged from 8 to 24 weeks. None of the trials reported posttreatment followup results. The two trials rated high risk of bias provided few details about trial methods.^85,88 Full risk of bias assessments for included trials are found in Appendix D.

Second-Generation Antidepressants Compared With Psychodynamic Therapies

Table 13 describes the four included trials (five articles) of an SGA compared with PSYD of various sorts.^{86,96,101,116,132} Of these four trials, one included an additional treatment arm that combined fluoxetine and PSYD.¹⁰¹ One trial took place in the United States;⁹⁶ three were conducted in outpatient psychiatry clinics,^96,101,116 and one was conducted in a primary care setting.^86,132 Three trials were funded in part by a government agency.^86,96,101

Table 13

Second-generation antidepressants versus psychodynamic therapies: Trial characteristics, main outcomes, and risk of bias ratings.

Subjects ranged in age between 18 and 66 years of age; the samples comprised at least 72 percent females. Trial enrollment ranged from 51 to 272 patients. Three trials compared SGA monotherapy with short-term (2 to 4 months) PSYD; the fourth compared SGA monotherapy with long-term (24 months) PSYD. All four trials were rated medium risk of bias.

Second-Generation Antidepressants Compared With Third-Wave Cognitive Behavioral Therapy

Two trials compared an SGA (sertraline and paroxetine) with a third-wave CBT (namely, 16 sessions of behavioral activation).^97,118 One took place in an outpatient psychiatry clinic in Iran over 49 weeks and received funding from two academic institutions; the other was conducted in the United States and funded by the government. The American study also contained a cognitive therapy arm as reported earlier in this section. The American study was rated medium risk of bias, and the Iranian study as high risk. The samples ranged from 100 to 143 patients (see Table 14), and 66 to 85 percent of participants were female. Dimidjian et al. allowed a full range of paroxetine (10 mg to 50 mg/day), but Moradveisi et al. capped the dosage of sertraline at 100 mg/day—half the maximum dosage typically allowed for MDD.

Table 14

Second-generation antidepressants versus third-wave cognitive behavioral therapy: Trial characteristics, main outcomes, and risk of bias ratings of trials.

Both studies defined remission as HAM-D-17 ≤ 7 and BDI ≤ 10. The Iranian study reported much higher rates of both response and remission compared with the American study (see Table 14). In fact, we find it suspicious that over 90 percent of patients in both treatment groups in Moradveisi et al. reported response (between-group p=0.42). However, if one assumes that trial dropouts failed to respond, then rates of response are 66.0 percent for SGA and 88.0 percent for behavioral activation (BA) CBT. In the American study, paroxetine and BA CBT were associated with roughly similar rates of response at 16 weeks.

In the American study, authors found a greater rate of remission for BA CBT patients—nearly half—compared with patients taking paroxetine (27%). The Iranian study authors also reported that fewer patients taking sertraline were in remission at 13 weeks, compared with patients receiving BA CBT (68.6 percent versus 91.1 percent; p<0.01). With the same assumption of trial dropouts as treatment failures, the rates of remission in the Iranian study were 48.0 percent and 82.0 percent, respectively. However, these results should be interpreted with caution because of the potentially insufficient dosage of sertraline allowed.

Both studies reported followup data beyond the acute treatment phase. In one,¹¹⁸ at the 49-week followup, roughly half as many SGA patients as BA CBT patients reported at least a 50 percent reduction in symptoms (46.5 percent versus 88.6 percent; p<0.01). Similarly, fewer than half the number of SGA patients than BA CBT patients were in remission at 49 weeks (27.9 percent versus 65.9 percent; p<0.01). If one assumes that trial dropouts failed to remit, then rates of remission are 24.0 percent for SGA and 58.0 percent for BA CBT. With the same assumption for response, the rates are 40.0 percent and 78.0 percent, respectively. Among patients who were in remission at 13 weeks, more SGA patients relapsed (defined as “no longer meeting the remission criterion [scores less than or equal to 7 on the HAM-D and less than or equal to 10 on the BDI]”) during 49 weeks of followup than BA CBT patients (60.0 percent versus 27.8 percent; p=0.02). Again, these results should be interpreted with caution in light of the upper limit of the sertraline dosage.

In the other,⁹⁷ patients who had initially received BA did not receive any treatment, and patients who had received SGA were randomized to continue SGA or be withdrawn to pill placebo. In that study, relapse was defined as either HAM-D-17 score of 14 or greater or psychiatric status rating of 5 or greater during the first year of followup.

During the first year of followup of that trial, the rates of relapse were 50 percent for prior BA, 53 percent for patients who were on SGA and continued to receive it during followup, and 59 percent for patients who received SGA during acute phase but were withdrawn to placebo during followup.

Finally, one of the trials⁹⁷ reported recurrence during the second year of followup, defined as either a HAM-D-17 score of 14 or greater or a psychiatric status rating of 5 or greater among those who did not relapse during year 1 of followup. The rates of recurrence during year 2 were 26 percent for prior BA and 52 percent for patients who were on SGA during the acute phase. Owing largely to small numbers of patients (12 in prior BA and 17 in prior SGA), the difference was not statistically significant, and results should be interpreted with caution.

Description of Included Trials

We evaluated four CAM therapies: acupuncture, omega-3 fatty acids, SAMe, and St. John’s wort. All involved a comparison of an SGA with the CAM therapy of interest as monotherapy. When data were available, we also included an evaluation of an SGA with a combination of a CAM therapy plus an SGA. For all reports, the SGA was an SSRI; however, the term SGA has been used throughout for consistency. We defined acupuncture broadly to include techniques provided by trained practitioners that provide stimulation to meridian points using traditional needles. We elected to group trials of manual and electroacupuncture together because of the paucity of publications in this area and the uncertainty surrounding any meaningful differences between the two techniques for treating patients with depression.

We identified 20 primary RCTs (22 articles) comparing an SGA with a CAM therapy for treating patients with MDD.^{91,92,104–106,109–114,117,120,122–128,135,136} Five trials (six articles) evaluated acupuncture (503 participants), two trials evaluated omega-3 fatty acids (102 participants), one trial evaluated SAMe (189 participants), and 12 trials (13 articles) evaluated St. John’s wort (1,806 participants). About one-half of the trials (11 of 20) compared fluoxetine with a CAM therapy. Other SGAs involved sertraline (3 trials), paroxetine (2), citalopram (2), and escitalopram (1). Importantly, many of these trials used moderate or low antidepressant doses. Trials enrolled participants according to a criteria-based diagnosis of MDD based on the DSM-IV or the DSM revised third edition (DSM-III-R) and a predefined cutoff point of the HAM-D. Most participants had moderate to severe depression as measured by the HAM-D. All trials excluded patients who had additional Axis I disorders, high suicidal risk, or progressive medical diseases or who used psychotherapy, electroconvulsive therapy, or psychotropic medications.

Second-Generation Antidepressants Compared With Acupuncture

Table 15 describes the five trials (six articles, two reporting on substantially the same participants) that compared patients treated with an SGA to those treated with acupuncture monotherapy or with acupuncture plus an SGA. All trials took place in China. Four sets of analyses were funded by Chinese government agencies;^{91,105,122,135} the other two did not report their funding sources.^123,124 Trial enrollment ranged from 75 to 160 participants. All trials performed primary outcome evaluations at 6 weeks.

Table 15

Second-generation antidepressants versus acupuncture: Study characteristics, main outcomes, and risk of bias ratings.

Four trials used fluoxetine; the Qu et al. and Chen et al. trials used paroxetine. Trials employed a variety of experimental designs–including a variety of types of acupuncture, points used, and frequency of treatment. Three trials used the HAM-D-24^91,122,124 and two used the HAM-D-17.^105,123 Chen et al.¹³⁵ reported on essentially the same dataset as the Qu et al. trial;¹⁰⁵ also, it described outcomes for only the SCL-90 (Symptom Checklist 90), so we excluded it from meta-analyses.

Second-Generation Antidepressants Compared With Omega-3 Fatty Acids

Second-Generation Antidepressants Versus Omega-3 Fatty Acids: Combination Comparisons

Two trials compared patients treated with either fluoxetine or citalopram with patients treated with combinations of omega-3 fatty acids plus an SGA; we rated both these trials as high risk of bias (Table 16). One trial took place in the United States (funded by the National Institutes of Health) and recruited participants from outpatient referrals and local advertisements.¹⁰⁶ The other trial was from Iran (described above).¹²⁰ Combined, the trials evaluated 90 participants receiving either SGA monotherapy or the combination intervention; patient ages ranged from 18 to 65 years, and about 70 percent were female; the interventions took place over an 8-week period. Omega-3 fatty acid supplements consisted of either 1,000 mg daily of pure EPA¹²⁰ or a combination of 1,800 mg EPA, 400 mg DHA, and 200 mg other omega-3 fatty acids daily.¹⁰⁶ Primary outcome evaluations were based on the HAM-D.

Table 16

Second-generation antidepressants versus omega-3 fatty acids: Study characteristics, main outcomes, and risk of bias ratings.

In the U.S. trial, at 8 weeks, changes in HAM-D favored the combination of omega-3 fatty acid supplement plus citalopram over citalopram monotherapy (data not reported, p<0.05).¹⁰⁶ The Iran trial reported superior 8-week treatment response rates for combination treatment with EPA plus fluoxetine (81 percent) over rates for either fluoxetine (50 percent) or EPA (56 percent) alone (p=0.005).¹²⁰ Similarly, the combination treatment produced greater reductions in HAM-D over 8 weeks than either monotherapy (data not reported, p=0.005). In summary, participants treated with a combination of omega-3 fatty acids plus SGA were more likely to benefit than participants treated with either SGA or omega-3 monotherapy.

Second-Generation Antidepressants Compared With S-Adenosyl-L-Methionine

One trial (high risk of bias) compared escitalopram (10 to 20 mg/day) to SAMe (1,600 to 3,200 mg/day).¹⁰⁴ The National Institutes of Health supplied funding. The trial recruited participants from outpatient referrals and local advertisements to academic hospitals in two U.S. locations. Patients ranged in age from 17 to 79 years. The sample was 50 percent female. The trial evaluated outcomes, based on the HAM-D, after 12 weeks of treatment.

Second-Generation Antidepressants Versus S-Adenosyl-L-Methionine: Monotherapy Comparisons

Treatment groups did not differ significantly in treatment response (34 percent versus 36 percent, p>0.05), remission (28 percent versus 28 percent, p>0.05), or reduction in HAM-D scores over time (6.3 versus 6.1, p-value not reported) (see Table 17). Results of our network meta-analyses also reported similar response rates for patients treated with SGAs and patients treated with SAMe (RR, 1.22; 95% CI, 0.66 to 2.26).

Table 17

Second-generation antidepressants versus SAMe: Study characteristics, main outcomes, and risk of bias ratings.

Second-Generation Antidepressants Compared With St. John’s Wort (Hypericum perforatum)

Overall, 12 trials (13 articles) compared an SGA with St. John’s wort (Table 18). Trials used a variety of commercially available standardized extracts (LI-160, WS5570, Ze117, STW3, Calmigen, Iperisan, Swiss herbal remedies), most often standardized to 0.12 to 0.28 percent hypericin; doses ranged from 300 mg to 1,800 mg of the standardized extract daily. Nine trials included 900 mg within their dosing range. Six trials used fluoxetine for comparison,^{109,112,114,117,127,128} four used sertraline,^{92,111,113,126} one used paroxetine,¹¹⁰ and one used citalopram¹²⁵ (see Table 18 for dosages). Of the trials included in meta-analyses, none used an SGA dose at the maximum recommended strength and most used doses at the low end of the dosage range. In all, these trials provided data on 1,806 participants, predominantly with severe depression. Three trials took place in outpatient psychiatry clinics,^109,110,112 six trials in outpatient primary care clinics,^{111,114,117,125,126,128} and three trials did not report the source of patients beyond outpatient communities.^92,113,127 Five trials were conducted in Germany;^{110,114,125,126,128} three in the United States;^92,109,113 and one each in Brazil,¹¹² Canada,¹¹¹ Denmark,¹²⁷ and Sweden.¹¹⁷ The maker of the supplement sponsored seven trials;^{109–114,117} the U.S. government sponsored one.⁹² Treatment duration ranged from 4 to 12 weeks. Most trials had a medium risk of bias, although we rated three trials as high^109,112,113 and two trials as low risk of bias.^117,125 In two cases, we gave a medium risk of bias rating to high-risk trials when evaluating response and remission.^109,113 We attempted to contact all study authors for additional study information and received additional data for two studies.^92,111

Table 18

Second-generation antidepressants versus St. John’s wort: Trial characteristics, main outcomes, and risk of bias ratings.

Second-Generation Antidepressants Versus St. John’s Wort: Monotherapy Comparisons

Overall, treatment effects with respect to treatment response, remission, and magnitude of change on the HAM-D scale were similar between patients treated with SGAs or St. John’s wort. We did not find any evidence with respect to other outcomes of interest such as quality of life or functional capacity.

We conducted random-effects meta-analyses of nine low or medium risk of bias trials that reported data on response (1,517 participants), typically defined as ≥50 percent decrease in HAM-D.^{92,110,111,113,114,117,125–127} Patients treated with SGAs and those receiving St. John’s wort had similar response rates (51.7 percent versus 54.4 percent; RR, 0.96; 95% CI, 0.83 to 1.10) after 4 to 12 weeks of treatment (Figure 10). Sensitivity analysis using SGA dose or treatment duration showed no statistical difference between SGA and St. John’s wort. Sensitivity analysis stratified by St. John’s wort preparation demonstrated a difference for Ze 117¹¹⁴ in favor of St. John’s wort when compared with other preparations (RR, 0.66; 95% CI, 0.51 to 0.87) but was used in only a single trial. When stratifying by study country of origin, we found no statistical difference in estimates between studies conducted in Germany and non-German countries (RR, 0.90; 95% CI, 0.76 to 1.06 RR, 1.07; 95% CI, 0.85 to 1.33, respectively).

Figure 10

SGA versus St. John’s wort: Response. CI = confidence interval; SGA = second-generation antidepressants; SJW = St. John’s wort

Likewise, random-effects meta-analyses of five low or medium risk of bias trials (768 participants) showed similar remission rates (typically defined as HAM-D ≤7) for participants on SGAs or St. John’s wort (30.2 percent versus 36.2 percent; RR, 0.82; 95% CI, 0.67 to 1.00) after 4 to 12 weeks of treatment (Figure 11).^{92,109–111,117} Sensitivity analysis including one high risk of bias trial¹¹² (40 participants) produced similar findings (29.4 percent versus 33.2 percent; RR, 0.88; 95% CI, 0.68 to 1.13).

Figure 11

SGA versus St. John’s wort: Remission. CI = confidence interval; SGA = second-generation antidepressants; SJW = St. John’s wort

Seven trials with low or moderate risk of bias reported data on change in HAM-D scores (1,369 participants).^{92,109–111,114,125–127} We found similar HAM-D reductions for patients treated with an SGA and those treated with St. John’s wort (Figure 12; mean difference −0.45; 95% CI, −1.45 to 0.55). Sensitivity analysis including two high risk of bias trials indicated no difference in conclusions (mean difference −0.65; 95% CI, −1.62 to 0.33).

Figure 12

SGA versus St. John’s wort: Change in HAM-D-17. CI = confidence interval; HAM-D = Hamilton Depression Scale for Depression; SGA = second-generation antidepressants; SJW = St. John’s wort.

Second-Generation Antidepressants Compared With Yoga

We found no eligible trials that compared an SGA with yoga.

Second-Generation Antidepressants Compared With Meditation

We found no eligible trials that compared an SGA with meditation therapy.

Second-Generation Antidepressants Versus Aerobic Exercise: Monotherapy Comparisons

Neither trial found a statistically significant difference in remission rates between sertraline alone and aerobic exercise alone: 68.8 percent (sertraline) versus 60.4 percent (exercise) in the 1999 trial and 47 percent (sertraline) versus 45 percent (supervised exercise) versus 40 percent (home-based exercise) in the 2007 trial. All three active groups in the 2007 trial tended to have higher remission rates than the placebo control group (31 percent) (p=0.057). The crude pooled risk ratio comparing sertraline treatment with the exercise conditions (pooling data from the two exercise groups in the 2007 trial with the one exercise-only group in the 1999 trial, a total of three arms) was 1.10 (95% CI, 0.87 to 1.39). All active treatment groups in the 2007 trial showed a clinically and statistically significant decline (p<0.0001) in HAM-D scores from baseline to 16 weeks; the sertraline group decreased by 6.1, supervised exercise by 7.2, home-based exercise by 7.1, and placebo by 6.1 points. There were no between-group differences in this decline (p=0.321). The 1999 trial found the magnitude of the decline in HAM-D to be comparable across groups; it did not provide specifics. Neither trial reported response rates. Based on network meta-analyses, patients in the SGA and exercise groups had similar response rates (RR 1.86; 95% CI, 0.81 to 4.27).

In both trials, patients receiving sertraline showed significantly lower levels of aerobic capacity (peak V0₂), as well as shorter treadmill times, than patients in the exercise groups (p<0.001). The Blumenthal et al. 1999 trial also assessed anxiety, self-esteem, life satisfaction, and dysfunctional attitudes. Although both the sertraline and the exercise groups improved, the groups did not differ on these measures. The companion report to the 2007 trial¹³⁸ found little evidence of between-group differences in neurocognitive measures; exercise participants performed better than those on sertraline on tests of executive function (Trail-making Test, p=0.02; Ruff 2 & 7 test, p=0.03) but not on measures of verbal memory or verbal fluency/working memory.

In a sensitivity analysis, the magnitudes of the RRs are slightly attenuated with inclusion of trials with a high risk of bias, but the interpretations do not change.

Second-Generation Antidepressants Versus Aerobic Exercise: Combination Comparisons

A single trial, Blumenthal et al. 1999,⁹³ included an arm comparing sertraline alone to a combination of sertraline plus exercise; it had 48 participants in the sertraline-alone group and 55 in the combined sertraline plus exercise group. Data were insufficient to estimate comparative benefits of SGA monotherapy versus combination therapy with SGA and exercise using network analysis. Patients in the sertraline-only group showed minimal (<3 percent) improvement in aerobic capacity; those in the combined group improved by 9 percent. The two groups did not differ in improvements in anxiety, self-esteem, life satisfaction, or dysfunctional attitudes scores.

Description of Trials

In all, four RCTs compared SGA with a psychological treatment and provided data for KQ 1b (Table 20).^{96,97,103,118} We rated three trials medium risk of bias^96,97,103 and one trial as high risk of bias.^97,118 One medium risk of bias trial compared SGA with either of two psychological treatments.⁹⁷ Included trials compared an SGA with a CBT (cognitive therapy),^97,118 a third-wave CBT (behavioral activation),⁹⁷ a PSYD,⁹⁶ and an integrative therapy (interpersonal therapy).¹⁰³ We found no trials eligible for KQ 1b that compared a SGA with behavior therapy/behavior modification or with a humanistic therapy.

Table 20

SGAs versus psychological interventions by depression severity: Trial characteristics, main outcomes, and risk of bias ratings of trials.

Two of the trials were conducted in the United States,^96,97 and two were conducted in other countries: one in Iran¹¹⁸ and one in Italy.¹⁰³ Two of the trials took place in outpatient primary care settings;^97,103 two were conducted in outpatient psychiatry clinics.^96,118 Three of the trials were funded entirely or in part by the government.^96,97,103 Three trials did not provide any information on treatment fidelity,^96,103,118 and only one trial reported adequate treatment fidelity.⁹⁷ None of the trials reported on functional capacity, quality of life, reduction of suicidality, relapse, or hospitalization. None of the trials excluded individuals with any comorbid anxiety disorder, although one trial reported that they did not include subjects with a primary diagnosis of panic disorder or obsessive-compulsive disorder.⁹⁷

None of the trials was designed to answer the primary question of whether depressive severity was a modifier of the comparative effectiveness of SGAs versus psychotherapy. However, two trials prespecified their plan to use depressive severity as a moderator.^97,103 The methods to analyze whether outcome measured by depressive severity varied. One trial stratified its sample into a high- and low-severity subgroup and assessed the comparative benefits of the SGAs versus psychotherapy within each subgroup.⁹⁷ Another trial examined potential moderators of remission with logistic modeling, including stratification of high versus low severity as one possible predictor.¹⁰³ A third trial used a mixed regression analysis model that tested whether the baseline depressive severity score moderated outcomes.¹¹⁸ Finally, one trial used hierarchical linear modeling to determine whether depressive severity had a moderating effect, considering both the full sample as well as the subgroup with higher depressive severity.⁹⁶

Generally, patient age ranged between 18 and 50 years old; trials reported a mean age between 31.4¹¹⁸ and 44.9 years.¹⁰³ In all trials, the majority of the patients were female. Two trials reported minority status (18.3 percent⁹⁷ and 48.1 percent⁹⁶).

Impact of Severity on Various Outcomes

One medium risk of bias trial (n=145), with one arm comparing paroxetine and CT, conducted subgroup analyses in patients with low- and high-severity MDD.⁹⁷ For the subgroup with high-severity MDD (i.e., those with a HAM-D-17 ≥20), those receiving paroxetine were less likely to achieve remission of MDD than those receiving CT (23 percent versus 36 percent, p=NS).⁹⁷ For the subgroup with low-severity MDD (i.e., those with a HAM-D-17 ≤19); remission rates did not differ significantly for patients treated with paroxetine or CT. Efficacy did not differ significantly between treatments in either subgroup when measured by treatment response or change in HAM-D-17. Because of the small sample size and the fact that authors conducted multiple parallel comparisons of subgroups and not a test of interaction, findings might be attributable to chance and need to viewed cautiously.

One medium risk of bias trial (n=287) reported subgroup analyses of patients with low- or high-severity MDD at baseline who were treated with either an SGA or IPT.¹⁰³ From regression analyses, Menchetti and colleagues¹⁰³ reported that the likelihood of remission varied as a function of depression severity; only those with less severe depression saw a worse outcome from SGA than from IPT. For patients with baseline HAM-D-21 <18, those receiving 2 months of citalopram or sertraline were 19 percent less likely to achieve remission than those receiving IPT (Standardised Rate Difference [SRD], 0.19; 95% CI, 0.04 to 0.34), consistent with a small-to-moderate effect size (ES = 0.25).¹⁰³ However, for patients with high-severity MDD (HAM-D-20 ≥18), the likelihood of remission did not differ between the two treatment groups [SRD, −0.06; 95% CI, −0.24 to 0.12].¹⁰³ The trial did not report treatment response or change in HAM-D score.

One medium risk of bias trial and one high risk of bias trial^97,118 compared an SGA with behavioral activation and provided subgroup analyses that considered the effects of depressive severity on treatment outcome. In one trial (n=100), Moradveisi and colleagues used regression modeling to assess the effect of baseline severity on change in depressive severity.¹¹⁸ The difference in treatment effects between the two types of interventions increased as a function of severity. In patients with less severe MDD at baseline, the difference in treatment effects at weeks 4, 13, and 49 were minimal. However, as baseline severity increased, patients receiving sertraline had less improvement in depressive severity as measured by both HAM-D and BDI at each followup point.¹¹⁸

The medium risk of bias trial (n=143) reported on the effect of baseline depressive severity on all three main outcomes.⁹⁷ In this trial, the authors reported that for subjects with high-severity MDD (defined as HAM-D-17 ≥20), those receiving paroxetine were less likely to remit than those receiving BA (23 percent versus 56 percent, p=0.002). In those with low-severity MDD, remission rates did not differ to a statistically significant degree between the two treatment groups. For the other two outcomes, treatment response or change in HAM-D-17 score, having either high- or low-severity MDD did not produce different outcomes for the two interventions.

One medium risk of bias trial (n=106) that compared supportive–expressive psychotherapy conducted subgroup analyses in high- and low-severity patients.⁹⁶ The trial did not report on either response to treatment or remission. Although the authors did not report specific changes in HAM-D scores stratified by subgroup, they did analyze depression severity as a potential moderator of change in HAM-D scores. Limiting the analysis to patients with high depression severity revealed no differences in rate of change of HAM-D. We contacted trial authors for additional data but did not receive any supplementary information.

Comparative Efficacy for Critical Efficacy Outcomes by Baseline Severity for Psychological Interventions and Second-Generation Antidepressants

We further investigated the role of depressive severity on outcomes by considering all trials from KQ 1a that both directly compared psychological interventions to SGAs and reported on key effectiveness outcomes (response, remission, and/or functional capacity). These studies did not directly assess depressive severity as a moderator; however, one might observe whether there is evidence of a relation between mean baseline depressive severity and the comparative effectiveness of the interventions (Table 21). We were not able to stratify by whether the depressive severity of the populations was specifically “moderate” or “severe,” because most populations were mixed (i.e., they had both moderate and severely depressed populations mixed together). Rather, for each comparison we list the range of mean baseline depressive severity and the findings. Of note, we found no differences in the comparative effectiveness between SGAs and psychological treatments in patients with moderate to severe MDD, which is consistent with findings of the few studies that we have for KQ 1b. However, as with our earlier KQ 1b findings, the evidence was very limited.

Table 21

Comparative efficacy for critical efficacy outcomes by baseline severity for psychological interventions and second-generation antidepressants.

Second-Generation Antidepressants Compared With S-Adenosyl-L-Methionine

One trial (Table 22), rated medium risk of bias, compared escitalopram (10 to 20mg/day) with SAMe (1,600 to 3,200 mg/day). The National Institutes of Health supplied funding. The trial recruited participants (N=129) from outpatient referrals and local advertisements to academic hospitals in two U.S. locations. Participant age ranged from 17 to79 years; the sample was 50 percent female. The trial evaluated outcomes, based on the HAM-D, after 12 weeks of treatment. Mean (SD) baseline HAM-D score was 19.2 (4.7) with a range from 4 to 32. No statistically significant interaction appeared between baseline HAM-D score and treatment groups for reduction in HAM-D scores over time (p=NS).

Table 22

SGAs versus SAMe by depression severity: Trial characteristics, main outcomes, and risk of bias ratings of trials.

Comparative Efficacy for Critical Efficacy Outcomes by Baseline Severity for Complementary and Alternative Interventions, Exercise, and Second-Generation Antidepressants

As with our psychological intervention comparison, we further investigated the role of depressive severity on outcomes by considering all trials from KQ 1a that directly compared CAM interventions or exercise to SGAs and reported on key effectiveness outcomes (i.e., response, remission, and/or functional capacity) (Table 23). Again, we found no differences in treatment effects in populations with moderate or severe MDD, which is consistent with findings of the few studies that we have for KQ 1b. This evidence, too, was extremely limited.

Table 23

Comparative efficacy for critical efficacy outcomes by baseline severity for complementary and alternative interventions, exercise, and second-generation antidepressants.

Description of Included Trials

In all, two trials provided four comparison studies reported in four articles. All four comparisons reported in three of the articles^107,140,141 involved data from the STAR*D study, which had multiple arms allowing several comparisons following a treatment failure. A different independent study reported data comparing various SGA switches.¹¹⁵

The Lenox-Smith and Jiang trial was conducted in a single outpatient psychiatry setting in Great Britain and was funded by the pharmaceutical industry. The STAR*D comparison involved outpatients from 41 psychiatric (60 percent) and primary care (40 percent) settings in the United States and was government funded.

Generally, patients were between 18 and 65 years of age (mean ages between 41 and 43 years). In both, the majority of patients were female. Mean baseline depressive severity was at least moderate. STAR*D comparisons involved mean baseline HAM-D scores between 15.8 and 17.8; the Lenox-Smith and Jiang trial had greater depression severity, with a mean HAM-D score of approximately 26 (severe). The total daily dose of each SGA medication reached or exceeded the minimum recommended dose for that medication as prescribed for adults, and the maximal dose did not exceed that noted in FDA labelling.

Whereas the Lenox-Smith and Jiang trial was a relatively standard RCT, the STAR*D study employed an equipoise randomization scheme that allowed some degree of patient preference. STAR*D was designed to allow multiple randomized comparisons of second-step therapies; the three relevant comparisons reported here^107,140,141 all involved patients who did not remit following 3 months of treatment with citalopram. Patients could not refuse a specific medication choice, but patients did have the option of refusing any of the available treatment strategies (switch to another SGA, switch to cognitive therapy, augment with a second medication, or augment with cognitive therapy), as long as at least two treatment options remained to allow randomization.

Second-Generation Antidepressant Switch Compared With Second-Generation Antidepressant Switch

Table 25 describes the trial characteristics, main outcomes, and risk of bias ratings for these analyses. The Lenox-Smith and Jiang trial lasted 12 weeks, with 396 patients randomized to one of two treatment arms.¹¹⁵ The trial compared venlafaxine ER (doses ranged from 75 to 300 mg daily; mean daily dose was 191 mg) to citalopram (20 to 60 mg; mean daily dose, 51 mg). The investigators measured response with the HAM-D; they did not report response rate for the two study arms but instead stated that response did not differ (reported as p=0.953). They did not report remission rate or time to remission for any outcome. The decrease in depressive severity, whether measured by HAM-D, MADRS (p=0.5002), or CGI-S (p=0.3014), did not differ by groups. We rated the risk of bias as low.

Table 25

Second-generation antidepressant switch versus another second-step switch strategy: Trial characteristics, main outcomes, and risk of bias ratings.

The Rush et al. study lasted an average of 14 weeks; it randomized 727 patients into one of three treatment arms.¹⁴⁰ The switch comparison randomized patients to either bupropion SR (150 to 400 mg; mean daily dose at end of study was 282 mg), sertraline (50 to 200 mg; mean daily dose at end of study was 136 mg), or venlafaxine XR (37.5 mg to 375 mg; mean daily dose at end of study was 194 mg). Response rates did not differ by treatment arm; as reported for the QIDS-SR, they ranged from 26.1 percent to 28.2 percent (p-value not reported). Similarly, remission rates did not differ between treatment arms, whether reported for either the HAM-D (p=0.16) or the QIDS-SR (p-value not reported). The mean change in HAM-D score was not reported; however, the percentage decrease in QIDS-SR was reported and did not differ among the three groups. Neither the time to response (ranging from 5.5 to 7.0 weeks) nor the time to remission (5.4 to 6.2 weeks) differed among the three options.

We rated this study as medium risk of bias, as we did for all the STAR*D studies described below, for two reasons: less than 80 percent of the sample provided outcomes at study completion and because the mean medication doses ultimately prescribed indicated that some medications did not reach the maximal dose recommended in the protocol so that comparable adequate doses may not have been achieved among the various arms. Appendix C documents the full risk of bias assessments for included trials.

Second-Generation Antidepressant Switch Compared With a Nonpharmacological Treatment Switch (Psychotherapy)

Thase et al. reported a STAR*D-based comparison (rated medium risk of bias) of switching to an SGA versus switching to a nonpharmacological strategy, namely CT¹⁰⁷ (Table 25). Randomization to a different SGA could assign patients to receive sertraline (50 mg to 200 mg; mean daily dose at end of study was 137 mg), bupropion SR (150 mg to 400 mg, mean daily dose at end of study was 270 mg), or venlafaxine XR (37.5 mg to 375 mg, mean daily dose at end of study was 221 mg); however, the comparisons of SGA with CT consolidated the medications into a single SGA group variable. Response rates assessed on the QIDS-SR showed no difference between SGA and CT (26.7 percent versus 22.2 percent p=0.84). Similarly, remission rates did not differ by treatment arm, whether measured by the HAM-D (27.9 percent versus 25.0 percent, p=0.69) or by the QIDS-SR (26.7 percent versus 30.6 percent, p=0.90).

HAM-D change in depressive severity was not reported, but the percentage decrease in QIDS–SR-16 did not differ between the groups (46.2 percent versus 40.7 percent, p=0.90). Neither the time to response nor remission differed for these two switch strategies.

Second-Generation Antidepressant Augmentation Compared With Second-Generation Antidepressant Augmentation

One eligible trial (another from the STAR*D series, also rated medium risk of bias) compared an SGA augmentation strategy with another SGA augmentation strategy (Table 25).¹⁴¹

This augmentation comparison randomized patients to the addition of either bupropion SR (150 mg to 400 mg, mean daily dose at end of study was, 268 mg) or buspirone, a nonbenzodiazepine anxiolytic (15 mg/60 mg; mean daily dose at end of study was 41 mg). Response rates did not differ by treatment arm; as reported for the QIDS-SR (31.8 percent versus 26.9 percent, p=0.21). Remission rates also did not differ (29.7 percent versus 30.1 percent, p=0.93, on HAM-D; 39.0 percent versus 32.9 percent, p=0.13, on QIDS-SR). The investigators did not report the mean change in HAM-D score; they did report the percentage decrease in QIDS-SR as favoring bupropion over buspirone (decrease of 25.3 percent versus 17.1 percent, p<0.04). Neither the time to response (ranging from 6.3 to 6.8 weeks) nor the time to remission (ranging from 5.4 to 6.3 weeks) differed between the two augmentation options.

Second-Generation Antidepressant Augmentation With Pharmacological Treatment Compared With Second-Generation Antidepressant Augmentation With Nonpharmacological Treatment Switch (Psychotherapy)

One eligible trial compared an SGA augmentation with a nonpharmacological SGA augmentation strategy, CT (Table 26).¹⁰⁷

Table 26

Second-generation antidepressant augmentation versus another second-generation augmentation strategy: Trial characteristics, main outcomes, and risk of bias ratings.

This augmentation comparison randomized patients to the addition of either a medication (bupropion SR, an antidepressant [150 to 400 mg, mean daily dose at end of study was 283 mg], or buspirone [15 to 60 mg, mean daily dose at end of study was 45.1 mg]) or CT (16 sessions). Response rates did not differ by treatment arm, as reported by the QIDS-SR (28.2 percent versus 35.4 percent, p=0.25). Remission rates also did not differ by HAM-D (33.3 percent versus 23.1 percent, p=0.20) or by QIDS-SR (33.3 percent versus 30.8 percent, p=0.78). Although the mean change in HAM-D score was not provided, the percentage decrease in QIDS-SR revealed no difference between the percentage decrease in depressive severity (39.6 percent versus 40.5 percent, p=0.83). Patients assigned to medication group did not differ from the CT group in terms of time to response; however, those receiving medication reached remission faster than those receiving CT (40.1 days versus 55.3 days, p=0.022).

Second-Step Switch Strategy Compared With Any Augmentation Strategy

We found no eligible trials that directly compared a SGA switch strategy with an augmentation strategy.

Network Meta-Analysis of Either Switch or Augmentation Comparisons Versus Placebo

We did not have enough eligible studies to conduct a network meta-analysis of the relevant treatment options compared with placebo.

Second-Generation Antidepressants Compared With Psychological Interventions

We first present the available evidence on the comparative risk of harms for SGAs and psychological treatments as a class. Next, we summarize the evidence for each included psychological intervention. As in KQ 1, we use classifications of the Cochrane Collaboration Depression, Anxiety and Neurosis Review Group.⁷⁰

Second-Generation Antidepressants Compared With Any Psychological Interventions

Second-Generation Antidepressants Versus Any Psychological Intervention: Monotherapy Comparisons

We conducted meta-analyses of low or medium risk of bias studies comparing overall discontinuation rates, discontinuation rates because of lack of efficacy, and discontinuation rates because of adverse events for patients treated with any SGA compared with those treated with any psychological intervention. Interventions for these comparisons were limited to fluoxetine, fluvoxamine, paroxetine, and sertraline (for the SGAs) and behavioral activation, cognitive therapy, problem solving therapy, rational emotive behavior therapy, and short-term psychodynamic psychotherapy (for the psychological interventions).

Overall discontinuation rates were numerically, but not statistically, higher following SGAs than psychological interventions, according to our random-effects meta-analysis (15.4 percent versus 11.4 percent; RR, 1.47; 95% CI, 0.94 to 2.30; Figure 16). When we used a fixed-effects meta-analytic model, however, the difference in overall discontinuation rates became statistically significant (RR, 1.41; 95% CI, 1.01 to 1.97).

Figure 16

SGA versus psychological interventions as a class: Overall discontinuation rates. CI = confidence interval; Discont’d = discontinued; SGA = second-generation antidepressant

Discontinuation rates because of lack of efficacy were numerically lower for patients treated with SGAs than for patients treated with psychological interventions, even though the difference did not reach statistical significance (2.1 percent versus 6.3 percent; RR, 0.52; 95% CI, 0.18 to 1.46; Figure 17).

Figure 17

SGA versus psychological interventions as a class: Discontinuation rates because of lack of efficacy. CI = confidence interval; SGA = second-generation antidepressant.

In contrast, discontinuation rates because of adverse events were more than twice as high for patients receiving SGAs than for those treated with psychological interventions, but the difference was not statistically significant (7.1 percent versus 2.1 percent; RR, 2.73; 95% CI, 0.89 to 8.38; Figure 18). The numbers of events of discontinuation because of lack of efficacy and discontinuation because of adverse events, however, were low; therefore, results should be interpreted with caution.

Figure 18

SGA versus psychological interventions as a class: Discontinuation rates because of adverse events. CI = confidence interval; Discont’d = discontinued; SGA = second-generation antidepressant

For sensitivity analyses, we added five high risk of bias trials to the meta-analytic models.^{89,90,95,102,116} The results of the analyses of overall discontinuation rates (RR, 1.16; 95% CI, 0.82 to 1.64) and discontinuation because of adverse events (RR, 2.55; 95% CI, 0.43 to 15.01) remained consistent with the results of their respective primary analyses presented above. No high risk of bias trials reported on discontinuation because of lack of efficacy. When a fixed-effects meta-analytic model was used, the difference between SGAs and psychological interventions in the sensitivity analysis became statistically significant for discontinuation because of adverse events^97,102 (RR, 2.66; 95% CI, 1.31 to 5.39).

Second-Generation Antidepressants Versus Any Psychological Treatment: Combination Comparisons

Four trials, including one with a high risk of bias,⁸⁹ comparing SGAs with combinations of SGAs and psychological interventions reported information on adverse events.^{89,101,108,121} SGAs included fluvoxamine, paroxetine, escitalopram, nefazodone, and fluoxetine, and psychological interventions included CBT, interpersonal therapy (IPT), and long-term psychodynamic therapy. After 12 weeks, overall discontinuation rates were similar for patients treated with SGAs (9.4 percent to 16.7 percent) and those treated with psychotherapy (15.4 percent to 17.1 percent).^108,121 In contrast, rates of discontinuation because of adverse events were numerically, but not statistically, higher among patients treated with escitalopram, fluvoxamine, or paroxetine (5.7 percent to 11.1 percent) than the same SGAs used in combination with CBT (0.0 percent to 3.8 percent).^108,121 The high risk of bias trial found similar results after 12 weeks of treatment whether patients received SGAs (36.2 percent) or IT (32.7 percent).⁸⁹

Second-Generation Antidepressants Compared With Cognitive Behavioral Therapy

Of 11 trials included for KQ 1a, 9 reported limited data on adverse events (see KQ 1, Table 11 for more details on trial design and dosing).^{87,90,95,97–99,102,108,121,129,133}

Second-Generation Antidepressants Versus Cognitive Behavioral Therapy: Monotherapy Comparisons

Eight trials, of which four had a high risk of bias rating,^90,95,98,102 provided limited information on the comparative risk of harms of SGA monotherapy compared with CBT.^{87,90,95,97–99,102,121,129,133} In these trials, SGAs were limited to escitalopram, fluoxetine, fluvoxamine, paroxetine, sertraline, and venlafaxine. None of the trials provided information on the comparative risk of specific adverse events, even common adverse events of SGAs. Only one trial reported in two publications provided data on the proportions of patients who experienced any adverse events.^87,133 About 15.7 percent of patients treated with an SGA experienced adverse events as did 0.9 percent of patients treated with CBT. Particularly for SGAs, reported adverse event rates appear to underestimate substantially the actual risk. A comprehensive systematic assessment of the risk of harms for SGAs reported that an average of 60 percent of patients treated with SGAs experience at least one adverse event during treatment.³³

Overall discontinuation rates were similar for patients treated with SGAs or CBT (16.0 percent versus 15.8 percent; RR, 1.00; 95% CI, 0.59 to 1.69, Figure 19). The findings did not change when stratified by time point (<12 weeks versus 12 to 16 weeks). Discontinuation rates because of lack of efficacy were numerically, but not statistically significantly, lower for patients treated with SGAs than for those treated with CBT (2.4 percent versus 11.4 percent; RR, 0.36; 95% CI, 0.06 to 2.21, Figure 20). In contrast, discontinuation rates because of adverse events were numerically higher for patients on SGAs than for patients treated with CBT, but the difference did not reach statistical significance (7.8 percent versus 2.7 percent; RR, 2.54; 95% CI, 0.39 to 16.47, Figure 21). The numbers of events for discontinuation because of lack of efficacy and discontinuation because of adverse events were very low. Therefore, results should be interpreted with caution.

Figure 19

SGA versus cognitive behavioral therapy: Overall discontinuation rates. CBT = cognitive behavioral therapy; CI = confidence interval; Discont’d = discontinued; SGA = second-generation antidepressant

Figure 20

SGA versus cognitive behavioral therapy: Discontinuation rates because of lack of efficacy. CBT = cognitive behavioral therapy; CI = confidence interval; SGA = second-generation antidepressant

Figure 21

SGA versus cognitive behavioral therapy: Discontinuation rates because of adverse events. CBT = cognitive behavioral therapy; CI = confidence interval; Discont’d = discontinued; SGA = second-generation antidepressant

In sensitivity analyses, we added high risk of bias trials to the meta-analytic models.^90,95,102 The differences in overall discontinuation rates (RR, 0.98; 95% CI, 0.72 to 1.35) and rates of discontinuation because of adverse events (RR, 2.97; 95% CI, 0.69 to 12.81) remained similar to the original analyses. As in the primary analysis, the findings of the sensitivity analysis for overall discontinuation did not change when stratified by time point (<12 weeks versus 12 to 16 weeks).^97,102

Second-Generation Antidepressants Versus Cognitive Behavioral Therapy: Combination Comparisons

The only trial that compared an SGA (escitalopram) with a combination of escitalopram and telephone CBT did not report information on specific adverse events.¹⁰⁸ After 12 weeks, overall discontinuation rates (13.0 percent versus 23.0 percent) were numerically lower for patients treated with SGAs than for those treated with the combination regimen. Discontinuation rates because of adverse events were similar for the two treatment groups (6.0 percent versus 4.0 percent).

Second-Generation Antidepressants Compared With Complementary and Alternative Medicines

Second-Generation Antidepressants Compared With St. John’s Wort

All 12 trials comparing SGAs with St. John’s wort provided data on harms or discontinuation rates (see KQ 1, Table 18 for more details on trial design and dosing).^{92,109–114,117,125–128} Two trials were rated as high risk of bias.^112,113

Second-Generation Antidepressants Versus St. John’s Wort: Monotherapy Comparisons

Enough data were available to warrant meta-analyses of overall rates of adverse events and rates of overall discontinuation, discontinuation because of adverse events and because of lack of efficacy, and overall rates of serious adverse events.

Patients treated with SGAs experienced higher overall rates of adverse events, overall discontinuation, and discontinuation because of adverse events than patients treated with St. John’s wort. Discontinuation rates because of lack of efficacy were similar between the treatment groups. In the following paragraphs, we describe the results of these meta-analyses in more detail.

Eight trials, all assigned a low or medium risk of bias rating, reported overall rates of adverse events.^{110,111,114,117,125–128} SGAs were limited to citalopram, fluoxetine, paroxetine, and sertraline. Our random-effects meta-analysis indicated a numerically but not statistically significantly higher overall risk of adverse events for patients treated with SGAs than those treated with St. John’s wort (46.6 percent versus 39.3 percent, respectively; RR, 1.17; 95% CI, 0.95 to 1.44; Figure 22).

Figure 22

SGA versus St. John’s wort: Overall risk for adverse events. AEs = adverse events; CI = confidence interval; SGA = second-generation antidepressant(s); SJW = St. John’s wort

Of note, a high degree of heterogeneity was present because three trials found a higher rate of overall adverse events for patients treated with St. John’s wort, although none of these trials’ risk ratios were statistically significant.^111,126,127 An exploratory analysis to identify the cause of the heterogeneity did not reveal any systematic differences between these three trials and the five showing a higher rate with SGAs; we surmise that the between-trial differences can probably be attributed to chance.

All 12 trials comparing SGAs with St. John’s wort extracts, of which three had a high risk of bias rating,^109,112,113 reported overall discontinuation rates. Random-effects meta-analysis findings based on low and medium risk of bias trials showed that patients treated with SGAs had a statistically significantly higher risk of overall discontinuation than those treated with St. John’s wort (15.5 percent versus 11.8 percent, respectively; RR, 1.28; 95% CI, 1.01 to 1.62; Figure 23). The results of our sensitivity analysis, which included the three high risk of bias trials, were similar (17.5 percent versus 13.6 percent, respectively; RR, 1.25; 95% CI, 1.02 to 1.54; forest plot not shown).

Figure 23

SGA versus St. John’s wort: Overall discontinuation rates. CI = confidence interval; Discont’d = discontinued; SGA = second-generation antidepressant; SJW = St. John’s wort

Eleven of 12 trials, of which two were rated high risk of bias,^109,113 reported rates of discontinuation because of adverse events.^{92,109–111,113,114,117,125–128} Our random-effects meta-analysis found a statistically significantly higher rate of discontinuation because of adverse events among patients treated with SGAs than those treated with St. John’s wort (6.9 percent versus 3.8 percent, respectively; RR, 1.70; 95% CI, 1.12 to 2.60; Figure 24). Our sensitivity analysis, which included the same two high risk of bias trials mentioned above, found similar statistically significant results (SGA: 6.8 percent versus St. John’s wort: 3.8 percent; RR, 1.69; 95% CI, 1.12 to 2.54; data not shown).

Figure 24

SGA versus St. John’s wort: Discontinuation because of adverse events. AEs = adverse events; CI = confidence interval; Discont’d = discontinued; SGA = second-generation antidepressant(s); SJW = St. John’s wort

Six trials reported rates of discontinuation because of lack of efficacy.^{92,110,111,125,126,128} Our random-effects meta-analysis found similar rates of discontinuation because of a lack of efficacy between patients treated with SGAs and those treated with St. John’s wort (2.3 percent versus 2.4 percent, respectively; RR, 0.91; 95% CI, 0.44 to 1.88; Figure 25).

Figure 25

SGA versus St. John’s wort: Discontinuation rates because of lack of efficacy. CI = confidence interval; Discont’d = discontinued; SGA = second-generation antidepressant; SJW = St. John’s wort

Second-Generation Antidepressants Versus St. John’s Wort: Combination Comparisons

We did not find any trials addressing this comparison.

Second-Generation Antidepressant Switching Strategies

In this section, we present the available evidence on the comparative risk of harms from SGA switch strategies compared with other switch strategies following failure of an adequate SGA trial.

Second-Generation Antidepressant Augmentation Strategies

In this section, we present the available evidence on the comparative risk of harms from SGA augmentation strategies compared with other augmentation strategies following failure of an adequate SGA trial.

Second-Step Switch Strategy Compared With Any Augmentation Strategy

We found no eligible trials that directly compared an SGA switch strategy with an augmentation strategy.

Second-Generation Antidepressants Compared With Psychological Interventions

Ten trials comparing SGA monotherapy with psychotherapy alone or in combination with SGAs provided explicit information about the occurrence or nonoccurrence of serious adverse events.^{85,88,96,97,99,101,103,107,108,116,129,131} None of these trials compared between-group differences in the rates of serious events.

Second-Generation Antidepressants Compared With Any Psychological Intervention

Second-Generation Antidepressants Versus Any Psychological Intervention: Monotherapy Comparisons

Eight trials reported the occurrence of one or more serious adverse events;^{85,88,97,99,102,103,107,116,129,131} of these, seven reported data on suicidal ideas or behaviors.^{85,88,97,99,102,103,116,129,131} Rates of suicidal ideas or behaviors ranged from 1.0 percent to 9.0 percent for patients treated with SGAs, and from 0.0 percent to 19.0 percent for patients receiving psychological treatments. One RCT conducted a comprehensive assessment of suicidal ideas or behaviors in patients treated with SGAs (escitalopram) or integrative therapy, specifically IPT.^85,131 We received data from the authors of one RCT that evaluated the presence of suicidal ideas or behaviors at all study timepoints, including baseline,⁸⁸ as well as data from the authors of three other RCTs that reported the incidence of suicidal ideas or behaviors at posttreatment followup.^102,103,116

We were able to conduct meta-analyses of overall rates of suicidal ideas or behaviors using all seven of the above trials that reported relevant data, three of which had a high risk of bias.^{85,88,97,102,131} These trials all compared patients receiving SGAs (paroxetine or sertraline) with those receiving different psychotherapies (CBT, IPT, short-term PSYD, and third-wave CBT). Our primary analysis including only the four low and medium risk of bias trials^{97,99,103,116,129} did not show a statistically significant difference in the rate of suicidal ideas or behaviors between the two groups (9.0 percent versus 7.5 percent, respectively; RR, 1.36; 95% CI, 0.87 to 2.14; Figure 26).

Figure 26

SGA versus any psychological treatment: Rates of suicidal ideas or behaviors. CI = confidence interval; SGA = second-generation antidepressant(s)

In our sensitivity analyses, we included the three high risk of bias trials mentioned above. The results were statistically similar to those of the primary analysis (RR, 0.83; 95% CI, 0.47 to 1.46).

Second-Generation Antidepressants Versus Any Psychological Intervention: Combination Comparisons

Only a single trial with a medium risk of bias compared the risk of any serious adverse events following SGA monotherapy or SGA treatment in combination with psychotherapy.¹⁰¹ Specifically, the rate of suicidal ideas or behaviors in patients treated with fluoxetine (4.4 percent) exceeded that of patients treated with long-term psychodynamic therapy (1.1 percent). While this was a fourfold difference, these findings should be interpreted with caution because of the very small number of events taking place.

Second-Generation Antidepressants Compared With Complementary and Alternative Medicines

Nine RCTs^{92,106,109–111,117,120,125,126} comparing SGA monotherapy with CAM interventions alone or in combination with SGAs provided information about serious adverse events.

Second-Generation Antidepressants Compared With St. John’s Wort

Second-Generation Antidepressants Versus St. John’s Wort: Monotherapy Comparisons

Seven trials comparing SGAs with various extracts of St. John’s wort provided data on serious adverse events.^{92,109–111,117,125,126}

Enough data were available to warrant meta-analyses of overall rates of serious adverse events. Specifically, we included five of the above trials in our analyses (one rated high risk of bias¹⁰⁹). These trials all compared patients receiving different SGAs (citalopram, fluoxetine, paroxetine, or sertraline) with those receiving St. John’s wort.^{109–111,125,126} Our primary analysis with only low and medium risk of bias trials did not show a statistically significant difference in the rate of serious adverse events between the two groups (1.2 percent versus 1.7 percent, respectively, for SGAs or St. John’s wort; Peto OR [odds ratio], 0.66; 95% CI, 0.12 to 3.58; Figure 27). Including the remaining, high risk of bias trial¹⁰⁹ in the sensitivity analysis did not affect the original findings (SGAs: 1.1 percent versus St. John’s wort: 1.7 percent; OR, 0.55; 95% CI, 0.12 to 2.47; forest plot not shown). Because of the low number of events, results should be interpreted with caution.

Figure 27

SGA versus St. John’s wort: Serious adverse events. CI = confidence interval; SGA = second-generation antidepressant; SJW = St. John’s wort

Second-Generation Antidepressants Compared With Exercise

No information about serious adverse events was available from trials comparing SGAs with exercise.

Second-Generation Antidepressant Switching Strategies

Three trials comparing an SGA switch strategy with an SGA or nonpharmacological therapy switch strategy following failure of an adequate SGA trial provided information about serious adverse events.^107,115,140

Second-Generation Antidepressant Augmentation Strategies

In this section, we present the available evidence on the comparative risk of harms from SGA augmentation strategies compared with other augmentation strategies following failure of an initial adequate SGA trial.

Second-Step Switch Strategy Compared With Any Augmentation Strategy

We found no eligible trials that directly compared an SGA switch strategy with an augmentation strategy.