Antidepressants and Social Functioning in Patients with Major Depressive Disorder: Systematic Review and Meta-Analysis of Double-Blind, Placebo-Controlled RCTs

To the extent that humans are social beings, social functioning (SF) is pivotal to a successful life. In fact, the WHO definition of health includes complete physical and mental but also complete social well-being [1]. While this is an ideal, the ability to fulfil one’s roles in private and occupational contexts is clearly desirable from both an individual and a societal perspective.

Many psychiatric disorders weaken this ability, and to a certain degree, they are defined by reduced sociability. Schizophrenia, autism, and personality disorders are cases in point, and another example is depression. Furukawa and co-authors, in a long-term study of 44 patients with major depression [2] (major depressive disorder; MDD) found a marked drop in the Social Adjustment Scale score 6 months after illness onset, without recovery during 10 years of follow-up. Similarly, in a Dutch study [3], patients diagnosed with MDD scored considerably lower in the Groningen Social Disability Schedule, covering a broad array of social obligations, and Saris et al. [4] reported comparable educational attainment between patients with depression and controls, but the former were less likely to live in a partnership, had a smaller social network, and were less involved in social activities.

Reduced SF is therefore an important part of the burden associated with depressive disorders. Antidepressant treatment of any kind, it is hoped, improves the ability of patients to interact socially, i.e., improves SF. Per Bech, writing in 2005 [5], urged the field to establish SF as an endpoint in antidepressant studies, and indeed, several trialists measure SF in their RCTs on antidepressant drugs in MDD. Important as the problem is, there is a dearth of summarizing research on the change of SF through antidepressant drug treatment, and, in 2020, Hieronymus et al. [6] called for more results on SF in patients with depression treated with antidepressants.

In pioneering work, Evans and co-authors meta-analysed 17 RCTs found until June 2015 [7]. They restricted their study to the work subscale of the Sheehan Disability Scale (SDS) to arrive at only a moderately positive effect of newer antidepressants. We are not aware of more current meta-analyses or of work considering SF in its entirety, including roles in family and leisure time. Also, in a related meta-analysis [8], our group found a correlation of 0.73 between antidepressants’ effects on changes in depression and in quality of life scores, and we aimed at testing this indicative finding with regard to depression and SF. To address these questions, we carried out a systematic review and meta-analysis of RCTs comparing SF between antidepressant and placebo arms among patients with MDD.

This investigation is part of a larger project on the effects of antidepressant drugs on SF and quality of life [8]. Before the study started, we registered the protocol at the Open Science Framework (https://osf.io/dvza6) [9]. We followed the Cochrane Handbook for Systematic Reviews of Interventions [10] and the PRISMA guidelines [11] in carrying out and reporting the systematic literature review as well as the meta-analyses. The present analysis is modelled after the study on quality of life and antidepressants published earlier [8, 9].

We did not exclude grey literature, and we did not restrict the literature search to certain dates or languages.

We searched for RCTs contrasting antidepressant drugs against placebo in patients with MDD that reported results on SF.

• Diagnosis of MDD based on an established diagnostic instrument, for example, DSM-IV, -5, or ICD-10.

• Presentation of quantitative SF data according to a validated tool, for example, SDS [12], Work Productivity and Activity Impairment Questionnaire (WPAI) [13], or SF-36 (subdomain SF) [14].

• SF results reported for antidepressant and placebo arms.

• Trial duration of at least 6 weeks, assuming that changes in SF take a certain amount of time and may lag psychopathology improvement.

• Trials on antidepressants in other fields of research, such as treatment of acute pain, anxiety, migraine prophylaxis, or stroke rehabilitation.

• Studies focussing on quality of life rather than SF, e.g., studies exclusively reporting SF-36 total scores [14].

• Studies on children and adolescents.

We searched the Cochrane Central Register of Controlled Trials (CENTRAL), NLM databases, specifically Medline and PubMed Central (via PubMed), and PsycINFO. In addition, reference lists of review articles and of all papers eventually included were hand searched for additional studies and grey literature.

We integrated generic and specific search terms for three domains: antidepressant drugs, randomized controlled trials, and SF. The search history is detailed in the study protocol and in online supplementary eTable 1 (for all online suppl. material, see https://doi.org/10.1159/000533494). The last search was conducted in March 2023 and covered articles published until December 31, 2022.

Independently, two authors (S.K., T.W.) screened titles and abstracts of all records retrieved. Both authors read full texts of all reports of studies potentially eligible for inclusion. In all unclear cases, S.K. and T.W. discussed inclusion with the senior author (C.B.).

Independently, two authors (S.K., T.W.) extracted data on the studies included employing a pre-specified and standardized Excel form resembling those used in previous studies of our group [8, 15‒19]. By study, we documented information on number, sex, and age of subjects, diagnosis, interventions, trial duration, psychometric instruments employed, results (point estimates and measures of dispersion), and meta-data regarding authors, journal, and publication year. If available, we collected intention to treat outcomes. In the event that study authors provided data on more than one time point during follow-up, we decided for the longest duration. We solved uncertain cases by discussion among the authors (S.K., T.W., C.B.).

If data points were provided in graphs only, we used Plot digitizer software [20]. We contacted authors via e-mail if results were missing from original reports.

We assessed study risk of bias according to the Cochrane Collaboration RoB tool [10] based on the following items: sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessors, incomplete outcome data, selective reporting of outcomes, and other sources of bias (e.g., enriched design). We assumed a low risk of bias only if all domains had been rated low. Studies were considered having a high risk of bias if at least one domain was rated high. In the event no item was rated high but at least one was rated uncertain, the study was classified as carrying an uncertain risk of bias. Assessments were carried out independently by two raters (S.K., T.W.) and double-checked by the senior author (C.B.).

The effects of antidepressant drugs, compared with placebo, on SF and depressive symptoms were computed as standardized mean differences (SMDs) (Cohen’s d), with 95% confidence intervals and 95% prediction intervals. We used global measures of social function, such as the Sheehan Disability Scale [12], unless SF was covered in subdomains only, for example, in SF-36’s SF items [14].

In the event RCTs reported on more than one SF instrument, we prioritized widely used scales, e.g., SDS. The same way we computed SMDs (Cohen’s d), 95% confidence as well as 95% prediction intervals of antidepressant drug versus placebo contrasts in depression rating scale scores, such as HAM-D [21] or MADRS (Montgomery-Asberg Depression Rating Scale) [22]. If more than one tool was employed in a trial, for comparability, we privileged HAM-D, then MADRS, and then others.

Primary outcome is the effect of antidepressants on SF, computed as SMD (Cohen’s d) between drug and placebo arms and its 95% confidence interval as well as its 95% prediction interval. We also calculated, as secondary outcome, the effect size regarding depression rating scales.

In calculating SMDs we followed the Cochrane Collaboration’s Handbook of Systematic Reviews and Meta-Analysis [10] based either, and preferably, on changes from baseline or on follow-up figures. If those figures were unavailable, we based calculations on correlation coefficients, contingency tables, p values, or other test variables (e.g., t values). If more than one verum arm had been investigated in one RCT, for example, various antidepressant medications or more than one dose of a drug, we combined antidepressant arms to avoid double counting in placebo groups. If standard deviations or standard errors were not provided in original papers, we calculated them from confidence intervals or imputed figures based on similar studies in the sample.

The RCTs included shared certain design properties, such as diagnoses, comparisons of antidepressants versus placebo, blindness, and a limited group of psychometric instruments, but they differed in other respects, for example, in their sex and age distribution or duration of follow-up. Therefore, for summary effects, we used random effect meta-analyses (DerSimonian and Laird). Main results are presented in forest plots, and heterogeneity is given in Q-statistics and I²-values, showing variance in excess of random error. We provide 95% prediction intervals in order to estimate the expected range of effects in populations similar to those sampled in the RCTs selected for the present meta-analysis.

To find evidence of small study effects, we used funnel plots, calculated Egger’s test, and carried out trim-and-fill analyses.

We contrasted effect sizes of studies with low versus studies with uncertain/high risk of bias. We also left each study out at a time to find a signal for undue reliance on specific studies.

We separated two different sorts of RCTs: (1) studies on the treatment of depressive episodes among patients diagnosed with MDD (“primary depression”), and (2) RCTs with patients diagnosed with MDD episodes comorbid with a primary somatic disorder or condition, e.g., Alzheimer’s disease, HIV, or elderly patients – where comorbidity is assumed – (“comorbid depression”).

Post hoc, we applied the GRADE instrument [23] to the outcomes of both meta-analyses – regarding primary and comorbid depression – in order to estimate, in a structured way, the quality of the evidence. GRADE contains five domains – risk of bias, inconsistency, indirectness, imprecision, and publication bias – that need to be rated and result in supporting or downgrading the initially assumed quality of the evidence which, as a default, is considered high for RCTs without important limitations [23].

We present descriptive statistics as means with standard deviations or percentages, as appropriate. For calculating Spearman’s rho, we used Jamovi, version 2.3, and for computing effect sizes, we used Comprehensive Meta-Analysis (CMA), version 2, as well as the Campbell Collaboration’s effect size calculator [24]. Meta-analyses were conducted with CMA, version 2, and with CMA prediction intervals. We present results of meta-analyses as summary estimates of SMDs, SMD (Cohen’s d) with 95% confidence intervals and 95% prediction intervals.

Among patients with MDD (P), to what extent is SF (O) improved by antidepressant drug treatment (I) in comparison with placebo (C)?

We selected 39 papers on 40 randomized controlled trials [25‒63] out of 1,188 titles and abstracts screened. Papers not selected were mostly not original work or not reporting SF measures (PRISMA flowchart in Fig. 1). Owing to clinical heterogeneity, we also excluded four RCTs on maintenance treatment [64‒67].

The papers selected appeared from 1995 to 2020 and included 16,586 patients at baseline, 10,537 randomized to antidepressant and 6,049 to placebo arms, with a weighted mean age of 46.8 years. 64.2% of participants were women. The SDS [12], was the most often employed SF tool, while the Short Form Health Survey, SF-36 [14], social role functioning subsection, the Social Adaptation Self-Evaluation Scale, SASS [68], the Work and Social Adjustment Scale, WSAS [69] as well as the ADCS Activities of Daily Living Scale, ADCS-ADL [70], were applied in three studies, two studies and one study each, respectively. Depressive symptomatology was predominantly assessed with HAM-D [21] and MADRS [22]. Other than treatment as usual, patients did not receive specific psychosocial interventions to improve SF.

Twenty-six papers presented 27 RCTs on acute treatment of MDD (“primary depression”), and 13 publications reported on 13 acute treatment trials among patients with depression comorbid with a physical condition such as kidney disease, chronic heart failure, or dementia (“comorbid depression”). All studies were double-blind. The trials, including their risk of bias, are summarized in online supplementary eTable 2.

The mean effect of antidepressants relative to placebo on SF among patients with primary depression is 0.25 [95% CI: 0.21–0.30], with moderate heterogeneity (I²: 39%) and a prediction interval from 0.10 to 0.41 (Table 1; Fig. 2). Leaving out one study at a time resulted in summary estimates ranging from 0.244 to 0.261, indicating that the summary estimate is not dependent on one specific, particularly influential RCT. There was no indication of publication bias from a funnel plot, from Egger’s test (p = 0.95), or from a trim-and-fill procedure (no study trimmed). The summary effect of the twenty studies carrying a high or uncertain risk of bias was lower compared to the seven low risk of bias trials: 0.23 ([0.18; 0.28]; I²: 27%) versus 0.32 ([0.24; 0.40]; I²: 35%) – a difference that fell short of statistical significance: p = 0.084 (Q-test, df: 1). The effect size regarding depressive psychopathology across all 27 RCTs was 0.28 ([0.23; 0.33] I² 59%).

Among the 13 studies including patients with a physical condition and major depression, the SMD in SF scale scores between antidepressant and placebo arms is 0.24 [0.10–0.37]. Heterogeneity is substantial (I²: 75%) and the prediction interval spans unity: −0.23–0.71 (Table 1; Fig. 3). When we successively removed each study from the analysis, values ranged from 0.201 to 0.271, with the strongest positive and negative influence from the RCTs published by Kornstein et al. [49] and Schneider et al. [58], respectively. While Egger’s test was negative (p = 0.26), the funnel plot suggested a missing study in the lower left quadrant, and, in a trim-and-fill procedure, the summary effect size was reduced to 0.21 [0.07; 0.35]. Ten studies were considered to have a high/uncertain risk of bias trials, while three studies had a low risk of bias. The latter revealed almost no effect on SF at all: 0.04 ([−0.16; 0.24] I²: 0%) versus. 0.29 ([0.13; 0.44] I²: 79%) (p = 0.052; Q-test, df: 1). With regard to depressive psychopathology, the summary SMD was 0.25 ([0.14–0.35], I²: 61%).

At the study level, antidepressants’ efficacy on SF and on depression – both measured in SMDs – are associated: 0.67 (Spearman’s rho, p < 0.001, online suppl. eFig. 1). Nineteen studies in the present sample additionally provided quality of life (QoF) figures measured with different instruments than SF, and SMDs of SF and QoL show a correlation of 0.63 (Spearman’s rho, p < 0.001, online suppl. eFig. 2). When controlling for change in quality of life scores, the correlation between improvement in SF and antidepressant efficacy is moderately reduced (partial correlation [Pearson]: 0.55, p = 0.019, N = 19).

As measured by GRADE [23], the quality of the evidence with regard to the outcome of the meta-analysis on primary depression was high but very low regarding comorbid depression after downgrades for risk of bias, inconsistency, and indirectness (Table 2 provides detailed results of the GRADE approach).

In a sample of 40 studies including about 16,500 individuals with MDD, patients receiving antidepressants experienced slightly better improvements in SF than those treated with placebo. This particularly holds true for the subgroup of RCTs on acute treatment of MDD without specific physical comorbidity – the largest group of studies in this meta-analysis. Here, low heterogeneity, a confirmative risk of bias analysis, and the absence of small study effects (reporting bias) support the analysis of all 27 primary depression studies that arrived at summary result of 0.25 SMD.

Among 13 trials with patients suffering from MDD comorbid with a physical disorder or condition, however, the best studies, that is, the three low-risk bias trials, yielded only a minimal but homogenuous effect (SMD: 0.04), calling into question the summary effect across all thirteen studies of 0.24 SMD. In contrast to primary depression, our GRADE rating of the outcome regarding comorbid depression revealed a very low certainty of evidence. From a clinical perspective, reduced efficacy of antidepressants on SF among patients with comorbid depression makes sense: an additional physical condition or illness, such as kidney or heart failure, may impede SF even after depressive symptoms have subsided.

In a recent discussion of the difficulties in defining a clinically relevant effect size of antidepressants, Hieronymus and co-authors have voiced the hope that SF and quality of life scales may be more sensitive to improvements caused by antidepressant drugs than psychopathology scores [6]. Our present results, as well as an earlier meta-analysis on quality of life [8], conducted in an earlier part of this project, are not in line with such a view. On the contrary, we see a nominally slightly lower benefit in regard to functioning than to psychopathology and, as an additional finding, a high correlation of both measurements, 0.67, indicating the measurements are not independent of one another. This finding echoes an earlier result: Changes in quality of life, both mental and global QoF, and in depression scores in antidepressant drug trials show a high correlation: 0.73 [8]. It is therefore only fitting that, with a correlation of 0.63, SF and QoF are also associated. While measuring different phenomena, the three parameters seem to converge when depressive symptoms subside or improve. Unfortunately, we cannot tell from our data whether SF indeed improves, whether it is largely the patients’ assessment of their social abilities that recovers along with depressive symptomatology or whether it is a combination of both. Objective data describing social roles, e.g., with respect to work status or family obligations, may help in sorting out what is at the heart of the dynamic, although changes in social domains often materialize only slowly. However, naturalistic long-term data from Japan [2] indicate social function has not recovered about 10 years after diagnosis.

From our clinical experience, the ability to fulfil one’s social roles is key to patients. Provided the small effect size seen in our data, it appears necessary to increase efforts to improve SF by non-pharmacological means, for example, by programmes directed at sustained employment or by organizing help to assist patients in managing their household and family duties.

An additional approach to psychosocial interventions is psychotherapy, in particular after medication treatment of acute depression. In a meta-analysis, Guidi and Fava [71] have shown that psychotherapy subsequent to response to antidepressant medication reduces the risk of relapse. It is plausible to expect this approach to have at least some added efficacy with regard to SF as well [72]. With regard to SF and in keeping with a psychotherapeutic approach, a long-term perspective seems to be appropriate for both the patient and the therapist because the ability to fulfil one’s social roles may take longer to recover than core symptoms of depression – especially longer than the six to 8 weeks of antidepressant treatment duration in most of the trials covered in this meta-analysis. Bech, in his review, reports that even though fluoxetine improved emotional role imitation in SF-36 among patients with major depression after 6 weeks of treatment, scores remained well below national norms [5].

Several limitations of this study should be noted: Firstly, we deviated from our originally posted protocol by subdividing the sample of studies into two groups: primary and comorbid depression trials. We do believe, however, that rather than a summary estimate across the entirety of RCTs, the subgroup results are more helpful because they refer to distinct types of patients as we see them in clinical practice. Secondly, we may have missed relevant studies. In comparison to screening titles and abstracts for primary outcomes, secondary outcomes, such as measures of SF, are less often mentioned in summaries, and although we did hand-search reference lists of the papers included, we consider it likely that there are more trials in the literature reporting SF than those covered in the present analysis. It also remains a possibility that data have been published in grey literature and not mentioned in the reference lists of included studies. At the same time, we collected 40 trials, more than twice as many as the only other systematic review we know about. The number of studies appears sufficiently large for primary and, to a lesser extent, comorbid depression. Related is the peril of publication bias in this field of research. Almost all studies were funded by drug manufacturers, and 26 out of 40 studies reported statistically significant results on SF, a skewed distribution, but only under the unlikely assumption that no effect exists. Reassuringly, Eggers’ test and funnel plots remained negative in primary and comorbid depression trials, and in tentative trim-and-fill analyses results did not change meaningfully. Thirdly, we have included a variety of SF instruments, even though the majority of investigators have chosen the Sheehan Disability Score. From a purely statistical point of view, comparing merely one SF scale is desirable, but it is common practice in meta-analysis to combine results from different scales, not least in depressive psychopathology, where a large number of instruments are in use (e.g., HAM-D, MADRS, BDI, PHQ-9, etc.). Had we restricted the analysis to studies measuring SDS scores, we had lost important contributions. At the same time, the low number of investigations employing other scales than SDS precludes reasonable subgroup analyses. Still, we note that, as with depression scales, concepts differ somewhat between SF scales, and what we present is a global estimate of SF rather than a fine grained analysis of its various domains.

Fifth, the studies sampled in this meta-analysis were published after 1994 and cover only newer antidepressants. It is thus unclear whether our results apply to tricyclics, MAO inhibitors, or drugs with antidepressant effects not belonging to the class of antidepressants. Finally, while all investigations were conducted double-blind, it is possible masking was compromised, although trial authors only very rarely comment on blinding success [73].

In forty placebo-controlled and double-blind RCTs of antidepressants among patients with MDD, we calculated a modest effect on SF. In primary depression, the effect ranges between 0.2 and 0.3 SMD, which is the same order of magnitude as the effect of antidepressants on depressive psychopathology (e.g., 0.30 [0.26–0.34]) [74]. Results for comorbid depression and maintenance trials appear uncertain at this point. Improvements in SF are closely associated with both improving depressive psychopathology and quality of life scores, suggesting construct overlap as well as redundancy in the way the parameters are currently investigated. Since fulfilling social obligations is of enormous importance to patients, the field needs to increase its efforts in finding means beyond antidepressant drugs to help our patients live a successful social life.

An ethics statement is not applicable because this study is based exclusively on published literature. In its reporting, this manuscript follows PRISMA.

The authors declare no conflict of interest.

This study has not been externally funded.

Conception and design and interpretation of data: C.B. and T.B. Acquisition of data: S.K., T.W., and C.B. Analysis of data: S.K. and C.B. Drafting the manuscript: C.B. Revising the manuscript critically for important content: S.K., T.W., and T.B. Final approval of the manuscript and agreement to be accountable: S.K., T.W., T.B., and C.B.