Introduction
Major depressive disorder (MDD) significantly affects overall mental and physical health (Gold et al., Reference Gold, Köhler-Forsberg, Moss-Morris, Mehnert, Miranda, Bullinger and Otte2020; Herrman et al., Reference Herrman, Patel, Kieling, Berk, Buchweitz, Cuijpers and Wolpert2022; Marx et al., Reference Marx, Penninx, Solmi, Furukawa, Firth, Carvalho and Berk2023) and exerts strain on health services and society (Greenberg et al., Reference Greenberg, Fournier, Sisitsky, Simes, Berman, Koenigsberg and Kessler2021). This may be attributable to its high prevalence (Ferrari et al., Reference Ferrari, Somerville, Baxter, Norman, Patten, Vos and Whiteford2013; Liu et al., Reference Liu, He, Yang, Feng, Zhao and Lyu2020) and recurrent nature (Hardeveld et al., Reference Hardeveld, Spijker, De Graaf, Nolen and Beekman2009), with relapse risk increasing with each major depressive episode (Hardeveld et al., Reference Hardeveld, Spijker, De Graaf, Nolen and Beekman2009; Solomon, Reference Solomon2000). Although approximately half of previously depressed never experience a second depressive episode, for the others, it takes a recurrent course (Monroe & Harkness, Reference Monroe and Harkness2022). Currently, antidepressant medication (ADM) is often the sole relapse prevention strategy used, despite persistently high relapse rates on ADM (Bockting et al., Reference Bockting, Klein, Elgersma, van Rijsbergen, Slofstra, Ormel and Nolen2018; Kato et al., Reference Kato, Hori, Inoue, Iga, Iwata, Inagaki and Tajika2021). Moreover, patients prefer psychological treatment (McHugh et al., Reference McHugh, Whitton, Peckham, Welge and Otto2013), experience side effects from ADM (Anderson et al., Reference Anderson, Pace, Libby, West and Valuck2012; Ferguson, Reference Ferguson2001), and can resist prophylactic properties with long-term usage (Kaymaz et al., Reference Kaymaz, van Os, Loonen and Nolen2008), underscoring the limitations of relying solely on ADM. Hence, better relapse prevention in patients with recurrent depression is vital, and psychological interventions may play a key role.
One of the most robust prognostic indicators of relapse is partial remission of MDD (Böttcher et al., Reference Böttcher, Breedvelt, Warren, Segal, Kuyken and Bockting2023; Buckman et al., Reference Buckman, Underwood, Clarke, Saunders, Hollon, Fearon and Pilling2018; Hardeveld et al., Reference Hardeveld, Spijker, De Graaf, Nolen and Beekman2009; Judd et al., Reference Judd, Akiskal, Maser, Zeller, Endicott, Coryell and Keller1998; Paykel et al., Reference Paykel, Ramana, Cooper, Hayhurst, Kerr and Barocka1995). Partial remission is defined as considerable residual depressive symptoms after a depressive episode (Frank et al., Reference Frank, Prien, Jarrett, Keller, Kupfer, Lavori and Weissman1991; Paykel et al., Reference Paykel, Ramana, Cooper, Hayhurst, Kerr and Barocka1995), such as depressed mood, suicidal ideation, and fatigue. Compared to full remission, partial remission is associated with considerable societal cost (Romera et al., Reference Romera, Perez, Menchón, Delgado-Cohen, Polavieja and Gilaberte2010; Stewart et al., Reference Stewart, Ricci, Chee, Hahn and Morganstein2003; Tranter et al., Reference Tranter, ChB, O’Donovan, Chandarana and Kennedy2002), a higher suicide risk (Sokero et al., Reference Sokero, Melartin, Rytsälä, Leskelä, Lestelä-Mielonen and Isometsä2005), diminished quality of life (Lenox-Smith et al., Reference Lenox-Smith, Martinez, Perahia, Dowsett, Dennehy, Lopez-Romero and Demyttenaere2014; Riihimäki et al., Reference Riihimäki, Sintonen, Vuorilehto, Jylhä, Saarni and Isometsä2016), and worse social functioning (Judd et al., Reference Judd, Akiskal, Zeller, Paulus, Leon, Maser and Keller2000; Lenox-Smith et al., Reference Lenox-Smith, Martinez, Perahia, Dowsett, Dennehy, Lopez-Romero and Demyttenaere2014; Romera et al., Reference Romera, Perez, Menchón, Delgado-Cohen, Polavieja and Gilaberte2010). Clinical guidelines do not discern between fully or partially remitted patients (Gelenberg et al., Reference Gelenberg, Freeman, Markowitz, Rosenbaum, Thase, Trivedi and Van Rhoads2010; NICE, 2022). However, this may be crucial to preventing relapse, as Kuyken et al. (Reference Kuyken, Warren, Taylor, Whalley, Crane, Bondolfi and Dalgleish2016) found symptom severity to moderate effectiveness of mindfulness-based cognitive therapy (MBCT) among 1258 remitted participants. Likewise, Jarrett et al. (Reference Jarrett, Kraft, Doyle, Foster, Eaves and Silver2001) found that partial and unstable remitters especially benefited from the preventive effects of continuation cognitive therapy (C-CT). More personalized care by stratifying for symptom severity, instead of a one-size-fits-all approach, may enhance therapeutic outcomes.
How could partial remission best be treated? Our recent meta-analysis – the only one available specifically examining partial remission – demonstrated that psychological interventions show short-term benefits on depressive symptoms (Gülpen et al., Reference Gülpen, Brouwer, Geurtsen, Van Dis, Denys and Bockting2023). However, these effects were not enduring at one-year follow-up, and importantly, psychological interventions did not reduce the relapse risk (Gülpen et al., Reference Gülpen, Brouwer, Geurtsen, Van Dis, Denys and Bockting2023). The absence of effects is likely due to only three studies reporting longer-term outcomes and three studies examining relapse. Another explanation may be that interventions are effective for some individuals but not others. Such moderating effects were not systematically examined in included studies (Gülpen et al., Reference Gülpen, Brouwer, Geurtsen, Van Dis, Denys and Bockting2023). Studying individual differences in effectiveness to elucidate what works for whom can help target preventive approaches or allow individuals to choose between treatment options.
To address the limited power in randomized controlled trials (RCTs) and previous meta-analytic research, and to examine individual differences in partial remission, an individual participant data meta-analysis (IPD-MA) approach is needed (Riley et al., Reference Riley, Stewart and Tierney2021). First, while our previous meta-analysis included aggregate data from RCTs specifically targeting partial remitters (Gülpen et al., Reference Gülpen, Brouwer, Geurtsen, Van Dis, Denys and Bockting2023), pooling individual participant data allows extraction of data on partial remitters within a higher number of RCTs that recruited both partial and full remitters, increasing statistical power. Second, IPD-MA overcomes aggregation bias and the large number of participant-level datapoints enables thorough assessment of individual-level predictors and moderators (Cuijpers et al., Reference Cuijpers, Ciharova, Quero, Miguel, Driessen, Harrer and Karyotaki2022). For example, using IPD-MA the International Taskforce for Relapse Prevention of Depression and Anxiety (ITFRA) demonstrated that the number of previous depressive episodes moderates relapse prevention outcomes (Breedvelt et al., Reference Breedvelt, Karyotaki, Warren, Brouwer, Jermann, Hollandare and Bockting2024). Finally, IPD-MAs facilitate cross-sectional comparison of fully and partially remitted participants. Identifying participant-specific and intervention-specific variables associated with remission status can help identify barriers to achieving full remission and further support personalization. Earlier efforts did not specifically target partial remission and therefore what works (for whom) remains uncertain (Breedvelt et al., Reference Breedvelt, Brouwer, Harrer, Semkovska, Ebert, Cuijpers and Bockting2021, Reference Breedvelt, Karyotaki, Warren, Brouwer, Jermann, Hollandare and Bockting2024; Kuyken et al., Reference Kuyken, Warren, Taylor, Whalley, Crane, Bondolfi and Dalgleish2016).
In this systematic review and IPD-MA of RCTs using individual participant data from the ITFRA consortium, we aimed to (1) explore which clinical and demographic predictor variables are associated with prospectively assessed relapse among individuals in partial remission of MDD; (2) understand the effectiveness of psychological relapse prevention on depressive relapse risk, depressive symptoms, and quality of life; (3) examine which (if any) individual moderator characteristics, clinical, or demographic, are associated with differential treatment efficacy; and (4) compare fully and partially remitted participants at baseline.
Method
Search strategy
This IPD-MA (PROSPERO: CRD42023463468) was conducted according to the PRISMA-IPD (Supplementary Table 1) and Cochrane recommendations (Higgins et al., Reference Higgins, Thomas, Chandler, Cumpston, Li, Page and Welch2019; Stewart et al., Reference Stewart, Clarke, Rovers, Riley, Simmonds, Stewart and Tierney2015). Building on previous ITFRA initiatives (Breedvelt et al., Reference Breedvelt, Warren, Brouwer, Karyotaki, Kuyken, Cuijpers and Bockting2020, Reference Breedvelt, Karyotaki, Warren, Brouwer, Jermann, Hollandare and Bockting2024), we updated searches and focused on partial remission. Included were:(1) RCTs; (2) comparing psychological interventions versus any control conditions (e.g. treatment-as-usual [TAU], waitlist, or ADM); (3) in relapse prevention, with time to relapse established using clinical interviews (primary outcome); (4) in adult patients (mean age 18–65 years) in remission of MDD, defined as no or subthreshold symptoms for at least eight weeks measured with validated instruments. TAU constituted a control condition, wherein participants received the care they would typically receive outside the trial context, which could encompass receiving no care or ADM. Non-English publications and trials targeting conditions other than or in addition to depression (e.g. substance use disorders) were excluded. We updated searches in Cochrane, Embase, APA PsychInfo, and PubMed through October 18, 2023, using four search strings (Supplementary Appendix 1), expert input and references of meta-analyses. We defined partial remission of MDD, in line with previous research (Frank et al., Reference Frank, Prien, Jarrett, Keller, Kupfer, Lavori and Weissman1991; Tranter et al., Reference Tranter, ChB, O’Donovan, Chandarana and Kennedy2002), as (1) a previous MDD, yet no current episode based on a clinical interview or clinician-based assessment; (2) with at least considerable residual symptomatology remaining, as determined by a depression severity scale using a prespecified Hamilton Depression Rating Scale 17-item (HDRS-17) cutoff score of ≥8 (or transformed scores on alternative scales).
Data collection, extraction, and handling
Authors of included studies were contacted to request data and, upon agreement, received a variable collection sheet stating instructions and requested variables (Supplementary Appendix 2). Shared data were recoded to harmonize and checked for accuracy using a two-step approach, verifying: (1) presence of variables and (2) whether data matched results and variables in published manuscripts (Supplementary Appendix 3). Authors were queried to resolve inconsistencies, confirm data, and request missing variables.
Quality assessment
Three researchers independently rated the risk of bias (RoB) using the Cochrane Risk-of-Bias 2 (RoB-2) tool (Sterne et al., Reference Sterne, Savović, Page, Elbers, Blencowe, Boutron and Eldridge2019), updating earlier assessments using version 1 (Higgins et al., Reference Higgins, Altman, Gøtzsche, Jüni, Moher, Oxman and Sterne2011). Five domains were evaluated: (1) randomization process, (2) deviations from intended interventions, (3) missing outcome data, (4) outcome measurement, and (5) selection of reported results. Domains were judged at low RoB, having some concerns, or high risk, resulting in an overall evaluation using the RoB-2 algorithm. Assessments were piloted, registered (Supplementary Appendixes 4 and 5, for justifications and side-by-side comparisons of RoB-1/RoB-2) and disagreements resolved by senior authors.
Data analysis
Analyses were conducted in Stata (v17) and R (v4.2.1), according to our preregistration (PROSPERO: CRD42023463468) and protocol (Breedvelt et al., Reference Breedvelt, Warren, Brouwer, Karyotaki, Kuyken, Cuijpers and Bockting2020). Participants were classed as in partial or full remission, with partial remission operationalized using prespecified HDRS-17 scores of ≥8. When HDRS-17 scores were unavailable, depressive symptoms scores from alternative scales (e.g., BDI-II) were transformed, using Supplementary Table 2 (Furukawa et al., Reference Furukawa, Reijnders, Kishimoto, Sakata, DeRubeis, Dimidjian and Cuijpers2020; Rush et al., Reference Rush, Trivedi, Ibrahim, Carmody, Arnow, Klein and Keller2003). Partial remission prevalence and relapse rates were calculated in the total sample. Partial and full remitters were compared using χ2 tests for categorical variables and independent samples t-tests for continuous variables. Primary analyses used follow-up data to 12-months, censoring beyond this (Supplementary Appendix 6 for follow-up times and censoring). Participants with missing time-to-relapse data were excluded. Placebo or tapering groups were removed.
Treatment efficacy was determined using 1-stage (merging participant-level data across trials in single models), and 2-stage IPD-MA as our primary analysis (computing effects for each trial separately and then pooling effect sizes), with both random-effects and fixed-effects models for each approach. As ≥2 studies were required to conduct pairwise comparisons (Valentine et al., Reference Valentine, Pigott and Rothstein2010), two clinically relevant comparisons were identified: (1) psychological versus nonpsychological interventions (primary comparison of interest) and (2) psychological interventions plus TAU versus TAU only (secondary comparison).
For 2-stage random-effects models, we used the DerSimonian and Laird method to pool individual results (DerSimonian & Laird, Reference DerSimonian and Laird1986), and applied the Hartung–Knapp–Sidik–Jonkman method to account for τ2 uncertainty (IntHout et al., Reference IntHout, Ioannidis and Borm2014). Hazard ratios (HRs) for each study, pooled HRs, heterogeneity using I 2, 95% confidence intervals (CIs), and forest plots were calculated based on 2-stage IPD-MA. Between-study variation due to heterogeneity, instead of sampling error, was examined using I 2 and considered absent (0–24%), low (25–49%), moderate (50–74%), or high (75–100%) (Ioannidis et al., Reference Ioannidis, Patsopoulos and Evangelou2007). Fixed-effects models used inverse-variance weighting. 1-stage IPD-MA used Cox proportional hazard survival models for our primary outcome.
Depressive symptoms at posttreatment and 60-weeks follow-up (or closest available; range 52–65, mean = 58.2 weeks), and quality of life scores at 60-weeks, were standardized using z-score transformations. Again, 1- and 2-stage IPD-MA were used. In a 1-stage approach using multilevel models, standardized β-coefficients were calculated with participants clustered within trials. Here, higher β-coefficients indicate greater effects of treatment allocation on dependent variables. In 2-stage models, pooled effect sizes (Hedges’ g) were calculated, with positive Hedges’ g indicating treatment superiority, and interpreted as small (0.20–0.49), moderate (0.5–0.79), or large (≥0.8) (Cohen, Reference Cohen1988).
Based on the 2-stage IPD-MA approach, sensitivity analyses were conducted. As we found a substantial portion of participants (n = 134, 6.6%) to have baseline transformed HDRS-17 scores above the MDD cutoff (≥16), analyses excluding these patients were conducted, resulting in attenuated but comparable results (Supplementary Table 3). Small-study effects bias was investigated visually using funnel plots, tested and adjusted using Egger’s test of the intercept (Egger et al., Reference Egger, Smith, Schneider and Minder1997), and with Duval and Tweedie’s trim-and-fill procedure (Duval & Tweedie, Reference Duval and Tweedie2000). Additionally, subgroup analyses – where possible – were performed for different treatment approaches, control conditions and study quality, using meta-regression and mixed-effect models.
To identify clinical and demographic predictors of relapse, predictors were entered in 1-stage survival models, using control-group data only. A study-level variable was included to account for patient clustering. Potential predictors were determined using two-sided p-values, with variables associated with time to relapse with p < .10. Variables were then combined in multivariable models, removing any variable with p > .05 to identify the final list of independent relapse predictors (Ahmed et al., Reference Ahmed, Debray, Moons and Riley2014). For inclusion, predictor and moderator data had to be available for 60% of cases in ≥3 studies, and inclusion had to be justified based on prior literature identifying them as potentially relevant predictors or moderators (Breedvelt et al., Reference Breedvelt, Warren, Brouwer, Karyotaki, Kuyken, Cuijpers and Bockting2020; Buckman et al., Reference Buckman, Underwood, Clarke, Saunders, Hollon, Fearon and Pilling2018).
Cox proportional-hazard assumptions were checked. Finally, moderation of all outcomes was investigated using 1-stage IPD-MA models, adding interaction terms between single moderating variables and binarized treatment allocations. In these models, interactions at the study level were estimated and pooled in random-effects IPD-MA. For models including interaction terms, predictors and moderators were centered around the study mean (continuous variables) or proportion with the covariate (binary variables), to facilitate interpretation and model estimation. In 1-stage IPD-MA interaction models, fixed effects on the study level (i.e., stratified by study) were added to avoid ecological bias, that is, possible discrepancies between group-level associations (across-trial) and individual-level associations (within) (Hua et al., Reference Hua, Burke, Crowther, Ensor, Tudur Smith and Riley2017; Riley et al., Reference Riley, Lambert and Abo-Zaid2010).
Results
Characteristics of studies
In total, 16,644 records, and 241 full-text articles were reviewed. Thirty RCTs (n = 4354) were identified, of which 20 agreed to share data (total n = 3171). Several author-stated factors contributed to an inability to share data (Supplementary Appendix 7), including time constraints and data protection regulations. Upon receipt of data, two studies had no primary outcome data (Biesheuvel-Leliefeld et al., Reference Biesheuvel-Leliefeld, Dijkstra-Kersten, van Schaik, van Marwijk, Smit, van der Horst and Bockting2017; Hoorelbeke et al., Reference Hoorelbeke, Van den Bergh, De Raedt, Wichers and Koster2021), and two used active psychological comparisons (Farb et al., Reference Farb, Anderson, Ravindran, Hawley, Irving, Mancuso and Segal2018; Shallcross et al., Reference Shallcross, Willroth, Fisher, Dimidjian, Gross, Visvanathan and Mauss2018).
This resulted in 16 RCTs (Figure 1 and Supplementary Table 4) with complete outcome data at 12 months for 2028 participants (Bockting et al., Reference Bockting, Schene, Spinhoven, Koeter, Wouters, Huyser and Kamphuis2005, Reference Bockting, Klein, Elgersma, van Rijsbergen, Slofstra, Ormel and Nolen2018; Bondolfi et al., Reference Bondolfi, Jermann, der Linden, Gex-Fabry, Bizzini, Rouget and Bertschy2010; de Jonge et al., Reference de Jonge, Bockting, Kikkert, van Dijk, van Schaik, Peen and Dekker2019; Godfrin & van Heeringen, Reference Godfrin and van Heeringen2010; Hitchcock et al., Reference Hitchcock, Smith, Elliott, O’Leary, Gormley, Parker and Dalgleish2021; Holländare et al., Reference Holländare, Johnsson, Randestad, Tillfors, Carlbring, Andersson and Engström2011; Huijbers et al., Reference Huijbers, Spinhoven, Spijker, Ruhé, van Schaik, van Oppen and Speckens2015; Jarrett et al., Reference Jarrett, Kraft, Doyle, Foster, Eaves and Silver2001, Reference Jarrett, Minhajuddin, Gershenfeld, Friedman and Thase2013; Klein et al., Reference Klein, Santiago, Vivian, Blalock, Kocsis, Markowitz and Keller2004; Klein et al., Reference Klein, Kok, Burger, van Valen, Riper, Cuijpers and Bockting2018; Ma & Teasdale, Reference Ma and Teasdale2004; Moore et al., Reference Moore, Lau, Haigh, Willett, Bosma and Fresco2022; Teasdale et al., Reference Teasdale, Segal, Williams, Ridgeway, Soulsby and Lau2000; Williams et al., Reference Williams, Crane, Barnhofer, Brennan, Duggan, Fennell and Russell2014). Five psychological relapse prevention interventions were examined: cognitive behavioral therapy (CBT; k = 2), C-CT (k = 2), MBCT (k = 7), MemFlex (k = 1), and preventive cognitive therapy (PCT; k = 4). Control conditions consisted of TAU (k = 10), ADM (k = 3), psychoeducation (PE; k = 2), relaxation group therapy (RGT; k = 1), or evaluation only (k = 2). This led to two comparisons: (1) psychological versus nonpsychological interventions (k = 16) and (2) psychological interventions plus TAU versus TAU only (k = 8).
Figure 1. PRISMA Flow Diagram of Individual Participant Data Meta-Analysis Study Selection and Inclusion.
Participant characteristics
The prevalence of partial remission of MDD was 34.8% (705/2028). Partial and full remitters were comparable on most variables (Table 1), but partial remitters more frequently identified as having a non-White background (χ 2 = 9.86, p = .002), were more highly educated (χ 2 = 5.53, p = .019), had longer durations of last depressive episodes (t(949) = 3.00, p = .003), experienced more psychiatric (χ 2 = 7.97, p = .047), and physical comorbidity (χ 2 = 5.23, p = .022), and showed lower therapy completion (χ 2 = 41.20, p < .001). Moreover, partial remitters displayed a higher relapse rate at 12 months (χ 2 = 32.70, p < .001), with 44.4% of partial versus 31.6% of full remitters relapsing. Examining the subset of partial remitters subsequently included in analyses, we found that on average they had depressive onset at age 25.9 years (SD = 12.20) and gone through 5.8 depressive episodes (SD = 8.01), with 77.5% having ≥3 episodes throughout their lifetime. Most previously received pharmacotherapy (64.5%) and psychological treatment (55.9%).
Table 1. Demographic and clinical characteristics, comparing patients in partial remission from MDD to those fully remitted
Note: Data are mean (SD), n, or %.
Abbreviations: BDI = Beck Depression Inventory; HDRS-17 = 17-item Hamilton Depression Rating Scale; MDD = major depressive disorder; SD = standard deviation.
a χ2-tests for categorical variables and independent samples t-tests for continuous variables.
b HDRS-17 scores are transformed from other scales when unavailable.
Predictors of relapse
In examining prospective relapse predictors among partial remitters (Supplementary Table 5), both clinician-rated (HR = 1.05, 95% CI 1.01–1.09, p = .016) and self-reported depressive symptoms (HR = 1.02, 95% CI 1.00–1.04, p = .088), and ≥3 prior depressive episodes (HR = 1.63, 95% CI 1.03–2.58, p = .036), were significantly associated with an increased relapse risk at 12 months at p < .10. Clinician-rated depressive symptoms (HR = 1.22, 95% CI 1.06–1.40, p = .005) and ≥3 prior depressive episodes (HR = 2.17, 95% CI 1.18–3.98, p = .012) remained significant at p < .05 when added in a multivariable model.
Treatment efficacy on preventing relapse
Examining preventive effects using 2-stage random-effects IPD-MA, psychological interventions were associated with a significantly longer time to relapse, compared to nonpsychological control (Figure 2 and Table 2; HR = 0.60, 95% CI 0.43–0.84, p = .006, I 2 = 32.7%, n = 672, k = 14), with 6.7 weeks difference in median survival time. Similarly, psychological interventions with TAU was associated with superior effects on time to relapse compared to TAU alone (HR = 0.51, 95% CI 0.32–0.73, p = .003, I 2 = 0.0%, n = 402, k = 8). Overall, results from 1-stage IPD-MA were similar (Table 2). No significant moderator effects were observed for either comparison using 1-stage IPD-MA models (Supplementary Table 6). Using subgroup analyses, no differences in effects were found based on intervention type (p subgroup = .839) or control conditions (p subgroup = .069). While visual inspection of the funnel plot showed minor asymmetry as evidence of small-study effects bias (Supplementary Figure 2), Egger’s test did not (p = .479). Duval and Tweedie’s trim-and-fill procedure attenuated the effect (HR = 0.72, 95% CI 0.50–1.04, p = .078), indicating that it was overestimated, possibly due to small-study effects bias.
Figure 2. Forest Plot of Random Effects Two-Stage Individual Participant Data Meta-Analysis for Time-to-relapse, Comparing Psychological with Non-psychological Control Interventions.
Note. Jarrett et al. (Reference Jarrett, Kraft, Doyle, Foster, Eaves and Silver2001) and Klein et al. (Reference Klein, Santiago, Vivian, Blalock, Kocsis, Markowitz and Keller2004) were excluded from the two-stage individual participant data meta-analysis models, as there were too few relapses observed. All 16 studies were pooled in the one-stage IPD-MA individual participant data meta-analysis depicted. CBT = cognitive behavioural therapy; C-CT = continuation cognitive therapy; MBCT = mindfulness-based cognitive therapy; PCT = preventive cognitive therapy.
Table 2 1- and 2-stage IPD meta-analysis models of treatment efficacy on time to relapse at 12-months
a Jarrett et al. (Reference Jarrett, Kraft, Doyle, Foster, Eaves and Silver2001) and Klein et al. (Reference Klein, Santiago, Vivian, Blalock, Kocsis, Markowitz and Keller2004) were excluded from the 2-stage models, as there were too few relapses observed.
Treatment efficacy on depressive symptoms and quality of life
Examining effects on depressive symptoms, using 2-stage random-effects IPD-MA, resulted in superior treatment outcomes for psychological interventions versus nonpsychological controls (Table 3 and Supplementary Figure 1), both at posttreatment (Hedges’ g = 0.29, 95% CI 0.04–0.54, p = .025, I 2 = 43.0%, n = 600, k = 15) and 60-week follow-up (Hedges’ g = 0.33, 95% CI 0.06–0.59, p = .020, I 2 = 38.2%, n = 544, k = 14). Again, results from 1-stage IPD-MA were similar (Table 3). Comparing psychological intervention plus TAU versus TAU only, pooled effect sizes using 2-stage models were small and nonsignificant at posttreatment (Hedges’ g = 0.43, 95% CI -0.01–0.88, p = .056, I 2 = 61.2%, n = 353, k = 7) and 60-week follow-up (Hedges’ g = 0.40, 95% CI -0.09–0.89, p = .092, I 2 = 60.1%, n = 338, k = 7). However, 1-stage outcomes were significant at posttreatment and 60 weeks (mean β = −0.53, p < .001, mean β = −0.45, p < .001, respectively). Effects on quality of life were nonsignificant for both IPD-MA models and no further moderator analyses were conducted, due to the limited data available (n = 196, k = 7; Table 3).
Table 3. Results from one- and two-stage IPD meta-analysis models of treatment efficacy on depression severity and quality of life, following partial remission of major depressive disorder
Abbreviations: CI = confidence interval; n.s. = not specified; PI = prediction interval; vs = versus.
a Standardized β coefficients using depression severity scores from the 17-item version Hamilton Depression Rating Scale, Montgomery Åsberg Depression Rating Scale, Inventory of Depressive Symptomatology – Clinician, Beck Depression Inventory I or II.
b Undefined as less than three studies with data available in this comparison.
Using moderator analyses to study differential treatment efficacy on depressive symptoms (Supplementary Tables 7 and 8), marital status significant interacted with efficacy at posttreatment in our first (psychological versus nonpsychological interventions; mean β = −0.49, p = .002) and second comparison (psychological intervention plus TAU versus TAU only; mean β = −0.59, p = .004). Thus, psychological interventions for patients in partial remission – compared to controls – seemed more effective in reducing depressive symptoms at posttreatment for those single, divorced, widowed, or separated. Furthermore, male patients (mean β = −0.52, p = .009), older patients (mean β = −0.01, p = .048), and patients with higher clinician-rated baseline depressive symptom severity (mean β = −0.05, p = .047), had fewer depressive symptoms at 60 weeks when receiving psychological interventions. In our second comparison, higher clinician-rated (mean β = −0.06, p = .017) and self-reported baseline depressive symptoms (mean β = −0.03, p = .027), and male gender (mean β = −0.71, p = .007), were associated with superior efficacy of psychological interventions.
Neither funnel plots (Supplementary Figures 3 and 4) nor Egger’s test provided evidence for small-study effects bias at posttreatment or 60-week follow-up (p = .423 and p = .361, respectively). Duval and Tweedie’s trim-and-fill procedure only slightly changed results at posttreatment (Hedges’ g = 0.33, 95% CI 0.07–0.59, k = 1 added), but not at 60 weeks (Hedges’ g = 0.33, 95% CI 0.06–0.59). Comparing treatment or control conditions was not possible due to a lack of power.
Quality assessment
Overall, risk of bias was considerable (RoB-2 plot in Supplementary Figure 5). Three RCTs were assessed at low risk, three had some concerns, and 10 were at high risk. Preregistrations, protocols, or statistical analysis plans dated prior to study completion were obtained for eight studies. Differences in relapse prevention effects were found for risk of bias (p subgroup = .006), with superior effects in high-risk studies (HR = 0.48, 95% CI 0.34–0.66, p = .001, k = 8), compared to studies having some concerns (HR = 0.49, 95% CI 0.14–1.78, p = .140, k = 3) or at low risk (HR = 1.18, 95% CI 0.37–3.76, p = .596, k = 3). Likewise, effects on depressive symptomatology differed between studies based on their RoB at posttreatment (p subgroup = .016) and 60-week follow-up (p subgroup = .002), with larger effects for higher-risk studies.
Discussion
This IPD-MA of 16 RCTs showed psychological interventions for partial remission of MDD, compared to control conditions, effectively reduce (residual) depressive symptoms immediately after treatment and at 60-week follow-up. Critically, psychological interventions extended depressive time to relapse over 12 months but did not improve quality of life more than control conditions. Interventions are more effective than control conditions in reducing long-term depressive symptoms for individuals with higher baseline symptomatology in partial remitted MDD, indicating they work even better for those most at-risk. We found no evidence that individual characteristics or treatment types have differential effects on relapse prevention. These findings extend our prior meta-analysis of treatment for partial remission (Gülpen et al., Reference Gülpen, Brouwer, Geurtsen, Van Dis, Denys and Bockting2023), including more studies and using participant-level data increasing statistical power. Despite prior reports of worse outcomes for partial remitters (Paykel, Reference Paykel2008; Tranter et al., Reference Tranter, ChB, O’Donovan, Chandarana and Kennedy2002), clinical guidelines and studies have not consistently differentiated them from those fully recovered (Gelenberg et al., Reference Gelenberg, Freeman, Markowitz, Rosenbaum, Thase, Trivedi and Van Rhoads2010; NICE, 2022). Our study provides compelling evidence for the advantages and generalizability of psychological interventions for individuals with partially remitted MDD. Therefore, it may be valuable for guidelines to include recommendations specifically for this at-risk group. As no differences between treatment types were observed, results indicate that partially remitted individuals have several effective treatments available to choose from (8 weekly sessions MBCT or PCT; 6–10 sessions CBT; 4 weeks MemFlex; or 10 sessions C-CT over 8 months).
Importantly, the favorable outcomes of psychological interventions were found despite partial remitters’ particular susceptibility to relapse (Buckman et al., Reference Buckman, Underwood, Clarke, Saunders, Hollon, Fearon and Pilling2018; Hardeveld et al., Reference Hardeveld, Spijker, De Graaf, Nolen and Beekman2009; Paykel et al., Reference Paykel, Ramana, Cooper, Hayhurst, Kerr and Barocka1995). Over 12 months, we found that 44.4% relapsed, compared to 31.6% in the fully remitted group. Among partially remitted individuals in our study, baseline depressive symptoms and the number of prior episodes are associated with a significantly higher relapse risk. This further highlights the relative importance of clinical prognostic factors compared to demographic factors in predicting relapse (Buckman et al., Reference Buckman, Underwood, Clarke, Saunders, Hollon, Fearon and Pilling2018; Hardeveld et al., Reference Hardeveld, Spijker, De Graaf, Nolen and Beekman2009). Overall, psychological interventions – alone or added to TAU – were, relative to control conditions, associated with a longer time to relapse over 12 months in partial remission of MDD. Preventive outcomes were comparable to those of previous IPD-MAs and meta-analyses (Biesheuvel-Leliefeld et al., Reference Biesheuvel-Leliefeld, Kok, Bockting, Cuijpers, Hollon, van Marwijk and Smit2015; Breedvelt et al., Reference Breedvelt, Karyotaki, Warren, Brouwer, Jermann, Hollandare and Bockting2024; Clarke et al., Reference Clarke, Mayo-Wilson, Kenny and Pilling2015; Kuyken et al., Reference Kuyken, Warren, Taylor, Whalley, Crane, Bondolfi and Dalgleish2016). Consistent with this work, we found no differences in efficacy between treatment types. Additionally, we did not find support for differential treatment effects based on age, gender, education level, marital status, age of onset, number of prior depressive episodes, or baseline symptom severity. This indicates their efficacy generalizes across these individual characteristics in partially remitted individuals. Individuals thus have a choice between treatment types, regardless of patient factors. This choice applies to individuals with one or two prior episodes, but also those with ≥3 previous depressive episodes typically considered at increased relapse risk.
Furthermore, our findings indicate the benefit of psychological interventions in reducing residual depressive symptoms immediately posttreatment and at 60-weeks follow-up. However, when comparing psychological interventions plus TAU versus TAU only, results differed between 1- and 2-stage models. Effects on symptomatology were significant using 1-stage, while borderline nonsignificant for 2-stage models. However, most trials are relatively small and we observed moderate heterogeneity using 2-stage models, conceivably leading to underestimation of effects. Additionally, what constitutes TAU varied across trials, often comprising ADM continuation (Supplementary Table 4), though detailed accounts are lacking (Petersson et al., Reference Petersson, Forsén, Björkelund, Hammarbäck, Hessman, Weineland and Svenningsson2023). This lack of clarity may hinder interpretability, replicability, and reliability of findings. In the context of relapse prevention, TAU commonly consists of ADM continuation or no care at all (de Jonge et al., Reference de Jonge, Bockting, Kikkert, van Dijk, van Schaik, Peen and Dekker2019). Interestingly, marital status interacted with posttreatment effects on depressive symptoms. Single, widowed, and separated individuals experienced greater efficacy. This group has previously been found to have a worse depression prognosis (Buckman et al., Reference Buckman, Saunders, Stott, Arundell, O’Driscoll, Davies and Pilling2021), possibly allowing more room for improvement among single or no longer married individuals in our study, leading to enhanced treatment outcomes. Our finding that psychological interventions are more effective in reducing depressive symptoms at 60 weeks for partially remitted individuals, at higher baseline levels of depressive symptoms (HDRS-17 cutoff around 16), is of importance as this highlights that interventions work best for those most affected. In contrast, IPD-MAs on depression found baseline severity to have little impact on symptomatology outcomes after treatment (Bower et al., Reference Bower, Kontopantelis, Sutton, Kendrick, Richards, Gilbody and Liu2013; Furukawa et al., Reference Furukawa, Weitz, Tanaka, Hollon, Hofmann, Andersson and Cuijpers2017). Current findings do hint effects are different for partial remitters, with superior effects for those with higher levels of symptomatology.
No differences were observed for quality of life, contrasting positive results previously reported (Kolovos et al., Reference Kolovos, Kleiboer and Cuijpers2016; Segal et al., Reference Segal, Dimidjian, Beck, Boggs, Vanderkruik, Metcalf and Levy2020). However, our results should be interpreted with caution as data was only sparsely available for this outcome (n = 196, k = 4), changes in quality of life may take longer to manifest compared to depressive symptomatology (McKnight & Kashdan, Reference McKnight and Kashdan2009), and studies varied in their conceptual measurement of quality of life (Supplementary Table 4). Four studies used the EQ-5D to measure quality of life, while this instrument is criticized as insensitive to change in mental health populations (van Kaam et al., Reference van Kaam, Rohrbach, van den Akker-Van Marle, van Furth and Dingemans2024). Additionally, although these analyses were planned, the lack of sufficient power precluded a robust moderation analysis for quality of life.
The current study, done within a systematic review framework, is the first utilizing IPD-MA for evidence synthesis to examine outcomes specifically for those experiencing partial remission of MDD. A key strength lies in the data consolidation – collected by different international groups – not previously published separately for the partial remission subgroup. With this, it helps bridge critical knowledge gaps without resorting to additional resource-intensive trials. Finally, our rigorous data checking provided independent scrutiny of data and ensured its quality, completeness, and consistency.
Nonetheless, some limitations deserve mentioning. First, while based on a predefined and empirical cutoff score established by earlier consensus (Tranter et al., Reference Tranter, ChB, O’Donovan, Chandarana and Kennedy2002), we selected a partial remission subsample. This post hoc analysis of RCTs may introduce selection bias, as selecting this subsample may be influenced by unknown characteristics and is not random. RCTs included individuals in remission and did not specifically target those only partially remitted. One’s symptom severity may affect participation in prevention trials, thereby hampering the generalizability of our findings. Relatedly, some RCTs imposed upper limits on depression severity as inclusion criteria, while others did not, introducing heterogeneity (though we were able to explicitly measure between-study heterogeneity). Second, our review included RCTs examining time to relapse and data for this variable was requested. Studies examining effects on symptomatology or quality of life, without primary outcome data available, were excluded. This may hamper the robustness of results for these outcomes, as such trials are known. Third, we were unable to conduct sensitivity analyses comparing studies providing data to those that did not, as only partial remitters were selected and aggregate data are not commonly published separately for this subgroup, potentially introducing availability bias. Fourth, we relied on a single measurement of depressive symptoms to stratify participants as partial versus full remitters. Given the considerable intraindividual variability in severity over time it would be prudent to use multiple measurements over time to determine remission status and stability. Nevertheless, real-world clinical decision-making would likely also be driven by a single measurement, perhaps including retrospective reporting. Fifth, we could not include all psychological modalities. Importantly, the absence of evidence for modalities not included, for example, interpersonal psychotherapy (no data provided), does not equal evidence of ineffectiveness. Finally, we were unable to examine some relevant factors, including prior psychological or pharmacological treatment during the acute phase of MDD. Future studies should consistently explore and report this to allow for more personalization and potentially staged treatment.
Conclusions
This first-ever IPD-MA examining psychological interventions – specifically for individuals in partial remission of unipolar depression – finds support for their effectiveness in reducing relapse risk as well as residual symptomatology. Crucially, findings underscore that these psychological interventions should be offered to partially remitted individuals, ensuring patient involvement through shared decision-making, as diverse interventions work regardless of patient characteristics. Despite its limitations, this study addresses critical gaps in our understanding of treatment efficacy for partial remission without the need for additional resource-intensive RCTs. Notably, it may provide clinical guidance for treatment decisions in patients that have remitted. Future research should validate these findings, considering additional factors such as dose–response relationships or prior treatment. While research has repeatedly found psychological relapse prevention interventions effective, our study provides the most compelling empirical evidence to date of its effectiveness for individuals in partial remission of MDD, further suggesting they have even superior effects for those with the poorest prognosis (i.e., higher levels of residual symptomatology).
Supplementary material
To view supplementary material for this article, please visit http://doi.org/10.1017/S0033291725000157.
Acknowledgments
The authors wish to thank all the authors that replied to our requests regarding data and for answering our questions. Additionally, the authors wish to thank Bram van Eekhout and Oskar Mook for their assistance in screening and selecting the articles, and RoB assessment.
Funding statement
This research received no specific grant from any funding agency, commercial, or not-for-profit sectors.
Competing Interest
All authors apart from JG (systematic reviewer), JJFB (systematic reviewer), EAVD, GJG, FW, and DAJPD were investigators on 1 or more of the original randomized clinical trials that contributed data and secured grant funding for these trials. CLB has presented clinical training workshops, some of which include a fee. CLB receives royalties from her books and co-edited books, and she developed PCT based on the cognitive model of A. T. Beck. WK is the Director of the Oxford Mindfulness Centre, a collaboration between the University of Oxford and not-for-profit charity, the Oxford Mindfulness Foundation. He was until 2015 an unpaid Director of the Mindfulness Network Community Interest Company. He receives royalties for several books on mindfulness and CBT published by Guilford Press. Since arriving in Oxford (2014) he has either donated any payments for training workshops and presentations to not-for-profit organizations or used them to fund research work. He has advised and consulted various groups on an unpaid basis (e.g., evidence to the UK Mindfulness All Party Parliamentary Group). His post, the Ritblat Professor of Mindfulness and Psychological Science is a Chair endowed by the University of Oxford and the Sir John Ritblat Family Foundation. RBJ is a paid consultant to the National Institute of Mental Health and UpToDate. She has stock equity in Amgen, Johnson and Johnson, and Procter and Gamble. No other disclosures were reported.
Ethical standard
The current study requires no specific ethical approval as it compromises secondary analyses of trials from different RCTs, for which ethical approval was obtained previously. However, the authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008.
Open access
