An individual participant data meta-analysis of psychological interventions for preventing depression relapse

Major depressive disorder (MDD) is a leading cause of global disability and is characterized by high relapse rates, making prevention of relapse a clinical priority. Antidepressant medication (ADM), psychological interventions, or their combination are commonly used to prevent relapse, either as continuation or sequential strategies. Prior systematic reviews and meta-analyses suggest psychological interventions can be viable alternatives or adjuncts to ADM with sustained effects. However, questions remain about which intervention to offer to whom. Individual participant data meta-analysis (IPDMA) can enhance personalization by enabling more precise assessment of predictors and moderators compared with aggregate meta-analyses. Previous IPDMAs have focused on specific interventions (e.g., MBCT) or specific comparisons (e.g., tapering ADM vs continuation), leaving a broader set of psychological interventions and moderators unexplored. This study conducts an IPDMA of RCTs comparing psychological interventions for previously depressed adults against ADM, treatment as usual (TAU), or evaluation-only control, aiming to determine effectiveness and identify for whom these interventions are most beneficial.

Two prior IPDMAs addressed relapse prevention. Kuyken et al. showed MBCT (with TAU or with ADM tapering) reduced relapse risk versus control (HR 0.69; 95% CI 0.58–0.82) and found baseline depressive symptom severity moderated outcomes. Another IPDMA comparing psychological intervention during ADM tapering versus ADM continuation found no difference in time to relapse and no moderators. Aggregate data meta-analyses have indicated psychological interventions (CBT, continuation cognitive therapy (C-CT), preventive cognitive therapy (PCT), MBCT) can prevent relapse, with mixed findings on moderators such as number of previous episodes. A broader evaluation across multiple psychological interventions and controls, and a wider moderator set, remained needed.

Design: Individual participant data meta-analysis (one- and two-stage) of randomized controlled trials comparing psychological interventions for relapse prevention versus control in adults with MDD in remission. Protocol registered on PROSPERO (CRD42019127844); PRISMA-IPD followed. Eligibility: RCTs in adults (mean age 18–65) in remission from MDD (no or subthreshold symptoms for ≥8 weeks or per study definition), comparing a psychological intervention with control (TAU, wait-list, ADM, psychological placebo, or another psychological intervention), with primary outcome time to relapse assessed via diagnostic interview (e.g., SCID). Only English-language studies included. Study identification: Searched PubMed, PsycINFO, Embase, and Cochrane Central on 23 Jan 2021 using terms for depression, relapse/recurrence, preventive interventions, and RCTs; screened reference lists and consulted an international task-force. Screening performed independently by multiple reviewers; conflicts resolved with senior author. Data collection: Contacted study authors for IPD; provided a variable collection sheet. Received datasets were verified against publications for completeness and accuracy (participant counts, means, s.d., relapse rates). Risk of bias: Assessed with updated Cochrane Risk of Bias tool (random sequence generation, allocation concealment, blinding of outcome assessors, incomplete outcome data, selective outcome reporting, other threats). Overall risk rated low if ≥5 criteria low risk. Outcomes and variables: Primary outcome was time to relapse (weeks) to 12 months post-randomization, censored at 12 months if longer follow-up available. Seventeen predefined variables were requested, including demographics and clinical characteristics (age, sex, ethnicity, education, employment, marital status, treatment group, number of previous episodes, age of onset, time in remission, duration of last episode, remission stability, prior psychological intervention, comorbid mental/physical conditions, baseline depression via BDI and HAM-D). Statistical analysis: Two-stage random-effects meta-analyses using Hartung–Knapp–Sidik–Jonkman variance estimator based on DerSimonian–Laird τ²; effect size hazard ratio (HR). Heterogeneity via I². Pairwise comparisons conducted when ≥2 studies available: (1) psychological interventions (alone, with TAU or ADM) vs any non-psychological control (TAU, ADM, evaluation-only), and (2) psychological interventions + TAU vs TAU only. One-stage fixed-effects Cox proportional hazards models (fixed effect for study) were used to identify predictors in control groups and moderators (treatment-by-covariate interactions). Predictors considered if ≥60% data availability across studies; predictors with P<0.10 in bivariable control-group analyses entered multivariable model; final predictors selected at P<0.05. Interaction terms between treatment and each predictor tested for moderation (P<0.05). Sensitivity analyses: Compared effects between studies providing IPD vs those not (using aggregate data), assessed small-study effects via funnel plots and Egger’s test, and performed sensitivity excluding a study with chronic depression participants. Risk-of-bias, subgroup, and fixed-effects analyses presented in appendices.

- Included studies: Of 28 eligible studies (n=4,053), 18 provided IPD (n=2,840). After exclusions (no time-to-relapse outcome or active psychological control), 14 RCTs with IPD (n=1,725; analyzable n=1,720) were included, testing PCT, CBT, MBCT, and C-CT against ADM, TAU, or evaluation-only controls. Mean participant age 45.1 years; 73% female; average previous episodes 4.8; 75% had ≥3 prior episodes. - Main effects (two-stage random-effects): Psychological interventions vs non-psychological controls significantly reduced relapse risk over 12 months: HR 0.60 (95% CI 0.48–0.74), P≤0.000; I²=14.9% (14 studies; n=1,720; 581 relapses). Psychological interventions added to TAU vs TAU only: HR 0.62 (95% CI 0.47–0.82), P=0.005; I²=28.3% (8 studies; n=1,191; 475 relapses). No significant differences between intervention types (MBCT, PCT, CBT, C-CT) in subgroup analyses. - Predictors of relapse (control groups; one-stage fixed-effects): In bivariable models, more previous episodes, being single/divorced/separated/widowed, lower age of onset, and higher residual depressive symptoms (HAM-D) increased relapse risk (P<0.10). In the multivariable model, independent predictors were number of previous depressive episodes (HR 1.03 per episode; 95% CI approx. 1.00–1.06; P=0.04) and baseline residual symptoms via HAM-D (HR 1.08 per point; 95% CI approx. 1.04–1.13; P=0.00). Marital status and age of onset were not retained after adjustment. - Moderators of treatment effect: No significant interactions for the broad comparison (psychological interventions vs non-psychological controls). For psychological interventions + TAU vs TAU, number of previous episodes moderated effect: those with ≥3 episodes benefited more (interaction HR 0.58 (0.34–0.99), P=0.047; similarly reported HR 0.55 (0.37–0.79), P=0.006 for three-or-more vs ≤2 episodes). No significant benefit detected for those with ≤2 episodes (two episodes: HR 0.85 (0.37–1.92), P=0.613; one episode: HR 1.48 (0.40–5.53), P=0.556), noting small subgroup sizes (n=182 with two episodes; n=32 with one). - Risk of bias: Generally low; blinding of participants/personnel high risk as expected for psychotherapy trials; outcome assessors typically blinded; low selective reporting and good ITT adherence. - Sensitivity analyses: No evidence of data availability bias (IPD-providing vs non-providing studies comparable). Funnel plots: little evidence for small-study effects at 12 months (Egger’s P=0.34); some evidence at 14 months (Egger’s P=0.01) without extreme asymmetry. Excluding a chronic depression trial did not change results.

This IPDMA demonstrates that psychological interventions (MBCT, PCT, CBT, C-CT), alone or added to usual care, significantly delay time to relapse in adults in remission from MDD compared with non-psychological controls. The findings support integrating psychological relapse-prevention strategies into ongoing care. No single psychological modality outperformed others, suggesting patients and clinicians can choose among evidence-based options according to preference and availability; however, limited sample sizes within specific modalities advise cautious interpretation. A key moderation finding is that individuals with three or more prior depressive episodes derive greater benefit from adding psychological interventions to TAU. This aligns partly with prior aggregate meta-analytic evidence but contrasts with some previous IPDMAs, possibly due to differences in included populations and expanded moderator assessment here. Predictors independent of treatment—higher residual depressive symptoms and greater number of prior episodes—should inform risk stratification. Contrary to expectations from some prior literature, baseline residual symptoms did not identify differential treatment benefits, implying that while such symptoms indicate higher overall risk, they may not guide selection among psychological interventions versus TAU. The results address the research question by quantifying overall effectiveness and by identifying for whom (those with ≥3 prior episodes) psychological interventions are especially beneficial, informing personalized relapse-prevention strategies.

Psychological interventions are effective in reducing 12-month relapse risk in adults with remitted depression, both versus non-psychological controls and when added to treatment as usual. The number of prior depressive episodes and residual depressive symptoms are key predictors of relapse risk, with evidence that patients with three or more prior episodes particularly benefit from adding psychological interventions to TAU. Clinical practice should incorporate psychological relapse-prevention strategies, especially for higher-risk patients, while allowing choice among validated modalities. Future research should include longer follow-ups, larger samples of patients with one or two prior episodes, additional candidate predictors (e.g., childhood trauma, socioeconomic status, cognitive styles), and consider IPD network meta-analyses to compare multiple interventions concurrently and explore mechanisms of change.

- Follow-up censored at 12 months; predictor and moderator performance over longer periods remains unknown. - Only RCTs randomizing after remission were included; naturalistic long-term follow-ups of acute treatments were excluded; gray literature not searched. - Some requested predictors (e.g., childhood trauma, socioeconomic status, employment details, cognitive processing styles) were unavailable or inconsistently measured; only variables with ≥60% availability were analyzed. - Limited numbers in certain subgroups (especially patients with one or two prior episodes) reduced power to detect moderation effects. - Not all eligible studies provided IPD (64% provided data); although sensitivity analyses suggested no availability bias, data sharing constraints limit completeness. - Could not assess interpersonal psychotherapy (IPT) or other less-studied interventions due to lack of IPD; conclusions cannot be drawn for these modalities. - Potential allegiance bias given inclusion of trial investigators as co-authors; mitigated by inclusion of diverse interventions, independent leadership/statistical oversight.