Taking psychosocial therapy apart: social support helps most, relaxation appears to harm

Design

Systematic review and Bayesian individual participant data component network meta-analysis of randomised trials, searched to March 2023 with additional datasets to May 2026

Population

34 trials, 30 contributing individual data on 10,612 adults with common mental disorders; 72.2% women, mean age 36.7

Primary outcome

Efficacy of individual treatment components in reducing symptoms at study endpoint, as incremental mean difference

Effect

Social support -9.48 (95% CrI -13.29 to -6.60), behavioural activation -4.15 (-7.48 to -0.05), problem management -4.08 (-5.37 to -2.83); relaxation +7.97 (3.35 to 12.11), cognitive reframing +5.17 (0.78 to 9.15)

Task-shared psychosocial interventions - therapy delivered by trained non-specialists - work for depression and anxiety, and nobody has known which parts of them do the working. This Bayesian component network meta-analysis dismantled them, using a purpose-built taxonomy of treatment elements and individual participant data from 30 of 34 randomised trials: 10,612 participants, 72.2% women, mean age 36.7.

Three components carried the benefit. Strengthening social support was far the largest (incremental mean difference -9.48, 95% credible interval -13.29 to -6.60), then behavioural activation (-4.15, -7.48 to -0.05) and problem management (-4.08, -5.37 to -2.83). Two appeared to make things worse: relaxation (+7.97, 3.35 to 12.11) and, with less confidence, cognitive reframing (+5.17, 0.78 to 9.15). Component effects varied systematically with baseline severity and sociodemographic characteristics, and personalised estimates are available through a public web application.

The detrimental findings need care. A component analysis estimates what adding an element does on average across the trials that contained it, and relaxation is often the comparator arm or the filler content of a lighter intervention - so 'relaxation is harmful' and 'interventions built around relaxation were weaker' are hard to separate. The credible intervals are wide, no ethnicity data were available, and this is not a randomised comparison of components.

What it changes is what you look for when commissioning or supervising a task-shared programme. If the manual is mostly breathing exercises and thought records, this analysis says it will underperform one built on activating the person and mobilising the people around them. That maps directly onto Indian task-sharing programmes, where lay counsellor manuals vary enormously and this is the first evidence about which ingredients to insist on.

• Look for social support mobilisation and behavioural activation in any task-shared manual you commission
• Do not build a lay-counsellor programme around relaxation training as its core component
• Match components to baseline severity - the effects varied systematically with it
• Treat the harm signal for relaxation cautiously: it is confounded with being used as a control condition
• Supervision should check which components are actually being delivered, not just session counts

The statistics, in plain English

Incremental mean difference is the change in symptom score attributable to adding one component, so -9.48 for social support is a substantial move on the pooled symptom scales and +7.97 for relaxation points the other way. A 95% credible interval is the Bayesian equivalent of a confidence interval: behavioural activation's, running to -0.05, only just excludes zero, so it is the least secure of the three beneficial components. The central caution is that components were not randomly assigned - trials chose which elements to include - so this is an association between what an intervention contained and how well it worked.

Read the rest in the app

You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.

Get Daily Doctor — free

Scan to keep reading on your phone. No account needed to start.