Psychological treatment protocols for social anxiety disorder

The short answer

• First choice for adults: individual cognitive therapy for social anxiety disorder (CT-SAD, the Clark and Wells model), or individual CBT on the Heimberg model. This is the NICE CG159 recommendation, and the network meta-analysis behind it supports it.
• Guided internet versions of CT-SAD and of CBT do about as well as face-to-face care in trials, and use much less therapist time.
• Group CBT works, but it is less effective and less cost-effective than individual CBT in the NICE analysis.
• Exposure on its own, without targeting self-focused attention and safety behaviours, is weaker than CT-SAD for social anxiety.

“CBT” is not one treatment. The protocols below differ in what they target, and in how much they help.

The protocols at a glance

The session counts for CT-SAD and the Heimberg model come from NICE recommendations 1.3.13 and 1.3.14.

CT-SAD: why it does well

CT-SAD is built directly on the Clark and Wells (1995) maintenance model: self-focused attention, a distorted observer-perspective self-image, safety behaviours, and anticipatory and post-event processing. Every component targets one link in that loop. What keeps social anxiety going explains the loop in plain language for clients.

The two pieces that set it apart from generic CBT are:

• Safety-behaviour and self-focus experiments early in treatment. The client holds a conversation twice, once with safety behaviours and self-focus, once without, and rates anxiety and how they think they came across. In a small study of eight people, exposure with safety behaviours dropped reduced anxiety more than exposure alone. An experimental study that manipulated safety behaviours and self-focus found both unhelpful in several ways.
• Video feedback, with preparation: predict how you looked in detail, then watch the recording as if it were a stranger.

The trials

In the NICE network meta-analysis (101 trials, 13,164 participants), individual CBT had the largest effect against waiting list (SMD −1.19). It was one of only two treatments that also beat an appropriate placebo: individual CBT against psychological placebo (SMD −0.56), and SSRIs/SNRIs against pill placebo (SMD −0.44). The linked economic model found individual cognitive therapy the most cost-effective of 28 options.

iCT-SAD is now delivered in routine care in NHS Talking Therapies in England (Yim 2024).

Heimberg’s CBT and CBGT

Heimberg’s approach rests on the Rapee and Heimberg (1997) model: the gap between how you think you come across and what you think the audience expects. Treatment combines cognitive restructuring with exposure simulations, first in session and then in real life.

• Group format (CBGT). In Heimberg 1998 (US, n = 133, 12 weeks), CBGT and phenelzine both beat pill placebo and educational-supportive group therapy. Phenelzine worked faster and was ahead on some measures at 12 weeks.
• Combined with phenelzine. In Blanco 2010 (n = 128), remission at 24 weeks was 53.1% with the combination, 25.7% with phenelzine, 23.5% with CBGT and 14.8% with placebo.
• Against mindfulness. In Goldin 2016 (n = 108), CBGT and mindfulness-based stress reduction improved symptoms by similar amounts; CBGT reduced subtle avoidance more.

In the network meta-analysis, group CBT against waiting list was SMD −0.92, below individual CBT. That is why NICE says not to offer group CBT in preference to individual CBT. Groups are still a reasonable option where individual therapy is not available.

Internet CBT (ICBT)

ICBT is the densest body of evidence for social anxiety, much of it from Sweden.

• A study that combined 37 trials (2,991 participants): ICBT g = 0.84 against passive controls and g = 0.38 against active controls.
• Hedman 2011: ICBT was non-inferior to group CBT.
• Five-year follow-up of 80 ICBT patients: LSAS-SR fell from 71.3 to 40.3, with the gains from 1 year kept.
• Across conditions, guided internet CBT and face-to-face CBT were equivalent (pooled g about −0.01), though only 3 of those trials were in social anxiety.
• Furmark 2009: a self-help book on its own and guided ICBT both beat waiting, with gains at 1 year. Several authors receive royalties for the book.

Guidance matters. In a Swedish trial of internet psychodynamic therapy, guided beat unguided (d = 0.46). Most guidelines that recommend self-help mean supported self-help: NICE specifies about 3 hours of support over 3 to 4 months.

Exposure-based protocols

Graded exposure helps. The question is whether it helps as much as cognitive therapy, and for social anxiety the answer is no.

• Ougrin 2011 combined 20 trials (1,308 people) comparing cognitive therapy and exposure across anxiety disorders. There was no difference for panic, PTSD or OCD, but cognitive therapy was better for social phobia, both short and long term.
• In Clark 2006, exposure plus applied relaxation got 42% free of the diagnosis, half the rate of CT-SAD.
• The Canadian guideline calls the evidence on exposure alone versus full CBT “equivocal”.

A likely reason: in social situations, people can do exposure while still using safety behaviours and watching themselves. The feared outcome is never tested. If you use exposure, build in dropping safety behaviours and moving attention outward. Drug add-ons meant to boost exposure learning have not changed end-of-treatment results: D-cycloserine sped up improvement but did not raise response or remission.

Other psychological treatments

• ACT: 12 sessions of ACT and of CBT both beat waiting, with no difference between them (n = 87, US).
• Short-term psychodynamic therapy: in the German SOPHO-NET trial (Leichsenring 2013, n = 495), remission was 36% with CBT, 26% with psychodynamic therapy and 9% on waitlist; response was 60%, 52% and 15%. NICE offers it only to adults who decline CBT and medication.

Caveats: allegiance and what “remission” means

• Developer allegiance. Most CT-SAD trials include David Clark or his colleagues, including the German and Swedish replications. Most ICBT trials come from the groups that built the programmes. Treatments tend to look best in the hands of their developers. Heimberg 1998 tested this directly across a pharmacological site and a CBT site and found few differences, which is reassuring for CBGT. For CT-SAD, routine-care data from the NHS are a start, but large independent trials are still missing.
• Remission is defined differently in different trials: below an LSAS threshold, a global rating, or no longer meeting criteria. Remission in CBT trials ranges from 36% to 100% depending on design. Compare like with like.
• Waiting-list comparisons inflate effects. Use the placebo- and active-controlled figures when advising clients.