Sertraline and exposure therapy in social phobia

I read with interest the article by Haug et al (2003), but was puzzled by the conclusion they drew from their data.

After a 24-week study comparing sertraline, sertraline plus exposure, exposure plus placebo, and placebo in patients with social anxiety disorder (Blomhoff et al, 2001), patients were followed up at week 52. In the summary the authors conclude that ‘Exposure therapy alone yielded a further improvement during follow-up, whereas exposure therapy combined with sertraline and sertraline alone showed a tendency towards deterioration after the completion of treatment’. This seems to be a misleading interpretation of their data.

Haug and colleagues did not mention the primary efficacy measures of their study in their paper. Reading the original paper by Blomhoff et al, I find that the primary efficacy measures were numbers of responders or partial responders on the Clinical Global Impression - Social Phobia (CGI–SP) and the Social Phobia Scale (SPS). In the first study, treatment with sertraline was superior to placebo, but exposure was not. For example, 45.5% of the patients treated with sertraline plus exposure were responders compared with 33.0% of the patients treated with exposure plus placebo. I wonder why it was not mentioned in the second paper whether the three active groups differed from placebo and from each other on the primary efficacy measures.

Instead, Haug et al report only relative changes of mean scores without adjusting for the large absolute differences at termination of the acute study (week 24). After 52 weeks, exposure patients only caught up to the already better scores of the sertraline groups. From both papers, I calculated the following total mean changes for weeks 0–52 by adding the mean changes for weeks 0 to 24 and the ones for weeks 24 to 52 and found: 1.68 for placebo, 2.02 for sertraline plus exposure, 1.92 for sertraline, and 1.88 for exposure plus placebo on the CGI–SP overall severity. For the SPS, I found the following mean changes: 12.09 for placebo, 15.56 for sertraline plus exposure, 14.12 for sertraline, and 15.91 for exposure plus placebo. These scores may change a little bit after correction for participants who withdrew from the trial. I doubt that any of these scores differs significantly from each other or from placebo. By no means is it true that ‘Exposure therapy given alone is more effective in the long term than when given in combination with sertraline’. The opposite is the case: it takes 1 year for the exposure patients to reach the level of improvement that the sertraline and the combination patients have already reached after half a year. Perhaps the patients treated with exposure only showed further improvement during the ‘treatment-free’ follow-up period because one-fifth of them now received treatment with selective serotonin reuptake inhibitors. Remarkably, there was no deterioration in the sertraline groups on the primary efficacy measures, despite the fact that only one-fifth of this group remained on medication.

I have calculated a Bonferroni-corrected critical P-value of 0.0073 when seven scales are used. Thus, all P-values <0.05 and <0.01 given in the paper may be not significant.

I would suggest that the authors analyse their primary efficacy measures and reinterpret their data.