We noted the findings of Kasper et al (Reference Kasper, Stein and Loft2005) and their conclusion that ‘escitalopram was efficacious in treatment of social anxiety disorder’ with interest. They reported a difference of 7.3 (P=0.005) on the Liebowitz Social Anxiety Scale (LSAS) from baseline to week 12, favouring escitalopram over placebo. They suggested that this difference was comparable to three previous studies that reported the efficacy of paroxetine in the treatment of social anxiety disorder (Reference Stein, Liebowitz and LydiardStein et al, 1998; Reference AllgulanderAllgulander, 1999; Reference Baldwin, Bobes and SteinBaldwin et al, 1999).

Unfortunately, without the confidence interval (CI), reliable interpretation of the above difference is not possible. Hence we calculated the standardised effect size, which was 0.22 (95% CI 0.01–0.43). Although the lower limit of the CI is not reassuring, by convention, the point estimate of 0.22 can be interpreted as ‘small’.

We appreciate that small effect sizes can be clinically relevant, especially if the condition treated is common and the putative treatment is easily available, cheap and without adverse effects. In addition, the given treatment must perform better than other options. We compared the above effect size with the effect sizes for the three studies quoted above. These were 0.83 (95% CI 0.53–1.13), 1.36 (95% CI 0.90–1.80) and 0.38 (95% CI 0.14–0.61), respectively.

We then looked at the number needed to treat (NNT) based on the responders as per the Clinical Global Impression–Improvement (CGI–I) scores. The NNT for the study by Kasper et al (Reference Kasper, Stein and Loft2005) is 7 (95% CI 4–20) and for the comparative studies, 4 (95% CI 3–6), 2 (95% CI 2–3) and 3 (95% CI 3–4), respectively. van der Linden et al (Reference van der Linden, Stein and van Balkom2000) reported a meta-analysis of the effectiveness of serotonin reuptake inhibitors (SSRIs) in the treatment of social anxiety disorder. They found a collective NNT of 4 (responders on CGI–I) and a mean effect size for all SSRIs of 1.0 (the SSRI/placebo difference at endpoint on the LSAS). None of the ten SSRI studies in the meta-analysis included escitalopram.

It is tempting to suggest that the placebo response in the study of Kasper et al (Reference Kasper, Stein and Loft2005) was high and distorts results. However, if randomisation is presumed to have been successful, an equivalent placebo effect would have occurred in the escitalopram group. The impressive P values reported by Kasper et al (Reference Kasper, Stein and Loft2005) are likely to be because their study was overpowered and they used analysis of covariance (ANCOVA) which is known to have greater statistical power.

Based on our analysis, among the different SSRI medications escitalopram is less likely to be effective in the treatment of social anxiety disorder. We suggest that P values can mislead and should not be interpreted as measures of magnitude of effect.