We thank Drs Lele and Joglekar for drawing our attention to the absence of the 95% CIs for the primary efficacy end-point (treatment effect measured as the difference in the Liebowitz Social Anxiety Scale (LSAS) scores from baseline) in our article on the treatment of social anxiety disorder with escitalopram (Reference Kasper, Stein and LoftKasper et al, 2005). The treatment difference between escitalopram and placebo was 7.3 (95% CI 2.2–12.4) with a standardised effect size of 0.30 (95% CI 0.09–0.51).

When comparing the results of this trial with the literature we looked at the size of the effect of the active treatment, that is, the adjusted change from baseline in LSAS scores, not the standardised effect size. These values are 33.0 (Reference AllgulanderAllgulander, 1999), 29.4 (Reference Baldwin, Bobes and SteinBaldwin et al, 1999) and 30.5 (Reference Stein, Liebowitz and LydiardStein et al, 1998), which are comparable to the 34.5 change in our study with escitalopram (Reference Kasper, Stein and LoftKasper et al, 2005). The main difference between these studies is the placebo response, which was largest in our study.

In interpreting differences in placebo response rate (and hence standardised effect sizes) it is important to recognise differences in study design. One of the paroxetine studies (Reference AllgulanderAllgulander, 1999) was a small (n=92) single-centre trial with a 40% placebo withdrawal rate (compared with 18% for paroxetine) and patients were also required to have been treated for at least 2 weeks. These factors may be responsible for the small placebo effect with the last observation carried forward (LOCF) analysis. In the studies of Allgulander (Reference Allgulander1999) and Stein et al (Reference Stein, Liebowitz and Lydiard1998) patients were not excluded if they had comorbid depression, which was the case in our study. Finally, in our escitalopram study the mean baseline LSAS scores in the placebo and treatment groups (95.5 and 96.3) were higher than in the paroxetine studies (70.4 and 78.5 in Reference AllgulanderAllgulander, 1999; 78.0 and 83.5 in Stein 83.5 in Reference Stein, Liebowitz and LydiardStein et al, 1998; and 86.1 and 87.6 in Reference Baldwin, Bobes and SteinBaldwin et al, 1999).

We would like to emphasise the appropriate powering of our study. ANCOVA is overpowered if the distribution is skewed but our data are fairly normally distributed. Allgulander (Reference Allgulander1999) state that their data were skewed and non-parametric tests were used.

In line with the results of our study additional recent data (Reference Lader, Stender and BurgerLader et al, 2004) confirm the efficacy of escitalopram in social anxiety disorder. In a 24-week study the placebo response was 43.4 compared with 60.8 with 20 mg escitalopram and 53.1 with 20 mg paroxetine (mean change from baseline). The treatment difference (observed cases) between escitalopram and placebo was 17.4 (95% CI 11.5–23.2) with a standardised effect size of 0.77 (95% CI 0.51–1.03). The treatment difference for escitalopram and paroxetine (observed cases) was 7.71 (95% CI 2.0–13.4) in favour of escitalopram with a standardised effect size of 0.34 (95% CI 0.09–0.59). After 12 weeks the number needed to treat (NNT) based on the responders as per Clinical Global Impression–Improvement (CGI–I ≤2, LOCF) scores for Kasper et al (Reference Kasper, Stein and Loft2005) was 6.4 (95% CI 4–19) and 4.8 (95% CI 3–10) for Lader et al (Reference Lader, Stender and Burger2004). To judge a single medication based on the NNT it is necessary to consider all available studies and, based on the evidence published in the literature, we therefore do not agree with the statement of Drs Lele and Joglekar that paroxetine is superior to escitalopram for the treatment of social anxiety disorder.