Archived Comments for:
A randomized double blind placebo controlled clinical trial of N-Acetylcysteine added to risperidone for treating autistic disorders
N-acetylcysteine (NAC) is a very promising psychotropic agent (Dean O et al, J Psychiatry Neurosci 2011). Nevertheless, the conclusion about its efficacy in autism, reported from a RCT by Ghanizadeh A & Moghimi-Sarani E in BMC Psychiatry 2013;13(196), does not seem to be warranted. The authors used a repeated measures ANOVA (Last Observation Carried Forward) approach to test the effect of NAC on the irritability subscale of the ABC inventory. They stated that the effect of groups was significant (p=0.035), but that the time*group effect was not (without showing the F- and p-values). They interpreted this as NAC reducing the irritability score more than placebo. However, as can been seen from Figure 2, it is the mean level of scores during the study, indicated by the group-effect in the statistical model, that differs between treatment groups (NAC or not-NAC). Contrarily, the slopes, indicated by the time*group effect, do not differ markedly. It is the latter effect that shows whether an agent changes the outcome variable more than another agent. Thus, unfortunately, their data give no evidence that NAC improves irritability more than placebo.
Another issue is the choice of outcome variable. The authors do not justify why they chose the irritability subscale and not the entire ABC scale or one of the other subscales as outcome variables. Probably, this was an a posteriori decision. based on the results shown in Table 2,"Between groups difference". However, as argued above, this difference is irelevant to the question of efficacy. Thus, ideally, the primary and secondary hypotheses should have been made a priori (e.g. Main outcome variable : Total ABC score; secondary outcome variables: subscales). As this was a pilot study, it seems justified not to adjust for multiple testing and consider the results as exploratory (Bender R & Lange S, J Clin Epidemiol 2001; Schulz KF & Grimes DA, Lancet 2005).
A last issue is the choice of statistical method. i.e., ANOVA/LOCF versus a linear mixed model (LMM) approach. LMM is now recommeded by the FDA as the optimal statistical method in randomised clinical trials (Siddiqui O et al, J Biopharm Stat 2009). The LMM method is more valid and efficient. Possibly the results and the conclusion will be more more favorable if the data were reanalysed with the outcome variables suggested above as well as applying the LMM approach?
Does N-acetylcysteine decrease irritability in autism?
18 November 2013
N-acetylcysteine (NAC) is a very promising psychotropic agent (Dean O et al, J Psychiatry Neurosci 2011). Nevertheless, the conclusion about its efficacy in autism, reported from a RCT by Ghanizadeh A & Moghimi-Sarani E in BMC Psychiatry 2013;13(196), does not seem to be warranted. The authors used a repeated measures ANOVA (Last Observation Carried Forward) approach to test the effect of NAC on the irritability subscale of the ABC inventory. They stated that the effect of groups was significant (p=0.035), but that the time*group effect was not (without showing the F- and p-values). They interpreted this as NAC reducing the irritability score more than placebo. However, as can been seen from Figure 2, it is the mean level of scores during the study, indicated by the group-effect in the statistical model, that differs between treatment groups (NAC or not-NAC). Contrarily, the slopes, indicated by the time*group effect, do not differ markedly. It is the latter effect that shows whether an agent changes the outcome variable more than another agent. Thus, unfortunately, their data give no evidence that NAC improves irritability more than placebo.
Another issue is the choice of outcome variable. The authors do not justify why they chose the irritability subscale and not the entire ABC scale or one of the other subscales as outcome variables. Probably, this was an a posteriori decision. based on the results shown in Table 2,"Between groups difference". However, as argued above, this difference is irelevant to the question of efficacy. Thus, ideally, the primary and secondary hypotheses should have been made a priori (e.g. Main outcome variable : Total ABC score; secondary outcome variables: subscales). As this was a pilot study, it seems justified not to adjust for multiple testing and consider the results as exploratory (Bender R & Lange S, J Clin Epidemiol 2001; Schulz KF & Grimes DA, Lancet 2005).
A last issue is the choice of statistical method. i.e., ANOVA/LOCF versus a linear mixed model (LMM) approach. LMM is now recommeded by the FDA as the optimal statistical method in randomised clinical trials (Siddiqui O et al, J Biopharm Stat 2009). The LMM method is more valid and efficient. Possibly the results and the conclusion will be more more favorable if the data were reanalysed with the outcome variables suggested above as well as applying the LMM approach?
Competing interests
No competing interests.