Letter to the editor regarding “Accelerated theta burst stimulation for the treatment of depression: A randomised controlled trial”

Abstract

I read with interest the article by Chen, Fitzgerald et al. [1] “Accelerated theta burst stimulation for the treatment of depression: A randomised controlled trial”. The article reports on a three arm randomised clinical trial involving recruitments across three clinical sites (in three different Australian cities). The authors stated they randomised 299 participants. However, they only reported proportions of response and remission (set as primary outcomes at week 4 endpoint) for 252 participants (although table 4 reports counts for 253 participants). Figure 1 Consort flow chart [1] didn t inform at what time point participants dropout/withdrawal had actually occurred. The assumption of missingness at random is untested. Further, the pattern of missing data is not discussed (monotonic, right censoring). For the treatment arms, the consort flow chart information on missing data is not consistent with the counts reported in table 4 [1]. The authors reported that baseline demographic characteristics were similar across the three treatment arms; however, table 3 [1] shows higher count for female participants and antidepressants use (reads 35/75, 96/10 for low intensity treatment arm, respectively). The authors didn t report model-based marginal or conditional estimates/means for treatment effects, i.e. the baseline covariates haven t been adjusted for. The number of antidepressants is a random count variable with Poisson distribution, yet it is reported as arithmetic mean with large variance. There are no reported estimates or effect sizes (winsorized or traditional) which compares inpatient vs. outpatient participants across the three recruitment sites. The proportions of response and remission are random variables, yet the authors didn t provide standard error estimates for these ratios. The authors stated they implemented a Chi-square test without elaborating further on what type of test they actually ran (Pearson, test of independence) and what were the test(s) degrees of freedom. More importantly, the response/remission rates are timedependent/variant as they were inferred from clustered or matched paired nominal data across 4 time point. Further, the Chi-square values don t seem to have been weighted by attrition rate. The definition of mixed model dictates that fixed and random effects are fitted; however, no random effect variables are defined (or declared independent across sites). The data structure is clearly hierarchal with site (the subject variable for the random effect) and subject (individuals) specific effects. However, this appears to have been ignored. In table 7 [1], the authors reported F statistic values for various models fixed effects (time and treatment group). However, there