Statistics
Independent sample T-tests were used to investigate differences between rTMS and sham treatment in terms of age and HDRS scores (table 2 and 3). Chi-Square tests were performed to check whether the treatment and sham groups differed in terms of time and reason for drop-out, electroconvulsive therapy (ECT) before rTMS treatment, antidepressive medication during the study, inpatient or outpatient, age older than 65 years, personality disorder, type of depression, or side effects (table 3).
A hierarchical linear model was used to analyze the change in HDRS scores over time, and the difference between sham and rTMS treatment, as well as the influence of other characteristics. This model is also known as a multilevel model, or random-effects regression model (Gibbons et al.1993, Snijders 1996). Its main advantage over the standard repeated measures analysis of variance (as, for instance, implemented in the SPSS GLM module) is that it can handle unbalanced data, that is, it does not require the same number of measurements for each subject. Thus, all available measurements can be analyzed, whereas in a standard repeated measures analysis of variance either subjects with incomplete data are omitted from the analysis, or measurement points are discarded (typically, the later time points because of drop-out). This is undesirable because it results in loss of power and disregards the effect of drop-out.
The five HDRS scores (week 1, 2, 4, 8, and 14) were entered in the hierarchical linear model, using the HDRS score before treatment (‘week 0’) as covariate. The change in HDRS score over time was modeled with a piecewise linear model distinguishing between the treatment period (week 1 and week 2) and the post-treatment period. The large fluctuations between patients and measurements were modeled with random effects. The effect of rTMS on the decrease in HDRS score over the two periods was investigated, as well as the influence of covariates, such as sex, age, and left-handedness.
