Discussion
The results of our study involving patients with a DSM IV diagnosis of major depression suggest that rTMS treatment has modest antidepressant effects. Depression scores decreased by 20% in both sham and rTMS groups during treatment but decreased further during 12 weeks of follow-up in the rTMS group only. Two studies conducted under similar conditions (left DLPFC, intensity 80%, 20Hz) revealed superior results for rTMS compared to placebo. George et al (1997)found a small decrease 2 weeks after cessation of treatment for 2 weeks. Berman et al, (2000), found a decrease over the 2-week course of rTMS, but, in contrast to the study of George et al (1997) and our study, benefits were no longer evident 2 weeks after the treatment had ended.
The inter individual variability may be associated with equivalent large variability in human neurophysiologic responses to specific brain interventions (Maeda et al, 2000). The degree of variability of rTMS effects in the human brain can be lowered somewhat by increasing the number of rTMS pulses to 1600 (Maeda et al, 2000), but still remains high.
The rather large number of drop-outs, which is quite common in a longitudinal study, caused missing observations. For this reason, we used the random effects model, which analyzes all available observations and which provides a more realistic approach to the analysis of longitudinal psychiatric data (Gibbons et al, 1993). Random regression models provide solutions for commonly observed problems of missing data, serial correlation, time-varying covariates, and irregular measurement times. Moreover, they accommodate systematic person-specific deviations from the average time trend (Gibbons et al, 1993). Mazumdar et al (1999) paid attention to the problem of missing values in longitudinal clinical trials. In data from clinical studies, the method of carrying the last observation forward has the major objection that strong time trends are often seen, and such time trends can very easily be confounded with treatment effects. Another model that is often used in intention-to-treat- analyses is the repeated-measures-model, which is mainly based on the assumption that observations are missing at random, an assumption that is rarely justified in psychiatric research.
Our rTMS treatment protocol was based on the method used in earlier studies (George et al 1997, Pascual-Leone et al 1996). Most studies, particularly those over the left dorsolateral prefrontal cortex use a figure of eight coil.
At the time this study was started, our method was more usual, but in the light of the current views of “power” of studies with TMS, our study is completely underpowered. Moreover, the sham rTMS we used was a 45 degree angle sham. Angling the coil at 45 degrees is not really a sham because it does induce stimulation of cortical tissue (Lisanby S, et al, 2001). Unfortunately, there appeared to be a fatal flaw in the methodology we used for stimulation. A circular coil creates the strongest stimulation at the outer edge of the donut, the stimulation being weakest or non-existent in the center, so in our method there would be no electrical activity at the DLPFC, but stimulation at distant regions of the cortex.
These methodological flaws might explain the modest improvement at the end of the first two weeks in the active group and the small difference at this point between the two groups. One more point that should be made is that 80% MT is now considered to be a very low stimulus intensity. Mark George found only a small decrease in de HDRS in 1997 when he used 80% MT but a larger decrease when in 2000 he used 100% MT. Others have had similar experiences.
At the time of our trial, improvement after the cessation of treatment had never been demonstrated before, nobody else had taken a routine, close look, such as we did. This may make our study a valuable contribution to TMS research.
