Short-term efficacy of repetitive transcranial magnetic stimulation (rTMS) in depression- reanalysis of data from meta-analyses up to 2010

Background According to a narrative review of 13 meta-analyses (published up to 2010), repetitive transcranial magnetic stimulation (rTMS) has a moderate, short-term antidepressant effect in the treatment of major depression. The aim of the current study was to reanalyse the data from these 13 meta-analyses with a uniform meta-analytical procedure and to investigate predictors of such an antidepressant response. Methods A total of 40 double-blind, randomised, sham-controlled trials with parallel designs, utilising rTMS of the dorsolateral prefrontal cortex in the treatment of major depression, was included in the current meta-analysis. The studies were conducted in 15 countries on 1583 patients and published between 1997–2008. Depression severity was measured using the Hamilton Depression Rating Scale, Beck Depression Inventory, or Montgomery Åsberg Depression Rating Scale at baseline and after the last rTMS. A random-effects model with the inverse-variance weights was used to compute the overall mean weighted effect size, Cohen’s d. Results There was a significant and moderate reduction in depression scores from baseline to final, favouring rTMS over sham (overall d = −.54, 95% CI: −.68, −.41, N = 40 studies). Predictors of such a response were investigated in the largest group of studies (N = 32) with high-frequency (>1 Hz) left (HFL) rTMS. The antidepressant effect of HFL rTMS was present univariately in studies with patients receiving antidepressants (at stable doses or started concurrently with rTMS), with treatment-resistance, and with unipolar (or bipolar) depression without psychotic features. Univariate meta-regressions showed that depression scores were significantly lower after HFL rTMS in studies with higher proportion of female patients. There was little evidence for publication bias in the current analysis. Conclusions Daily rTMS (with any parameters) has a moderate, short-term antidepressant effect in studies published up to 2008. The clinical efficacy of HFL rTMS may be better in female patients not controlling for any other study parameters. Electronic supplementary material The online version of this article (doi:10.1186/s40359-014-0039-y) contains supplementary material, which is available to authorized users.


Table of Contents
PRISMA checklist (Moher et al. 2009 Objectives 4 Provide an explicit statement of questions being addressed with reference to participants, interventions, comparisons, outcomes, and study design (PICOS).

METHODS
Protocol and registration 5 Indicate if a review protocol exists, if and where it can be accessed (e.g., Web address), and, if available, provide registration information including registration number.
-Eligibility criteria 6 Specify study characteristics (e.g., PICOS, length of follow-up) and report characteristics (e.g., years considered, language, publication status) used as criteria for eligibility, giving rationale.

3
Information sources 7 Describe all information sources (e.g., databases with dates of coverage, contact with study authors to identify additional studies) in the search and date last searched.
Tables S1-S2 Search 8 Present full electronic search strategy for at least one database, including any limits used, such that it could be repeated.
Table S1 Study selection 9 State the process for selecting studies (i.e., screening, eligibility, included in systematic review, and, if applicable, included in the meta-analysis).
3; Figure 1 Data collection process 10 Describe method of data extraction from reports (e.g., piloted forms, independently, in duplicate) and any processes for obtaining and confirming data from investigators.
3 Data items 11 List and define all variables for which data were sought (e.g., PICOS, funding sources) and any assumptions and simplifications made.
3; Tables 1-2 Risk of bias in individual studies 12 Describe methods used for assessing risk of bias of individual studies (including specification of whether this was done at the study or outcome level), and how this information is to be used in any data synthesis.

4; 10
Summary measures 13 State the principal summary measures (e.g., risk ratio, difference in means). 4 Synthesis of results 14 Describe the methods of handling data and combining results of studies, if done, including measures of consistency (e.g., I 2 ) for each meta-analysis.

Risk of bias across studies
15 Specify any assessment of risk of bias that may affect the cumulative evidence (e.g., publication bias, selective reporting within studies).

10
Additional analyses 16 Describe methods of additional analyses (e.g., sensitivity or subgroup analyses, meta-regression), if done, indicating which were pre-specified.

Study selection
17 Give numbers of studies screened, assessed for eligibility, and included in the review, with reasons for exclusions at each stage, ideally with a flow diagram.
3-4; Figure 1, Table S2 Study characteristics 18 For each study, present characteristics for which data were extracted (e.g., study size, PICOS, follow-up period) and provide the citations.

DISCUSSION
Summary of evidence 24 Summarize the main findings including the strength of evidence for each main outcome; consider their relevance to key groups (e.g., healthcare providers, users, and policy makers).

11; 13-16
Limitations 25 Discuss limitations at study and outcome level (e.g., risk of bias), and at review-level (e.g., incomplete retrieval of identified research, reporting bias).

15-16
Conclusions 26 Provide a general interpretation of the results in the context of other evidence, and implications for future research. 11; 13-16

FUNDING
Funding 27 Describe sources of funding for the systematic review and other support (e.g., supply of data); role of funders for the systematic review.

16
Mathematical approach used in the current meta-analysis The approach to meta-analysis and all formulae in this document are based on the method of Hedges' et al. (Borenstein et al. 2009).

Combining of data from independent (active rTMS) subgroups within studies
Some studies in the current meta-analysis compared one sham group with more than one active rTMS groups. Thus, for the purposes of the overall meta-analysis the data from the multiple active subgroups were combined into one active rTMS group to compute only one effect size for the study. Using one example study (Padberg et al. 1999), the mean (M) and standard deviation (SD) of depression scores for the two active subgroups (stimulation frequencies of either 10 Hz or .3 Hz) were combined into one group at each of the two points in time (pre and post-treatment) according to the following formulae (Borenstein et al. 2009 p. 222): • The combined mean depression score for the 'active-pre' group (M 1+2 ) was computed by weighing the mean depression score of subgroup 1 (M 1 ; 10 Hz) and subgroup 2 (M 2 ; .3 Hz) based on the sample size of each subgroup (N 1 and N 2 ): The combined mean depression score for the other group ('active-post') was computed the same way.
• The combined standard deviation of the mean depression scores for the 'active-pre' group (SD 1+2 ) was computed using individual SD and N values of subgroup 1 (10 Hz; SD 1 and N 1 ) and subgroup 2 (.3 Hz; SD 2 and N 2 ): The combined standard deviation of the mean depression scores for the other group ('active-post') was computed the same way.
• The combined sample size for the 'active-pre' group (N 1+2 ) was computed by adding the sample sizes of the two subgroups (N 1 + N 2 ). The combined sample size for the other group ('active-post') was computed the same way.
The study by Stern and colleagues (Stern et al. 2007), was performed on three active subgroups (10 Hz left DLPFC, 1 Hz left DLPFC, and 1 Hz right DLPFC). Thus, the two left-stimulation subgroups (10 Hz and 1 Hz) were combined first and these (combined) scores were then combined with the scores of the rightstimulation subgroup according to the formulae described above.

Combining of data in dependent subgroups at different points in time (pre and post)
Since data were collected from the same groups (sham or active) twice (pre and post treatment) it was necessary to reduce them to one score/group for the purposes of meta-analysis. Following the approach of Holtzheimer and colleagues (Holtzheimer et al. 2001), such reduction in scores was performed by expressing the severity of depression scores as difference scores: mean depression at baseline (pretreatment) -last session (post-treatment) in each group separately (M S in sham or M A in active groups).
The total sample size N of each group (N S or N A ) was either the sample size at baseline or the mean sample size at baseline and last session if any patients dropped out of the study.
The SD of the mean difference scores was computed for each group separately (SD S or SD A ) as follows (Borenstein et al. 2009 p. 234): The correlation coefficient r=.5, between the pre-and the post-treatment depression scores, was chosen as the most optimal coefficient that neither overestimates the SD (r=.0) nor underestimates the SD (r=1.0) (Borenstein et al. 2009 p. 237, Table 24.7). The value of .5 was also close to the mean of correlations between pre-and post-treatment scores conducted in studies that reported the scores for all individual patients (these studies are listed in the table below). Specifically, the correlation coefficients were r=.37 (all patients), r=.38 (sham), and r=.49 (active group). The M and SD of the difference scores and the total N per group (sham and active) were used to compute the effect sizes.

Effect size computation (standardised mean difference)-Cohen's d and Hedges' g
The standardised mean difference, Cohen's d, was computed for the sham and active rTMS groups in each study as follows (Borenstein et al. 2009 p. 26): , where M S and M A refer to the mean severity of depression difference score (pre-post) in sham and active rTMS groups respectively, SD pooled is the SD of the severity of depression difference score (pre-post) pooled for the two groups using the standard deviations of sham (SD S ) and active rTMS (SD A ) groups, and N S and N A are the sizes of sham and active groups respectively. The variance of d (V d ) was computed as follows (Borenstein et al. 2009 p. 27): In addition to d, the standardised mean difference, Hedges' g (that is an unbiased version of d in small-N studies) and its variance (V g ) were computed for the sham and active stimulation groups in each study as follows (Borenstein et al. 2009 p. 27): , where J is the correction factor.

Combining multiple outcomes within studies
Some studies in the current analysis utilised multiple scales to measure depression severity (HAMD, BDI, and MADRS). In such cases, the effect sizes d and their variance (V d ) were computed separately for each scale. Subsequently, one mean effect size d was computed/study using an arithmetic mean. The variance of such a mean effect size (V dmean ) was computed according to the following formula for combining multiple outcomes within the same studies (Borenstein et al. 2009 p. 227): , where r=1.0 (correlation coefficient between outcomes in the same cases). Subsequent meta-analysis was computed on such a mean effect size of multiple outcomes/study and its variance.

Meta-analysis: random-effects model with inverse-variance weights
The weight in each study (W d ) was computed according to the random-effects model as follows (Borenstein et al. 2009 p. 73): , where V d is the within-study variance (variance of d) and T 2 is the between-study variance which was computed according to the method of moments (or the DerSimonian and Laird method (DerSimonian and Laird 1986)) and using df=k-1 (k=number of studies) as follows (Borenstein et al. 2009 p. 73-74): The overall mean weighted effect size (M d ) and its variance (V Md ) were computed for subgroups of studies as follows (Borenstein et al. 2009 p. 73-74): The lower and upper 95% confidence intervals of M d (LCI Md and UCI Md ) were computed as follows (Borenstein et al. 2009 p. 73-74): Finally, the z-score for M d was computed, to test the null-hypothesis that M d =0 meaning that rTMS is not effective at reducing depression scores compared to sham, according to the following formula (Borenstein et al. 2009 p. 74):

Computation of R 2 in meta-regression
Univariate linear meta-regressions were computed using the random-effects model to find out if the weighted effect sizes (outcome) could be predicted using the various study characteristics (clinical, demographic, and the rTMS parameters) in the current meta-analysis. The slope of the straight line (the line of best fit), B*, was tested for statistical significance according to the following formula (Borenstein et al. 2009 p. 197): The null-hypothesis tested was that B* is not different from zero (meaning that the predictor does not predict the outcome). Since univariate regressions were conducted (using one predictor only), it was assumed that the statistical significance of the slope of the regression line was equivalent to the statistical significance of the regression model.
The practical significance of the statistically significant regression models was tested using the equivalent of the R 2 index in linear regression. The formula for R 2 in meta-regression takes into account the between-study variance in the weighted d unexplained by the regression model containing the predictor (T 2 model shown as 'Tau-squared' in the output of the meta-regression module in CMA) and the total withinand between-study variance among the weighted d (T 2 total that is computed together with other heterogeneity statistics in the standard random-effects model of all studies involved in the meta-regression) as follows (Borenstein et al. 2009 p. 202): The R 2 in meta-regression shows the proportion of the between-study variance in weighted d explained by the predictor.  Notes: There were 55 sources in total because data from two abstracts excluded from the current analysis ( (Haag et al. 1997) and (Avery et al. 2000)) were later included in published articles included in the current analysis ( (Padberg et al. 1999) and (Holtzheimer et al. 2004)).