The present systematic review and meta-analysis was conducted with the primary aims of quantifying efficacy of TMS in the treatment of adolescent depression and exploring respective patient- and trial-level moderators while considering the data currently available on a study (aggregated data) and patient-level (IPD), respectively. As a secondary aim, we also synthesized available data on rates of treatment response (aggregated data), and explored respective moderators (IPD).
First, meta-analysis of aggregated data suggested TMS to significantly reduce depression severity in adolescent patients, which was further in line with significant response rates. However, analyses of aggregated data also suggested potential biases, such as small-study effects. Indeed, the studies included were characterized by small sample sizes and large standard errors, respectively. Of note, all studies reported a statistically significant mean change in depression severity under active rTMS treatment from baseline to post-treatment, providing strong indications for potential publication bias—which was also suggested by visual inspection of funnel plots and formal testing of funnel plot asymmetry, respectively. Collectively, the present studies were also very heterogeneous with respect to dosing protocols for TMS. Of note, only two out of 10 studies included in the present synthesis applied a double-blind, randomized, and sham-controlled study design, and there have been considerable concerns about open-label trials to inherently inflate effect sizes and to be prone to several further biases, including regression to the mean, investigator biases, and, critically, confounding of active treatment with placebo effects [
37,
83]. Sham stimulation in rTMS trials can be considered in analogy to pill placebo in pharmacological trials [
37], and respective trials suggest larger response rates to pill placebo in adolescent as compared to adult depression [
84,
85]—with reported rates of placebo response in adolescents ranging between 22 and 59% [
86‐
93]. Considering rTMS treatment, there is only one large-scale randomized controlled study currently available that would inform on the response rate to sham stimulation in adolescent depression, suggesting a sham response rate of 36.4% [
54]—which falls well within the range of response rates reported for pill placebo. As a comparison, meta-analytic studies suggest response rates to sham rTMS in adult patients with depression to be at around 10 or 11% [
25,
94].
Based on IPD from five individual study samples, accounting for repeated measures within individuals clustered within trials, and considering three different depression scales (i.e., HDRS, CDRS-R, and BDI-II), we found treatment efficacy as well as response to be associated with certain patient and trial-level characteristics. Most consistently observed was the influence of patients’ age, with younger individuals exhibiting a higher reduction in depression scores as well as a higher likelihood of treatment response compared to individuals of older age. This finding somewhat aligns with existing meta-analytic evidence on TMS in adult patients suggesting young age to present a good prognostic factor [
39], as well as with findings from an evidence synthesis of several short-term randomized controlled trials of antidepressants, reporting higher placebo response rates in younger as compared to older adolescents (after the exclusion of one large fluoxetine trial) [
93]. Of note, presently, most studies included samples of older-aged adolescents (mean age of 17.45 ± 1.99 years), with only one study including relatively younger adolescents between 12 and 14 years. However, based on the present findings as well as growing evidence which points towards a favorable safety profile of TMS for the treatment of adolescent depression [
52], future studies should also consider younger individuals with MDD. Furthermore, although only observed when considering HDRS but not CDRS-R or BDI-II scores, depression severity was a significant moderator of pre- to post-treatment change after TMS. This finding also aligns with a considerable amount of evidence in adults, suggesting TMS to be particularly valuable in severe cases of adult MDD, and in patients with TRD [
25,
27,
95]. TMS applied to the DLPFC in adolescent MDD might reverse some of the aberrant functional connectivity between prefrontal and subcortical regions, which, during the period of adolescence that is characterized by peak PFC plasticity, might result in long-term clinical improvement, especially in severe cases of depression [
37]. Considering the present variability of results based on the outcome measure considered, several potential explanations for this finding exist. First, the present findings based on IPD analyses in general might simply present spurious associations that might have occurred based on a multiple testing situation, and future studies with pre-planned hypotheses testing will be needed to confirm the observed associations. Furthermore, there might be poor concordance between different instruments, i.e., clinician-rated (such as the HDRS or MADRS) and patient-reported (BDI-II) outcome measures of symptoms of depression [
96]. Furthermore, specifically considering the CDRS-R, which currently presents the most commonly used scale in adolescent depression research, existing studies on the psychometric properties of this instrument when used in adolescents (but which was originally developed for use in children aged 6–12 years) are of low methodological quality, and thus it currently remains unclear whether the CDRS-R appropriately measures depressive symptom severity in adolescent MDD [
97]. Finally, differences in IPD analysis results considering different measures might also go back to underlying sample characteristics, which presently has not been explored further but should be considered in future antidepressant trial. Besides patient factors, several trial-specific, TMS-related factors were identified to significantly moderate treatment efficacy as well as the likelihood of treatment response. These included the laterality of stimulation, the specific TMS modality applied, treatment duration, and the number of stimulation sessions applied, respectively. Concerning the latter, greater efficacy of TMS has also been previously reported for protocols applying a higher number of stimulation sessions, as well as a greater number of pulses per session [
98‐
100]. In the five studies included in IPD meta-analysis, the number of sessions ranged from 10 to 30 sessions, with a treatment duration of 14–42 days. It has been previously suggested that increasing the number of sessions per day from one to multiple might increase efficiency [
37]. Yet, as presently observed, a longer treatment duration might also increase efficacy (although potentially confounded by the number of sessions applied). Consequently, future studies should investigate the relative importance of the number of TMS sessions and overall treatment duration considering treatment efficacy to determine a session/duration ratio that optimizes stimulation protocols for both efficacy and efficiency. Concerning laterality of stimulation, the present finding of greater efficacy observed with unilateral compared to bilateral stimulation is in line with results from a double-blind, randomized, and sham-controlled study in adults with TRD [
101]. In the respective study, only unilateral stimulation was significantly more effective compared to sham stimulation at treatment termination and was correlated with a higher percentage of patients who showed remission. Further, unilateral but not bilateral TMS showed higher antidepressant efficacy compared to sham stimulation in the respective study. However, these and the present findings are somewhat contradictory to existing meta-analytic evidence, suggesting that bilateral stimulation might not be statistically significantly different from unilateral stimulation in adult MDD [
102‐
104]. Given potential neurodevelopmentally driven differences between the pathophysiology of MDD in adolescents compared to adults, it is important that future studies further investigate whether unilateral compared to bilateral stimulation might be differentially effective in adolescent MDD. Similar, standard rTMS compared to TBS was associated with greater treatment response. Of note, only one of the included studies used TBS. Primary studies addressing the comparative efficacy of rTMS versus TBS are warranted.
The present results must be interpreted within the constraints of considerable limitations inherent to this evidence synthesis. First, given the current lack of randomized controlled trials considering TMS treatment in minors with depression, we failed to conduct a synthesis of results from rigorous randomized controlled trials comparing active rTMS treatment with sham stimulation. Instead, we conducted a quantitative synthesis of results from mainly uncontrolled studies. While non-randomized studies are increasingly recognized as a potential source of insights into real-world performance of novel therapeutics, and thus are certainly of value for healthcare decision making especially in the case of novel and innovative treatments [
105], high-quality, randomized controlled trials unequivocally provide the most reliable evidence on the relative efficacy and safety of medical interventions. The present results must therefore be considered highly cautiously. As a further critical limitation, individual-patient data were retrieved only for half (i.e., five out of ten) of the studies that were included in the present meta-analysis, and most critically, we failed to include IPD data from randomized controlled trials—which certainly would have significantly improved the quality of the data included and allowed us to also consider effects of sham stimulation in IPD meta-analysis. Barriers to obtaining IPD on the one hand were encountered due to a lack of responsiveness of the corresponding authors even to repeated data requests. Furthermore, barriers were also encountered in the form of data sharing policies of study sponsors from the industry. Either problem is strongly impeding on the quality of meta-science, and endeavors to further ameliorate and facilitate practices of data sharing are warranted (for a scoping review and practical guide on the matter, we refer the reader to, e.g., [
106]).