Large-scale analysis of interindividual variability in theta-burst stimulation data: Results from the 'Big TMS Data Collaboration'

Abstract

Background: Many studies have attempted to identify the sources of interindividual variability in response to theta-burst stimulation (TBS). However, these studies have been limited by small sample sizes, leading to conflicting results.

Objective/hypothesis: This study brought together over 60 TMS researchers to form the 'Big TMS Data Collaboration', and create the largest known sample of individual participant TBS data to date. The goal was to enable a more comprehensive evaluation of factors driving TBS response variability.

Methods: 118 corresponding authors of TMS studies were emailed and asked to provide deidentified individual TMS data. Mixed-effects regression investigated a range of individual and study level variables for their contribution to iTBS and cTBS response variability.

Results: 430 healthy participants' TBS data was pooled across 22 studies (mean age = 41.9; range = 17-82; females = 217). Baseline MEP amplitude, age, target muscle, and time of day significantly predicted iTBS-induced plasticity. Baseline MEP amplitude and timepoint after TBS significantly predicted cTBS-induced plasticity.

Conclusions: This is the largest known study of interindividual variability in TBS. Our findings indicate that a significant portion of variability can be attributed to the methods used to measure the modulatory effects of TBS. We provide specific methodological recommendations in order to control and mitigate these sources of variability.

Keywords: Big data; Theta-burst stimulation; Transcranial, and magnetic stimulation; Variability.

Figure 1.. PRISMA flowchart.

Adapted from Liberati et al. (2009). Note that some studies employed more than one TMS protocol, resulting in more datasets than studies. IPD = individual participant data.

Figure 2.. Distribution plots.

Histograms show distribution of normalised MEP data for iTBS and cTBS protocols. Y-axis shows the percentage of responses within each normalised MEP bin of 10 along the X-axis.

Figure 3.. Regression flowchart for iTBS.

Figure shows the method employed to arrive at the final regression model, demonstrating IVs accounting for interindividual variability in iTBS normalised MEP. IVs were omitted if they did not include at least three studies within each IV level. Stage 1 regressions analysed the variance in normalised MEP explained by each IV separately, while controlling for age and gender. IVs were dropped from the model if they did not obtain a p-value < 0.10. Step 2 regressions again iterated through IVs that were dropped in stage 1, to see whether these IVs now obtained a p-value < 0.10 controlling for IVs in the starting stage 2 model. Thus, the final regression model comprised of IVs that obtained a p-value < 0.10 in predicting TBS-induced plasticity in either stage 1 or 2 regressions. See Table 3 for results.

Figure 4.. Relationships between normalised MEP amplitude and continuous IVs in final iTBS regression model.

Our main analyses using linear regression showed that baseline MEP was a significant predictor of iTBS normalised MEP, while age did not reach significance (Table 3). However, post-hoc analyses demonstrated a significant cubic relationship for age, and a significant quadratic relationship for baseline MEP amplitude. These bivariate scatterplots are presented to give an indication of relationships only. See regression analyses for results controlling for all IVs in final model. Green line fits a smoothed ‘lowess’ curve through data (smoothing level = 0.8 - default).

Figure 5.. Marginal means for iTBS normalised MEP amplitude.

Marginal means provide an estimate of normalised MEP, adjusted for all variables in the final regression model. Orange bar shows the overall marginal mean for iTBS. Grey and white bars show marginal means for each level of the IVs TMS machine, target muscle, and TBS intensity, and neuronavigation (NN) use, which were included in the final model. * denotes a significant difference between levels (p < 0.05. All samples showed significant facilitation from 100 (all p < 0.001). Note that when excluding two MagPro machine studies that used biphasic pulses (for pre/post MEPs) there was significantly higher normalised MEP for MagPro machine (see Results). Error bars show 95% confidence intervals. Brackets show (studies/participants).

Figure 6.. Regression flowchart for cTBS.

Figure shows the method employed to arrive at a final model, demonstrating IVs accounting for interindividual variability in cTBS normalised MEP. See Methods and Figure 3 caption for further explanation of method.

Figure 7.. Relationships between normalised MEP amplitude and continuous IVs in final cTBS regression model.

Our main linear regression analyses showed that baseline MEP, but not age was a significant predictor of cTBS normalised MEP (Table 4). Post-hoc analyses also demonstrated significant cubic and quadratic relationships for baseline MEP amplitude. Green line fits a smoothed ‘lowess’ curve through data.

Figure 8.. Marginal means for cTBS normalised MEP amplitude.

Blue bar shows the overall marginal mean for cTBS. Grey and white bars show marginal means for each level of the IVs TS intensity, pulse waveform, and age group. These IVs were not included in the final regression model, but demonstrated moderate effect sizes between levels. * denotes a significant difference from 100 (suppression) (p < 0.05). Error bars show 95% confidence intervals. Brackets show (studies/participants). Slight discrepancy in age group N is due to a number of participants aged at the median value (30). These were placed into the ‘older’ group.

Figure 9.. TBS-induced plasticity at different timepoints.

iTBS normalised MEP amplitude was significantly different from 100 at each timepoint (* denotes p < 0.05) and there were no significant differences between timepoints. cTBS plasticity was only significant at 0–5 and 5–10 minute timepoints. 0–5 minute and 5–10 minute timepoints demonstrated significantly greater cTBS plasticity than all other timepoints, except for the 30–40 minute timepoint (p = 0.10 and p = 0.08, respectively). No other pairwise comparisons were significant for cTBS. Error bars show 95% confidence intervals. Brackets show (studies/participants).