This analysis demonstrates a pooled ‘large-scale analysis’ using Big NIBS data. Findings provide proof of principle that pooled analyses can identify patterns in data and sources of variability that individual, single-site studies cannot.

Information

PTSD data were sourced from five previously published clinical trials on the therapeutic effect of TMS on PTSD symptoms. Below we have collated and analysed all of these individual participant data across the studies, providing an average overall effect of the TMS on PTSD symptoms, and demonstrating the factors that account for variability in individual patient response. This is a small number of studies so results should be treated with caution.

Methods

Data were analysed as per the method described in detail in Corp et al. (2020, 2021). Briefly, mixed-effects linear regression analysed IVs that could explain the DV of symptom improvement % after a clinical rTMS intervention (higher % = symptom improvement). Mixed-model regression preserves the nesting of participants within studies, given that it is inappropriate to analyse individual-participant data as if they all came from a single study.

Studies analysed

Table 1. Data included from studies sharing PTSD data. F = females.

 

 

 

 

 

 

Average Effect of rTMS on PTSD Symptoms

First column shows mean and 95% CIs from all participants who received active TMS for PTSD symptoms. Second and third columns show direct statistical comparison between studies that included both active and sham conditions. Brackets show (studies/participants). * = p < 0.05. Both active samples showed significant symptom improvement compared to zero, yet sham did not (p = 0.129).

 

 

 

 

 

 

 

 

 

 

 

 

Distribution plots. Symptom improvement scores from shared individual participant data. Active TMS N = 112; sham TMS N = 20. Symptom improvement % calculated as: (Baseline score – Follow up score) / (Baseline score)*100.

 

 

 

 

 

 

 

 

Active TMS Regression Model - Factors That Matter

B-values show the difference in symptom improvement % between IV levels after adjusting for all other Is in the model. E.g., on average, unblinded sessions showed 16.73% higher symptom improvement compared to blinded sessions. See Figure 2 for all comparisons between levels. Bold denotes significance (p < 0.05). Studies = 5; participants = 112.

 

 

 

 

 

 

Factors That Matter - Marginal Means

Data correspond to above regression table, showing differences between levels of the IVs included in the regression model. rTMS frequency was analysed post-hoc rather than in the Table 2 model due to low N. Brackets show (studies/participants). Error bars show 95% confidence intervals. * = p < 0.05. All samples showed significant symptom improvement compared to zero.

 

 

 

 

 

 

 

 

 

 

 

 

 

Other IVs Not Included in the Model