The dynamics of the EEG response to TMS perturbation differentiate individuals eventually found to be ‘resilient’ from those ‘vulnerable’ to the mental health impact of the COVID-19 pandemic
Assessments of mental health using the four-item patient health questionnaire (PHQ-4), an ultra-brief depression and anxiety screening self-report questionnaire, were obtained prior to the COVID-19 pandemic and up to three additional times during the pandemic. If during all timepoints across the confinement, the PHQ-4 score was lower or equal than before the pandemic outbreak, subjects were classified as resilient (n=32). Conversely, if a given subject had a higher score at any time point during the pandemic, they were classified as vulnerable (n=32). Because not all participants completed stimulation at both target locations, the subgroups used in this analysis were smaller for each target (for L-DLPFC, 23 resilient and 25 vulnerable; for L-IPL, 22 resilient and 23 vulnerable). To make sure that the mental health impact of the pandemic was related to the levels of stress perceived during the outbreak, we correlated the average score of the three pandemic PHQ-4 timepoints, with the scores of the 14-item perceived stress scale 36, which was also completed by participants during the pandemic, and found a strong positive correlation (Rs=.69; p<.001), indicating that subjects experiencing more stress during the pandemic also had a larger mental health impact. Overall, participants had a low to moderate level of perceived stress during the pandemic (Mdn=14; range from 2 to 32).
Point-by-point non-parametric permutation testing (1000 permutations) with cluster correction for multiple comparisons 37 on the TMS-EEG evoked time-series, revealed a single broad cluster (i.e., 202-269 ms post-stimulus) surviving correction for multiple comparisons, only during stimulation of the left dorsolateral prefrontal cortex (L-DLPFC; see Figure 2, A). Inspection of the topographical distributions in source space for the surviving cluster, confirms that vulnerable individuals had a qualitatively stronger frontal activation than resilient ones (see Figure 2, B). There were no significant clusters revealed after analysis of the responses to the left inferior parietal lobule (L-IPL) control target (see Figure 2, C and D), nor for the distributed responses to stimulation on either target (Figure S1).
Local EEG response to TMS perturbation of the left dorsolateral prefrontal cortex predicts mental health during the pandemic’s lockdown confinement
Without classifying participants into resilient or vulnerable, a multiple linear regression model was fit to determine the potential of TMS evoked EEG perturbation of the L-DLPFC to predict mental health outcomes after the COVID-19 pandemic outbreak and the strict lock-down confinement imposed to curb community transmission of the virus. The model’s response variable was the mean of the total scores for the three PHQ-4 questionnaires, which were completed by participants during the lock-down confinement. Candidate predictors were the local and global brain EEG reactivity to the TMS pulse — recorded before the pandemic outbreak —, as well as their interaction with the stimulation target definition method (i.e., functional or anatomical). Additionally, we included age, gender, and years of formal education as predictors, because these are demographic and individual factors partially predictive of resilience to stress 17. Finally, we included as a predictor the number of months before the pandemic since each subject underwent TMS-EEG. This was included to control for the possibility that the amount of time passed from stimulation to pandemic would have an impact in the prediction.
The full linear regression model for the L-DLPFC stimulation target significantly predicted mental health during the pandemic (F(8,47)=3.1, p=.006, R2ad j=.243), and revealed as significant predictors the local brain reactivity to TMS (t=2.31, p=.025) and years of formal education (t=-2.98, p=.005). See supplementary table S1 (model “Full DLPFC”) for detailed results. Both predictors were independent from each other, as reveled by the lack of correlation between them (R2=.187, p=.167). Subsequently, we tested a reduced model (F(2,53)=10.5, p<.001, R2adj=.257) retaining only as predictors local brain reactivity (t=2.33, p<.024) and education (t=-2.86, p=.006). See supplementary table S1 (model “Reduced DLPFC”) for detailed results. Likelihood ratio test comparing the two models showed that the full model did not provide a better fit than the reduced one (χ2(6)=4.98, p=.546). The lower Akaik and Bayesian information criteria (AIC and BIC, respectively) values for the reduced model further suggest a better and more parsimonious fit (AICfull=79.83, AICreduced=72.82; BICfull=98.06, BICreduced=78.89). Fig. 3 illustrates the linear relationship between the significant predictors and the response variable in the reduced model. Analysis of variance of the reduced model revealed that local L-DLPFC reactivity explained 12.79% of total variance in mental health during the pandemic, while education explained 15.64%.
To assess the brain specificity of our findings, we fitted a model replacing the predictors for local and global EEG reactivity with those measured when stimulating the L-IPL. The regression model for this control stimulation target did not significantly predict mental health during the pandemic (F(8,46)=0.4, p=.915, R2adj=-.097). See supplementary table S1 (model “Full IPL”) for detailed results.
Finally, to demonstrate the specificity of the stimulation itself, a model was fitted where we added the local baseline pre-TMS activity as an additional predictor to the reduced L-DLPFC model. The resulting model, while still significant (F(3,47)=7.09, p<.001, R2adj=.25), revealed that baseline pre-TMS EEG activity did not significantly contribute to predict mental health during the pandemic (t=.67, p=.507). See supplementary table S1 (model “Reduced+Baseline”) for detailed results.