Prolonged P3 latency predicts clinical response to repetitive transcranial magnetic stimulation in tinnitus

The clinical efficacy of repetitive transcranial magnetic stimulation (rTMS) for tinnitus is variable, necessitating predictive biomarkers. We assessed whether pre-treatment multimodal electroencephalography (EEG) could predict rTMS response.

Fifty-eight tinnitus patients underwent a 10-session rTMS protocol targeting the dorsolateral prefrontal cortex and left temporo-parietal lobe. Before treatment, we extracted a comprehensive set of 325 features, encompassing event-related potentials (ERPs), spectral power, microstate metrics, and clinical features. Predictive modeling was performed using five machine learning classifiers with 5-fold cross-validation. Feature importance was ranked, and an iterative feature selection procedure was conducted to optimize the feature set. The predictive power of the top-ranking feature was further validated by constructing a univariate model.

Adaptive Boosting performed best. P3 latency was the top predictor. Responders had longer pre-treatment P3 latencies than non-responders (389 ms vs. 370 ms, p < 0.001). The univariate prediction model based on a P3 latency cut-off of ≥384 ms achieved an accuracy of 0.79 and an area under the curve (AUC) of 0.69 ± 0.14. The multimodal model constructed through feature optimization, which incorporated the top 15 features, demonstrated superior predictive efficacy, achieving an accuracy of 0.84 ± 0.10 and an AUC of 0.91 ± 0.09.

Pre-treatment P3 latency robustly predicts clinical response to rTMS in tinnitus patients.

This study establishes P3 latency as a foundational, clinically actionable neurophysiological biomarker for stratifying rTMS response, facilitating a shift toward precision neuromodulation by optimizing patient selection and avoiding unnecessary interventions.