CORR Insights®: Is Prior Nonoperative or Operative Treatment of Dysplasia of the Hip Associated With Poorer Results of Periacetabular Osteotomy?

Non-invasive brain stimulation (NIBS) techniques are a collection of methods that can modulate neuronal activity and are increasingly used at therapeutic interventions for psychiatric disorders (). Among those techniques, repetitive Transcranial Magentic Stimulation (rTMS) is currently approved for treatment-resistant Major Depressive Disorder (MDD); however, only 30–50% of MDD patients achieve remission (). This heterogeneity underscores the need for predictive biomarkers to optimize resource allocation and reduce patient exposure to ineffective treatments. The article by Zhao et al. published in the current issue of Biological Psychiatry CNNI began addressing this gap by proposing an explainable ML framework that combined EEG-derived phase locking value (PLV) and clinical features and achieved a remarkably high classification accuracy (i.e., 97.33%) (). In particular, the inclusion of PLV, a robust electroencephalogram (EEG) parameter reflecting neural synchrony, enhanced the model’s ability to differentiate between responders and non-responders while the study’s reliance on multiple machine learning (ML) models, including Support Vector Machine (SVM), further reinforced the credibility of its findings. By achieving consistently high accuracy with different ML model, this study shows that precision psychiatry, where biomarkers can inform individualized interventions, is a reachable objective. It also represents a significant step toward personalized treatment for MDD patients by integrating several EEG metrics and ML methods to predict response to rTMS. Prior studies relied on single-modality predictors—either EEG clinical variables—limiting their predictive power. For example, Bares et al. achieved 76% accuracy using prefrontal theta cordance (), while Nobakhsh et al. reported 89.6% accuracy with direct directed transfer function, an effective brain connectivity measure computed by combining the delta and theta frequency bands (). Concerning predictive models based solely on clinical parameters, including age, treatment refractoriness, and baseline symptom severity (e.g., Hamilton Depression Rating scores) correlate moderately with outcomes, but appear to lack sufficient standalone accuracy. In contrast, in the study by Zhao et al. the integration of eight EEG metrics (e.g., PLV, PSD, entropy measures) and clinical variables (i.e., BDI, age, gender) leverages complementary information. PLV, which quantifies phase synchronization between brain regions, emerged as the most influential EEG biomarker, particularly in delta (F3-P7, F3-P4) and beta (P3-P8) bands and significantly outperformed previous studies that relied on either EEG or clinical features alone. This finding also indicates that aberrant frontoparietal connectivity in EEG delta band may predict antidepressant response to rTMS. SHAP (SHapley Additive exPlanations) is an interpretability framework for ML that explains individual predictions by attributing each feature a contribution value. By using SHAP, in this study Zhao et al. addresses a critical limitation of “black box” ML models that limit their interpretability. Specifically, by quantifying feature contributions, SHAP revealed that reduced delta-band PLV (F3-P7, F3-P4) was the key parameter associated with rTMS treatment response. Such interpretability enhances the plausibility of the reported findings and identifies testable mechanisms for future studies.