Machine learning-based identification of a psychotherapy-predictive electroencephalographic signature in PTSD

Although psychotherapy is at present the most effective treatment for posttraumatic stress disorder (PTSD), its efficacy is still limited for many patients, due mainly to the substantial clinical and neurobiological heterogeneity in the disease. Development of treatment-predictive algorithms by leveraging machine learning on brain connectivity data can advance our understanding of the neurobiological mechanisms underlying the disease and its treatment. Doing so with low-cost and easy-to-gather electroencephalogram (EEG) data may furthermore facilitate clinical translation of such algorithms for patients with PTSD. This study investigates whether individual patient-level resting-state EEG connectivity can predict psychotherapy outcomes in PTSD. We developed a treatment-predictive EEG signature using machine learning applied to high-density resting-state EEG collected from military veterans with PTSD. The predictive signature was dominated by theta frequency EEG connectivity differences and was able to generalize across two types of psychotherapy—prolonged exposure and cognitive processing therapy. Our results also advance a biological definition of a PTSD patient subgroup who is resistant to psychotherapy, which is currently the most evidence-based treatment for the condition. The findings support a path towards clinically translatable and scalable biomarkers that could be used to tailor interventions for each individual or drive the development of novel treatments (ClinicalTrials.gov registration: NCT03343028 ). Using machine learning, Zhang et al. identify EEG signature to predict psychotherapy outcomes in PTSD, paving the way towards the development of scalable biomarkers.