AI Reveals Two Distinct Brain Subtypes Behind Suicide Risk in Bipolar Depression
Researchers used resting-state brain scans and genetic data from 802 people and a generative AI model to sort suicidal thinking in bipolar depression into two reproducible brain-based subtypes. One was driven by hyperconnectivity involving the visual cortex and linked to greater anxiety and weaker cognitive flexibility; the other involved excessive coupling between the default mode and central executive networks and was linked to brooding rumination. Both patterns replicated in two independent cohorts, and the signal tracked fluctuations in individual suicide risk over time. The genetic and treatment-response signals (including an exploratory link to TMS response) are preliminary and need prospective testing before anything can change in care.
Source: Bioengineer (reporting on Wang et al., BMC Medicine, 2026)
Peer-Supported Online CBT Eased Anxiety After Unexplained Chest Pain
In the PACER randomized trial — 375 emergency department patients with low-risk chest pain and moderate-to-severe anxiety — peer-supported online cognitive behavioral therapy reduced anxiety more than an enhanced primary-care referral over 12 months of follow-up. Therapist-led telehealth CBT also improved more than the referral, though the primary-outcome difference was not statistically significant. The peer-supported approach helped most in patients who entered with severe anxiety. The study had no usual-care group, so some improvement could partly reflect time or regression to the mean.
Source: News-Medical.net (reporting on Musey et al., JAMA Internal Medicine, 2026)
An AI Chatbot Outperformed Group Therapy for Student Anxiety — But Context Matters
In a randomized trial of 977 Israeli university students, those using the "Kai" therapy chatbot scored about 2.1 points lower on a 21-point anxiety scale after 12 weeks than students in weekly in-person group therapy led by licensed psychologists. Much of the gap came from anxiety creeping up in the group-therapy and waitlist arms rather than from large improvements in the chatbot arm, and average effects were small. Caveats: outcomes were all self-reported, no one could be blinded, and the lead author consults for the chatbot's maker and holds stock options.
Source: MentalHealthDaily (reporting on Shoshani et al., npj Digital Medicine, 2026)