EEG microstates as a suicidality biomarker in OCD — a Beijing study points at the salience network

OCD has one of the highest hidden suicide rates of any anxiety-spectrum disorder, but most clinicians screen for it the way they screen for depression: ask the questions, watch the eyes, document. Self-report misses a lot. This Tsinghua/CAS Beijing group asked a different question — does resting-state EEG, decomposed into microstates, separate OCD patients with active suicidality from those without?

EEG microstates are short (60–120 ms) quasi-stable scalp topographies that recur across the resting brain. Four canonical microstates — A, B, C, D — map fairly reliably onto large-scale resting networks: auditory, visual, salience, and attention/executive. So a "microstate signature" is a low-cost, network-level fingerprint of how a brain idles.

What the data shows

Both OCD groups differed from controls on multiple temporal parameters and transition probabilities — the disorder leaves a signature. Inside the disorder, the discriminator was different: STB patients had a shorter microstate B duration than non-STB patients, plus elevated microstate C occurrence relative to HC. Read clinically, the salience network is being recruited more often, the visual-network state is held less time, and the brain at rest is switching between configurations differently. This is not a brand-new theory — salience-network hyperactivity has been a recurring finding in suicidal ideation across diagnoses — but it is one of the first demonstrations that the same fingerprint shows up in OCD specifically, and that it is detectable with off-the-shelf 64-channel EEG, not 7T MRI.

The ML classifier numbers tell the rest of the story. 96–97% balanced accuracy against healthy controls is unsurprising — OCD itself is detectable. The 65.88% accuracy distinguishing STB from non-STB is the meaningful number: low enough to remind us this is not a screening test, high enough to suggest the signal is real and could become an adjunct.

For your practice

You will not be running EEG microstate analysis in your office next week, and that is fine — that is not the point. Two things matter for clinical work today.

First: when you assess suicide risk in an OCD patient, anchor on what the network model predicts. Salience-network upregulation maps onto subjective experience of intrusive distress that "won't go away" — the patient feels persistently flagged-as-threatened by their own thoughts. Asking "how loudly are the thoughts demanding your attention right now, even when you try to think about something else?" probes the construct better than asking whether the patient feels hopeless. OCD-related suicidality often runs through exhaustion, not classical depression.

Second: keep an eye on the next two years of EEG-microstate work. Several Asian and European labs are pushing toward clinically deployable resting-state EEG protocols — 5–10 minutes, dry electrodes, automated pipeline. If those mature, an OCD-suicidality adjunct biomarker could land in tertiary clinics within five years, well ahead of fMRI biomarkers that still cost €600 a head. The clinical question for our generation of therapists is not whether to trust the machine, but how to integrate a probability score with clinical interview without the score quietly taking over.