Opparounds

Open Dialogue versus treatment as usual for adults presenting in crisis to mental health services in England (the ODDESSI Trial): a multisite cluster-randomised trial

Pilling S, et al. · The Lancet Psychiatry · August 2026

Open Dialogue is a network-based crisis model that brings the patient, clinicians, and their social network into the same room to plan and review care together, rather than routing decisions through the clinician alone. This is the first randomised trial of the model: 494 adults presenting in mental health crisis across NHS trusts in England and the South were randomised to Open Dialogue or treatment as usual and followed for two years. On the primary outcome, time to first relapse after initial recovery, Open Dialogue was no better than usual care — the headline result is a null one. But patients who received Open Dialogue had more than three times the odds of never being admitted to a psychiatric inpatient bed, lower odds of being re-referred to crisis services, and better self-rated recovery and quality of life.

Read together, the trial separates two things that are usually assumed to move as one: staying well, and how the system responds when you're not.

Read the paper →  doi:10.1016/S2215-0366(26)00229-4

Prediction in routine care: a naturalistic comparison of clinicians and a machine learning model for patients receiving treatment for generalised anxiety disorder

Delamain H, Buckman JEJ, et al. · Psychotherapy Research · August 2026

Before treatment began, clinicians at two UK routine psychological therapy services predicted post-treatment GAD-7 scores and outcome category (recovery, reliable improvement, deterioration) for 100 patients with generalised anxiety disorder, rating their own confidence and explaining their reasoning. Those predictions were then compared against a previously validated machine-learning model trained on national Talking Therapies data. Among the 66 patients with complete outcomes, the model was numerically more accurate than clinicians at predicting final severity, reliable improvement, and recovery — though the gap didn't reach statistical significance — and both clinicians and model were poor at predicting who would get worse. Clinicians' average confidence in their own predictions was 67%, and that confidence had no relationship to whether they were actually right. Clinicians leaned on context the model never saw — motivation, engagement, life stress — while the model weighted variables, like the wait between referral and first assessment, that not one clinician named as important.

Read the paper →  doi:10.1080/10503307.2026.2715402

Linezolid is an MAOI you'll forget to ask about

Linezolid, the oxazolidinone reached for when MRSA or VRE won't respond to anything else, is also a weak, reversible, nonselective MAO inhibitor — a mechanism buried in the antimicrobial label that most non-psychiatrists never register. Serotonin syndrome after starting linezolid on a patient already taking an SSRI, SNRI, or other serotonergic agent doesn't necessarily show up on day one: case reports put onset anywhere from within 24 hours to over two weeks into the course, so tolerating the first few doses is not reassurance.

Where the infection allows it, the standard is to hold serotonergic antidepressants for two weeks before starting linezolid, or switch to a non-serotonergic regimen; when linezolid genuinely can't wait, the plan is close monitoring, not stopping and restarting the antidepressant on your own timeline. This is the interaction that gets missed in consult work precisely because the primary team is thinking about vancomycin failure, not the sertraline in the chart — by the time altered mental status and clonus appear, several doses are already in. Screen the med list yourself the moment linezolid gets ordered; don't wait for ID to ask.

Habituation was never the mechanism — expectancy violation is

Exposure therapy was taught for a generation as a habituation exercise: stay in contact with the feared stimulus until anxiety comes down, repeat until the fear response extinguishes. The inhibitory learning model that has largely replaced this in the exposure literature reframes the target entirely — durable improvement doesn't track within-session anxiety reduction at all. It tracks how thoroughly the exposure violates the patient's specific feared prediction.

That reframe has a direct technical consequence. Safety behaviors — the hand sanitizer kept in a pocket, the reassurance glance at a companion, the mental review after leaving the stove — don't just soften an exposure's intensity. They're what lets the old prediction survive contact with disconfirming evidence unscathed. The move isn't longer exposures; it's removing the behavior that lets the feared belief go untested.