A Machine Learning Using NLP to Predict Long-Term Patient Response

A single high dose of the psychedelic compound COMP360 (synthetic psilocybin) is safe and effective for some people with treatment-resistant depression, but only after a specific point in the trial. Predicting who will benefit is therefore important. Researchers analyzed transcripts of therapist-patient conversations recorded one day after dosing, using a machine-learning algorithm to measure two dimensions of emotion (arousal and valence) in both speakers. The approach demonstrates a potential method for forecasting treatment outcomes.

Study at a glance

Abstract

Historical stories and contemporary clinical studies have shown substantial promise for the therapeutic use of psychedelics for the treatment of mood disorders such as depression. Comp360 is COMPASS Pathways' unique synthetic formulation of psilocybin. A recent randomised double-blind phase-II b research showed that it was safe and effective for those with treatment-resistant depression. The medication is effective for some patients, but only after a certain point in the trial, therefore accurate outcome prediction is crucial for determining which patients will need to switch treatments. The audio recordings of the participant and therapist's psychological support session were used to create transcripts. This session took place one day after the COMP360 was administered. The therapist and participant's two-dimensional sentiment (arousal and valance) were computed from the transcript using a zero-shot algorithm such as machine learning classifiers that relied on the BART big language method.