Generative Chatbot Eases Depression and Anxiety in Clinical Trials

A team led by Nicholas Jacobson at Dartmouth College has revealed promising results from clinical trials of a generative chatbot designed to tackle depression, anxiety, and eating disorders. The study, detailed in NEJM AI, involved 210 patients grappling with significant symptoms of major depressive disorder, generalized anxiety disorder, or disordered eating behaviors. Half of them—106 participants—engaged with Therabot, a chatbot fine-tuned with expert input, over four weeks, while the rest were placed on a waitlist, gaining access only after the trial concluded. Progress was measured at weeks four and eight, with symptom changes tracked using sophisticated mixed-effects models.

Those interacting with Therabot saw a striking drop in symptoms compared to the control group. For major depressive disorder, scores fell by an average of 6.13 points by week four (versus 2.63 for the waitlist) and 7.93 by week eight (versus 4.22). Generalized anxiety disorder improved by 2.32 points at week four (against a mere 0.13) and 3.18 by week eight (versus 1.11). Eating disorder symptoms plummeted by 9.83 points at week four (compared to 1.66) and 10.23 by week eight (versus 3.70). Participants logged over six hours on average with the bot, rating its therapeutic bond as comparable to that of a human therapist.