Use of Generic Large Language Model Chatbots for Mental Health Issues Among Patients Seeking Psychotherapy: Cross-Sectional Mixed Methods Analysis in a German University Hospital Outpatient Clinic.

BACKGROUND

Large language model-based chatbots (LLM-CBs) are increasingly used as mental health support tools. Risks and harms are discussed especially for unsupervised use and for vulnerable groups. A naturalistic characterization of the use of LLM-CBs by patients with mental health disorders is currently lacking.

OBJECTIVE

The goal of this cross-sectional study is to investigate demographic and clinical characteristics of users of LLM-CB for mental health and their associations with use behavior.

METHODS

Across 6 months, qualitative and quantitative data on LLM-CB use for mental health were collected in a consecutive sample of psychotherapy outpatients in routine treatment with at least 1 diagnosed mental health disorder. Results are reported descriptively and supported by exploratory inferential analyses.

RESULTS

Within a total of 812 included outpatients who reported on LLM-CB use for mental health, 245 (30.1%) reported having used LLM-CBs for mental health (once: 69/217, 31.8%; weekly: 118/217, 54.4%; and daily: 30/217, 13.8%). Users were more likely to be younger, female, better educated, students, or unemployed. In total, 61.4% (121/197) have used LLM-CBs in self-reported personal crisis situations, of whom, 67.7% (82/121) found the experience unhelpful. LLM-CB use showed a weak, imprecise association with the number of diagnoses in the included diagnosis groups and was higher for anxiety and eating disorder diagnoses and lower for somatoform disorders. Post hoc multivariable analyses suggest that higher use in eating disorders and lower use in somatoform disorders were attenuated after adjusting for age, while higher use for anxiety disorders persisted. LLM-CBs were generally evaluated positively across various dimensions, especially when self-reported personal crisis situation use was experienced as helpful. Qualitative analysis of an open-ended question on use experience shows LLM-CB use across 4 domains reported in the selected subsample: general psychological assistance, disorder-specific assistance, information assistance, and everyday assistance.

CONCLUSIONS

This naturalistic characterization of LLM-CB use in mental health outpatients implies that harms and benefits of LLM-CB use are associated with demographic and clinical profiles. The results suggest that asking about LLM-CB use may be a useful consideration in psychosocial assessment, particularly regarding crisis use, symptom-reinforcing use, and replacement of human support.