Large language model-assisted support among caregivers of patients with epilepsy: associations with caregiver burden and psychosocial outcomes.

OBJECTIVE: To investigate the association between large language model (LLM)-assisted support and caregiver burden as well as psychosocial outcomes among family caregivers of patients with epilepsy. METHODS: This single-center, prospective cohort study enrolled primary caregivers of patients with epilepsy. Participants were classified into an LLM group and a control group based on whether they had used DeepSeek for epilepsy caregiving-related support. Caregiver burden was assessed using the Zarit Burden Interview (ZBI). Secondary outcomes included anxiety, depression, perceived stress, social support, quality of life, and caregiving self-efficacy, measured at baseline, 1 month, and 3 months. Adjusted generalized estimating equation (GEE) models were used to evaluate longitudinal associations between LLM use and the study outcomes, and dose-response analyses were performed within the LLM group. RESULTS: A total of 296 caregivers were included (149 in the control group and 147 in the LLM group). The two groups were comparable at baseline. At 3 months, the LLM group had lower caregiver burden (median ZBI: 23.00 vs. 25.00), anxiety, depression, and perceived stress scores, as well as a higher perceived social support score, than the control group. Adjusted GEE analyses revealed significant between-group differences in changes in caregiver burden and depression at both follow-up assessments and in anxiety, perceived stress, and psychological quality of life at 3 months, favoring the LLM group. Within the LLM group, no significant dose-response associations were observed between weekly usage time or the weekly number of effective conversations and caregiver burden at 3 months. CONCLUSIONS: Among caregivers of patients with epilepsy, use of LLM-assisted support was associated with lower caregiver burden and better psychosocial outcomes over 3 months. LLMs may serve as a low-threshold, scalable supportive tool in epilepsy family caregiving, though further studies are needed to evaluate long-term effectiveness and safety.