JMIR Publications

ABSTRACT

Background:

Large language model (LLM)–based conversational agents are increasingly used in healthcare to support multi-turn dialogue. In oncology, where patients and caregivers experience complex informational and emotional needs throughout the disease trajectory, conversational agents may support information provision, symptom consultation, and emotional assistance. However, research specifically examining multi-turn conversational agents designed for cancer patients and caregivers remains limited.

Objective:

This scoping review aimed to map the research landscape of LLM-based multi-turn conversational chatbots developed for cancer patients and caregivers, focusing on system design, intervention purposes, evaluation approaches, safety considerations, and transparency of LLM-related components.

Methods:

This scoping review followed the Joanna Briggs Institute methodology and PRISMA-ScR guidelines. Six databases (PubMed, Embase, Scopus, Web of Science, CINAHL, and PsycINFO) were searched for studies published between January 2022 and January 2026. Studies were included if they described LLM-based chatbots designed for cancer patients or caregivers that supported multi-turn conversational interaction. Two reviewers independently conducted the study selection and data extraction.

Results:

Eight studies met the inclusion criteria. Most studies focused on prototype development, with limited research evaluating clinical outcomes. ChatGPT-based models were the most commonly used LLMs, and retrieval-augmented generation techniques were applied in several studies. Chatbots were primarily designed for emotional support or information provision. Evaluation approaches varied widely, including response quality, psychological outcomes, and user experience. However, no studies evaluated interaction-level characteristics such as conversational continuity or context retention. Reporting on safety risks, mitigation strategies, and LLM-specific components such as prompt design and model parameters was often limited.

Conclusions:

Research on LLM-based multi-turn conversational chatbots for cancer patients and caregivers remains at an early stage, characterized by prototype-oriented development and heterogeneous evaluation approaches. Future studies should adopt standardized reporting frameworks and develop evaluation methods addressing interaction-level performance, safety, and clinical effectiveness.