Published: May 11, 2025

Recent advances in large language models (LLMs) such as GPT-3/4 have spurred development of AI chatbots and advisory tools medicine. These systems are posited to assist or augment physician–patient communication, potentially improving empathy, clarity, responsiveness. However, their actual impact on communication outcomes remains uncertain. To systematically review meta-analyze peer-reviewed studies (2020–2025) evaluating how LLM-based interventions affect including trust, patient understanding. Following PRISMA 2020 guidelines, we searched PubMed/MEDLINE for published from 2025 examining LLM chatbot applications clinical contexts. Eligible designs included randomized, observational, cross-sectional, qualitative studies. Two reviewers independently screened titles/abstracts, assessed full texts, extracted data study design, population, type, measures, outcomes. We conducted a synthesis random-effects meta-analysis, reporting pooled standardized mean differences (SMD) odds ratios (OR) with 95% confidence intervals (CI). From 312 records, 10 (N=10) were included, all quantitative predominantly cross-sectional. Populations ranged patients chronic conditions healthcare professionals laypersons. Outcomes empathy (7 studies), clarity/information quality (6), satisfaction usefulness (4), trust perceptions (2). In six direct comparisons AI- versus physician-generated responses, LLMs rated significantly higher five One found replies judged empathetic 45.1% cases 4.6% physician (OR ~9.8, P<.001). Similarly, ChatGPT-4 answers scored 5-point scale than human-written responses (mean 4.18 vs 2.70, neurology showed scores (CARE +1.38, P<.01) ChatGPT answers. Only one no significant difference. content was also longer more information-rich, patient-perceived clarity On the other hand, GPT-4 simplified pathology reports, increasing comprehension (7.98 5.23/10, P<.001) reducing consultation time by 70%. sometimes less concise readable low-literacy patients. analyses (4 studies, n=2,604), positive effect (SMD +1.05, CI 0.45–1.65) improved understanding +0.82, 0.30–1.34). Patient results mixed. No directly long-term trust. Current evidence suggests can enhance producing empathetic, detailed, understandable responses. improvements may positively influence experience engagement. generate overly lengthy occasionally inaccurate advice, emphasizing need oversight. While meta-analytic findings promising, robust controlled trials needed confirm benefits, assess outcomes, define optimal integration strategies.

Language: Английский