An evaluation of DeepSeek and healthcare professionals' Q&A capabilities in improving patient-family satisfaction in the ICU.

Background: Families of intensive care unit (ICU) patients often require timely medical updates, creating additional communication demands for clinicians under heavy workloads. Large language models (LLMs), such as DeepSeek, may help support clinician-family communication, but their quality and reliability require empirical evaluation. Methods: Two ICU clinicians with more than 10 years of experience developed 30 representative questions reflecting common consultation scenarios encountered by family members of ICU patients. Responses were independently generated by clinician and DeepSeek-V3.1. Medical accuracy, information completeness, clarity and empathy, and safety assessment were evaluated by recruited simulated patients and ICU physicians, while satisfaction was evaluated by 600 family members of ICU patients. Results: DeepSeek showed significantly better performance than clinicians in information completeness, clarity and empathy, and patient-family satisfaction ( Conclusion: DeepSeek provided more complete, clear, and empathetic written responses to standardized ICU family consultation questions, with higher perceived family satisfaction than clinician responses. These findings suggest that LLMs may support ICU communication, but clinician oversight remains essential for individualized, prognostic, or safety-sensitive discussions. Real-world studies are needed to confirm their safety, reliability, and clinical applicability.