Exploratory research in clinical and social pharmacy · 2026
Evaluation of large language model responses to patient questions on oral anticoagulant therapy: a comparative expert assessment.
СтатьяНазвание и аннотация: Перевод в очереди
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Khan Mohammed Amer, Chaudhary Vaibhav, Maazuddin Mohammed, Sultan Ihtisham, Mehnaaz Saamiya, Pal Biplab
Аннотация
BACKGROUND: Large language model (LLM)-based chatbots are increasingly used by patients for health information; however, their reliability in high-risk cardiovascular therapies such as oral anticoagulation remains uncertain. This study evaluated the perceived accuracy, clarity, and completeness of LLM-generated responses to common patient queries compared with standard-derived expert responses (SDERs).
METHODS: A cross-sectional comparative study evaluated responses generated by ChatGPT-4.5, Gemini Pro 2.5, and DeepSeek-V3 to 11 frequently asked questions related to five oral anticoagulants: warfarin, dabigatran, apixaban, rivaroxaban, and edoxaban. Responses were generated using a standardized patient-focused prompt. SDERs were developed by cardiologists and clinical pharmacists using authoritative references. All responses were anonymized and independently assessed by two blinded clinical pharmacists using a five-point Likert scale evaluating accuracy, clarity, and completeness. Interrater reliability was assessed using linearly weighted Cohen's κ, and group comparisons were analyzed using the Friedman test with post hoc adjustments.
RESULTS: Interrater reliability ranged from fair to almost perfect (κ = 0.31-0.84). ChatGPT-4.5 achieved the highest mean ratings across all evaluation domains, particularly for completeness. Significant differences were observed among response sources for accuracy, clarity, and completeness (p < 0.05). Warfarin-related queries demonstrated significant differences in accuracy and completeness across response sources, whereas responses for direct oral anticoagulants showed no significant differences.
CONCLUSION: ChatGPT-4.5 received the highest mean expert ratings for patient education regarding oral anticoagulants. Performance differences were most evident for warfarin-related queries, whereas responses for direct oral anticoagulants were broadly comparable across sources.
METHODS: A cross-sectional comparative study evaluated responses generated by ChatGPT-4.5, Gemini Pro 2.5, and DeepSeek-V3 to 11 frequently asked questions related to five oral anticoagulants: warfarin, dabigatran, apixaban, rivaroxaban, and edoxaban. Responses were generated using a standardized patient-focused prompt. SDERs were developed by cardiologists and clinical pharmacists using authoritative references. All responses were anonymized and independently assessed by two blinded clinical pharmacists using a five-point Likert scale evaluating accuracy, clarity, and completeness. Interrater reliability was assessed using linearly weighted Cohen's κ, and group comparisons were analyzed using the Friedman test with post hoc adjustments.
RESULTS: Interrater reliability ranged from fair to almost perfect (κ = 0.31-0.84). ChatGPT-4.5 achieved the highest mean ratings across all evaluation domains, particularly for completeness. Significant differences were observed among response sources for accuracy, clarity, and completeness (p < 0.05). Warfarin-related queries demonstrated significant differences in accuracy and completeness across response sources, whereas responses for direct oral anticoagulants showed no significant differences.
CONCLUSION: ChatGPT-4.5 received the highest mean expert ratings for patient education regarding oral anticoagulants. Performance differences were most evident for warfarin-related queries, whereas responses for direct oral anticoagulants were broadly comparable across sources.