AI adoption among French oncology pharmacists: A national survey.

چکیده اصلی

BackgroundLarge language model (LLM)-based generative artificial intelligence (AI) tools have rapidly entered clinical practice. In oncology pharmacy, a specialty marked by high technical complexity, strict regulatory constraints, and an expanding clinical role spanning cytotoxic compounding, antineoplastic prescription validation, clinical pharmacy activities (medication reconciliation, pharmaceutical consultations, patient education), and advanced therapy medicinal product management, AI adoption carries specific risks and opportunities that remain poorly documented, and no national data exist for France. No national survey has characterized AI knowledge, use, or professional expectations among French oncology pharmacists.MethodsCross-sectional, anonymous web-based survey of French Society of Oncology Pharmacy (SFPO) members, conducted from March 30 to April 26, 2026. The 23-item questionnaire covered respondent profile, AI knowledge and tool awareness, current use and institutional framework, task-specific priorities and expectations, and barriers and governance needs. Descriptive statistics were applied to quantitative items; open-text responses underwent thematic coding.ResultsOf 374 SFPO members, 78 responded (response rate 20.9%; completion rate 71.8%, n = 56), predominantly senior pharmacists (92.2%) in public hospitals (75.3%), with 41.6% reporting more than 10 years of oncology experience. AI adoption was nearly universal (94.7%); 50.7% used a tool at least several times weekly. However, 65.3% operated within a shadow AI framework, while only 17.3% had access to fit-for-purpose validated tools. Email and administrative management dominated, cited by roughly 40% in the open-text time-allocation item and ranked first in the weighted prioritization exercise (weighted score 6.05; 26.2% ranking it first). The primary expectation was time savings on repetitive tasks (83.9%). Key barriers were the fear of erroneous outputs, in particular hallucinations (64.3%) and concerns over patient data confidentiality (62.5%; mean security concern score 3.21/4). A governance charter was supported by 96.4% of respondents, and 98.2% expressed interest in SFPO-organized specialty-specific training.ConclusionsAI adoption among French oncology pharmacists is already extensive but almost entirely unregulated, creating a critical governance gap with direct patient safety implications. The data reveal a pragmatic, specialty-specific demand for validated use-case libraries, integrated tools, and institutional governance frameworks. Governing this transition is an urgent task for hospital administrations and learned societies alike.

ترجمه فارسی چکیده

پس از ورود سریع ابزارهای هوش مصنوعی مبتنی بر مدل‌های زبانی بزرگ (LLM) به عملیات بالینی، در داروسازی اورولوژی که با پیچیدگی فنی بالا و محدودیت‌های نظارتی سخت مشخص می‌شود، پذیرش هوش مصنوعی ریسک‌ها و فرصت‌های خاصی را به همراه دارد که هنوز به خوبی مستند نشده است و داده‌های ملی برای فرانسه وجود ندارد. این مطالعه با استفاده از یک پرسشنامه آنلاین ناشناس از اعضای انجمن داروسازی اورولوژی فرانسه (SFPO) انجام شد. از میان ۳۷۴ عضو، ۷۸ پاسخ دادند (نرخ پاسخ ۲۰.۹٪). پذیرش هوش مصنوعی تقریباً جهانی بود، اما ۶۵.۳٪ از کاربران در چارچوب «هوش مصنوعی سایه» فعالیت می‌کردند و تنها ۱۷.۳٪ به ابزارهای تایید شده دسترسی داشتند. نگرانی‌های اصلی شامل ترس از خروجی‌های غلط و هالوسیشن‌ها و نگرانی‌های مربوط به محرمانگی داده‌های بیمار بود. ۹۶.۴٪ از پاسخ‌دهندگان از یک منشور حاکمیت حمایت کردند.

متن کامل اصلی