Generative AI in Pediatric and Neonatal Intensive Care Units: Tunisian Cross-Sectional Survey

Résumé

Background:
Generative artificial intelligence (AI), particularly large language models, is increasingly used in healthcare. In pediatric and neonatal intensive care, where clinical workload is high and decisions are often time-sensitive, its real-world use by young physicians remains poorly documented.

Objective:
To assess the use of generative AI among young physicians working in pediatric and neonatal intensive care, with a focus on clinical applications, risk-related practices, institutional governance, and training needs.

Methods:
We conducted a national cross-sectional online survey among residents and senior physicians practicing in pediatric intensive care units, neonatal intensive care units, or mixed units in Tunisia. Participation was voluntary and anonymous. The questionnaire explored demographic and professional characteristics, frequency and contexts of AI use, clinical applications, verification practices, use of patient data, perceived safety, institutional rules, and training priorities. A composite AI risk score was calculated among AI users.

Results:
A total of 94 physicians completed the survey, including 79 residents and 15 senior physicians. Overall, 88 respondents (93.6%) reported using generative AI for medical purposes during the previous three months. Among AI users, 72 (81.8%) used it at least weekly and 83 (94.3%) reported at least one clinical use. The most frequent clinical applications were explaining pathophysiological mechanisms, generating differential diagnoses, and discussing treatment or dosage. Insufficient verification of AI-generated answers was reported by 23 users (26.1%), while 58 (65.9%) had entered real patient clinical data into an AI tool. AI had influenced a real diagnostic or therapeutic decision in 56/87 respondents (64.4%). Overall, 18 AI users (20.5%) were classified as high risk. Clear institutional rules were absent for 85 respondents (90.4%), and only 11 (11.7%) had received formal training.

Conclusion:
Generative AI is already widely used by young physicians in pediatric and neonatal intensive care and extends into clinically influential domains. Its adoption appears to be occurring faster than the development of formal training and institutional governance. Structured education, confidentiality safeguards, and clear local guidance are needed to support safe and responsible use.