Can Artificial Intelligence Create Neonatal Resuscitation Simulations? 🤖👶

https://www.youtube.com/watch?v=1MKIMHG4kQA

Exploring the role of Large Language Models in neonatal education

Simulation-based training (SBT) is one of the cornerstones of modern neonatal resuscitation education. High-quality simulation helps healthcare professionals improve:

• Technical skills

• Team communication

• Crisis resource management

• Delivery room preparedness

Importantly, simulation-based neonatal resuscitation training has been associated with improved educational performance and better neonatal outcomes.

However, creating realistic, high-quality simulation scenarios is time-intensive and requires experienced instructors. This has led researchers to explore whether Large Language Models (LLMs)—such as ChatGPT and DeepSeek—could help generate neonatal resuscitation simulation scenarios dynamically.

🔬 What did we study?

In this multicenter exploratory pilot study, we evaluated the feasibility and challenges of LLM-generated neonatal resuscitation simulations.

Scenarios were generated using:

• ChatGPT-4o

• DeepSeek-R1

These were compared with equivalent scenarios derived from:

• Neonatal Resuscitation Program® (NRP®)

• RETAIN, a neonatal resuscitation serious game platform

The study included simulation cases such as:

• Extremely premature infant

• Placental abruption

• Born before arrival

• Meconium-stained amniotic fluid

All scenarios were standardized, anonymized, randomized, and evaluated by experienced neonatal resuscitation instructors from five centers who were blinded to the source of the scenarios.

📊 Key findings

✅ ChatGPT performed surprisingly well

Compared with traditional NRP® scenarios:

• ChatGPT-generated scenarios achieved comparable overall ratings

• No statistically significant difference was found in overall evaluation scores

Interestingly:

• ChatGPT achieved higher scores in debriefing design

• It performed particularly well in:

  • Creating clear learning objectives

  • Structuring debriefing discussions

  • Organizing scenario flow

⚠️ Important limitations remain

Despite promising results, several important gaps were identified.

ChatGPT:

• Occasionally struggled to provide consistent dynamic vital signs

• Still required instructor oversight and validation

DeepSeek:

• Demonstrated lower overall scores

• Showed deviations from NRP® algorithms

• Performed less well in providing appropriate clinical information

Importantly:

• Both LLMs demonstrated some degree of AI hallucination

• Human expert review remained essential before implementation

🧠 Why this matters

Artificial intelligence may offer exciting opportunities for neonatal education by:

• Reducing instructor workload

• Rapidly generating scenarios

• Creating customizable simulations

• Supporting global access to neonatal training resources

However, this study highlights an important message:

👉 AI-generated simulation is promising—but not yet ready to function independently.

Careful instructor oversight remains critical to ensure:

• Clinical accuracy

• Guideline adherence

• Educational quality

• Patient safety principles

🌍 The future of AI in neonatal simulation

This work represents an early step toward integrating AI into simulation-based neonatal education.

Future research will need to examine:

• Educational outcomes in learners

• Long-term retention

• Team performance

• Real-world implementation

• Safety and reliability of AI-generated content

As neonatal education continues to evolve, the combination of:

• Simulation

• Serious games

• Artificial intelligence

• Human expertise

may transform how neonatal resuscitation training is delivered around the world.

Technology can support education—but expert clinicians and educators remain at the center of neonatal resuscitation training.

https://pubmed.ncbi.nlm.nih.gov/42186080/