Large language models in spine care and research

Area of Science:

• Artificial Intelligence in Medicine
• Natural Language Processing
• Clinical Informatics

Background:

• Transformer-based Large Language Models (LLMs) excel at understanding complex clinical language.
• This review examines the technical underpinnings of medical LLMs, including architecture and training.

Purpose of the Study:

• To review the technical foundations of medical LLMs.
• To survey applications of LLMs in spine surgery and spinal care.
• To explore multimodal models and discuss implementation challenges.

Main Methods:

• Review of technical foundations: transformer architecture, attention mechanisms, training paradigms, retrieval-augmented generation.
• Survey of documented applications in spine surgery and spinal care.
• Exploration of multimodal models integrating diverse data types.
• Discussion of implementation challenges: accuracy, privacy, bias, and regulatory constraints.

Main Results:

• LLMs demonstrate moderate guideline concordance (46-67%) for diagnostic support in spine care.
• Applications include automated note generation, literature review (68% novelty), and patient-friendly material creation.
• Multimodal models show superior performance in holistic diagnostic and prognostic tasks.
• Key challenges include accuracy, hallucinations, computational/privacy/regulatory constraints, and bias.

Conclusions:

• LLMs offer significant potential to enhance decision support, workflow efficiency, research, and patient communication in spine care.
• Safe and equitable deployment necessitates addressing model limitations and regulatory hurdles.
• Interdisciplinary collaboration and robust evaluation are crucial for responsible implementation.