Abstract
Introduction
Pediatric surgeons face substantial administrative workload. Large language models
(LLMs) may streamline documentation, family communication, rapid reference, and education,
but raise concerns about accuracy, bias, and privacy. This review summarizes practical,
near-term uses with clinician oversight.
Materials and Methods
Narrative review of LLMs in pediatric surgical workflows and scholarly writing. Sources
included MEDLINE/PubMed, Scopus, Embase, Google Scholar, and policy documents (WHO,
FDA, EU). Searches spanned January 2015 to August 2025, English only. Peer-reviewed
and multicenter studies were prioritized; selected high-signal preprints were labeled.
Data screening and extraction were performed by the author; findings were synthesized
qualitatively.
Results
Across studies, LLMs reduced drafting time for discharge letters and operative note
registries while maintaining clinician-rated quality; they improved readability of
consent forms and postoperative instructions and supported patient education. For
decision support, general models performed well on structured medical questions, with
stronger results when grounded by retrieval. Common limits included coding performance,
case-nuance/temporal reasoning, variable translation outside high-resource languages,
and citation fabrication without curated sources. Privacy risks stemmed from logging,
rare-string memorization, and poorly scoped tool connections. Recommended controls
included a clinician-in-the-loop “review and release” workflow, privacy-preserving
deployments, version pinning, and ongoing monitoring aligned with early-evaluation
guidance.
Conclusion
When outputs are grounded in structured EHR data or curated retrieval and briefly
reviewed by clinicians, LLMs can responsibly reduce administrative burden and support
communication and education. Early adoption should target high-volume, low-risk, auditable
tasks. Future priorities must include multicenter pediatric datasets, transparent
benchmarks (accuracy, calibration, equity, time saved), and prospective studies linked
to safety outcomes.
Keywords
pediatric surgery - large language models - clinical documentation - artificial intelligence
- decision support
