Page 73 - Read Online

Page 487 Bala et al. Art Int Surg. 2025;5:476-89 https://dx.doi.org/10.20517/ais.2025.44

mentions of complications, offering a richer understanding of real-world patient outcomes.

However, this study has some limitations that should be acknowledged. First, the evaluation was confined to
the FDA MAUDE dataset, without external validation across diverse clinical settings or international
regulatory databases. Second, the k = 20 parameter for MMR was chosen based on common practices rather
than systematic optimization, which may not be optimal for all clinical text types. Third, the study focused
on English-language reports from a single regulatory system, limiting applicability to multilingual or
international contexts. Finally, while RAG reduces hallucination, it cannot eliminate the inherent
subjectivity of patient-reported narratives or guarantee complete accuracy in safety-critical applications,
where human expert oversight remains essential.

Looking ahead, future work can explore fine-tuning the generative model on domain-specific corpora to
further enhance contextual understanding. Incorporating external clinical ontologies or knowledge graphs
could also support more precise identification and normalization of adverse event terms. Moreover,
expanding the framework to handle multilingual reports or to integrate temporal progression of symptoms
could provide deeper insights into long-term patient outcomes . Lastly, combining this system with
[34]
clinician feedback loops may support semi-automated adverse event validation pipelines for regulatory or
pharmacovigilance purposes. Additionally, expanding validation beyond the FDA MAUDE dataset to
include diverse clinical settings and direct engagement with healthcare stakeholders will be essential for
establishing practical utility and supporting broader clinical adoption of this RAG-enhanced approach in
healthcare decision-making systems.

DECLARATIONS
Authors’ contributions
Conceptualized the study, developed the methodology, conducted the analysis, and wrote the initial draft:
Bala I
Contributed to the development of the methodology and provided a critical review: Sharma E
Supervised the research, provided guidance throughout the study, and contributed to the review and editing
of the manuscript: Mitchell L
All authors read and approved the final manuscript.

Availability of data and materials
The dataset used in this study is publicly available from the U.S. Food and Drug Administration’s
Manufacturer and User Facility Device Experience (MAUDE) database (https://www.accessdata.fda.gov/
scripts/cdrh/cfdocs/cfmaude/search.cfm), which can be accessed online at the FDA website. The specific
dataset and analysis codes are available from the corresponding author upon reasonable request.

Financial support and sponsorship
This research received no specific grant from any funding agency in the public, commercial, or not-for-
profit sectors.

Conflicts of interest
The authors declared that there are no conflicts of interest.

Ethical approval and consent to participate
Not applicable.