BACKGROUND
Language
barriers
contribute
significantly
to
healthcare
disparities
in
the
United
States,
where
a
sizeable
proportion
of
patients
are
exclusively
Spanish-speaking.
In
orthopedic
surgery,
such
impact
both
patient
comprehension
and
engagement
with
available
resources.
Previous
studies
have
explored
utility
large
language
models
(LLMs)
for
medical
translation
but
yet
robustly
evaluate
AI-driven
simplification
materials
Spanish
speakers.
OBJECTIVE
This
study
utilizes
Bilingual
Evaluation
Understudy
(BLEU)
method
assess
quality
investigates
ability
AI
simplify
education
(PEMs)
Spanish.
METHODS
PEMs
(n
=
78)
from
American
Academy
Orthopaedic
Surgery
(AAOS)
were
translated
English
using
two
LLMs
(GPT-4
Google
Translate).
The
BLEU
methodology
was
applied
compare
translations
professional
human-translated
PEMs.
Friedman’s
test
Dunn’s
multiple
comparisons
used
statistically
quantify
differences
quality.
Readability
analysis
feature
subsequently
performed
text
success
features
on
scores.
capability
an
LLM
written
also
assessed.
RESULTS
As
measured
by
scores,
GPT-4
showed
moderate
translating
into
less
successful
than
Translate.
Simplified
demonstrated
improved
readability
compared
original
versions
(P<.001)
unable
reach
targeted
grade-level
simplification.
Feature
revealed
that
total
syllables
average
per
sentence
had
highest
able
reduce
complexity
(P
<.001).
CONCLUSIONS
While
Translate
outperformed
accuracy,
as
may
provide
significant
simplifying
texts
We
recommend
considering
dual
approach
order
improve
Spanish-speaking
accessibility
surgery.
