This study employs a two-arm randomized controlled trial to evaluate whether artificial
intelligence (AI)-generated plain language summaries (PLSs) can improve patient
comprehension of ophthalmology notes. Eligible participants are recruited during their
routine visits at the Jules Stein Eye Institute, and once screened using standardized
clinical criteria, they are randomly assigned to either receive the standard
ophthalmology note (SON) or the SON supplemented with an AI-generated PLS. The
randomization process uses a computer-generated sequence with concealed allocation to
ensure unbiased group assignment.
The AI system used in this study is deployed locally on a secured UCLA intranet. It
leverages a large language model (LLM) that has been customized and validated for
generating plain language explanations of complex ophthalmologic information. All
processing occurs on UCLA-approved, encrypted devices, and no data are transmitted
externally. Before the PLS is provided to participants, each summary is reviewed by an
ophthalmologist to verify accuracy and ensure that essential clinical details are
correctly and clearly communicated.
Data collection is performed using survey instruments. The survey includes a series of
5-point Likert scale items, open-ended questions, and structured response sections
designed to assess comprehension of diagnosis, treatment plans, and follow-up
instructions. Participants complete the survey immediately after their clinic visit, and
a follow-up telephone interview is conducted approximately seven days later by trained
research staff to capture additional feedback on clarity and retention of the information
provided. The study does not employ audio or video recording; all responses are either
directly recorded by research personnel or entered electronically into a secured
database.
Statistical analyses will be conducted using standard software packages to compare
outcomes between the intervention and control groups. Primary analyses include
independent t-tests or Mann-Whitney U tests for continuous variables, chi-square tests
for categorical variables, and multivariable regression models to adjust for confounding
variables such as age, education level, and baseline health literacy. The sample size was
calculated to detect clinically meaningful differences in comprehension scores, with
power analyses indicating a need for between 460 and 2030 participants depending on the
effect size.
Data security is maintained through rigorous measures. Electronic data are stored on
encrypted, UCLA-secured laptops and in a secure Box repository. All data handling follows
UCLA policies and IRB guidelines for data retention and destruction, with identifiable
information destroyed using secure methods once it is no longer required.
Quality control procedures include periodic audits of data entry, regular review meetings
with study personnel, and cross-checks of survey responses against clinical records where
applicable. An independent monitoring process is in place to ensure compliance with the
study protocol and to address any deviations promptly.
Overall, this study is designed to provide robust evidence on the feasibility and
effectiveness of AI-generated PLSs in enhancing patient understanding of complex medical
information. By integrating technical safeguards, rigorous statistical methods, and a
streamlined data collection process, the research aims to deliver insights that may lead
to improved patient communication strategies and more effective health care delivery
across multiple specialties.