AI Agent, Eye Disease, Large Language Models, Ophthalmology, Retinal Disease
Conditions
Keywords
medical AI agent, large language model, ophthalmology, tool integration, clinical decision support, real world study
Brief summary
This study is a multicenter randomized controlled trial evaluating the effectiveness and safety of EyeAgent, a multimodal artificial intelligence (AI) agent designed to assist ophthalmologists in clinical decision-making. Participants will be recruited from ophthalmology clinics and hospitals in Hong Kong and mainland China. The AI agent acts as a digital co-pilot, analyzing patient images and clinical history to provide diagnostic and management recommendations. The trial aims to determine whether the use of the AI agent improves diagnostic accuracy, treatment decision-making performance, report generation, workflow efficiency, and user satisfaction compared to standard clinical practice.
Detailed description
This multicenter, randomized controlled trial aims to evaluate the integration of EyeAgent, a multimodal artificial intelligence (AI) agent, in real-world clinical settings. The AI system is designed to support clinicians by analyzing patient data, including ocular images and electronic health records, to aid in image interpretation, diagnosis, and treatment planning. A total of 300 participants will be randomly assigned to either an AI-assisted care arm or a standard care arm. In the AI-assisted arm, clinicians review the comprehensive report generated by AI agent as a supportive tool before finalizing their independent decisions. The study comprehensively measures diagnostic accuracy, the rate of inappropriate treatment decisions, report generation, workflow efficiency, and user questionnaire. By comparing these two groups, the trial aims to provide robust evidence on the effectiveness and practical utility of AI-driven clinical decision support in ophthalmology, with the goal of enhancing both the quality and efficiency of patient care.
Interventions
DEVICEEyeAgent AI system
EyeAgent is a multimodal AI agent assistant for ophthalmology that integrates imaging, electronic health records, and curated clinical knowledge. In this arm, EyeAgent supports clinicians in clinical consultation, including report generation, diagnostic interpretation, and treatment planning.
Sponsors
The Hong Kong Polytechnic University
Study design
Allocation
RANDOMIZED
Intervention model
PARALLEL
Primary purpose
OTHER
Masking
DOUBLE (Subject, Outcomes Assessor)
Eligibility
Sex/Gender
ALL
Age
6 Years to 75 Years
Healthy volunteers
Yes
Inclusion criteria
1. Outpatient participants aged 6 to 75 years. 2. Participants who undergo ophthalmic examinations for medical purposes during the study period. 3. Participants who can produce clear ophthalmic images in both eyes. 4. Agree to participate in this study with written informed consent: 1. Participants aged 18 years or older provide their own consent. 2. Participants aged 6-17 years require consent from a parent or legal guardian.
Exclusion criteria
1. Participants who are reluctant to participate in this study. 2. Participants presenting with acute or emergency ocular conditions requiring immediate intervention. 3. Participants with poor quality of ophthalmic images, including blurriness, artifacts, underexposure, or overexposure. 4. Other unsuitable reasons determined by the evaluators.
Design outcomes
Primary
| Measure | Time frame | Description |
|---|---|---|
| Diagnostic accuracy rate | Immediately after the intervention. | Proportion of diagnoses consistent with a reference expert panel. |
| Rate of inappropriate treatment decisions | Immediately after the intervention. | The frequency of treatment recommendations (e.g., injection, laser therapy, or observation) that deviate from clinical guidelines as determined by the senior expert panel gold standard. Expert adjudication is conducted post-hoc after the enrollment phase concludes. |
Secondary
| Measure | Time frame | Description |
|---|---|---|
| Report quality | Within 1 month after enrollment. | Quality of clinical reports assessed using a structured Expert Report Quality Rubric evaluating five domains: accuracy, completeness, safety, reasoning, and interpretability. Each domain is scored on a 3-point scale (1 = poor, 2 = acceptable, 3 = good). Total scores range from 5 to 15. Higher scores indicate better report quality. |
| Clinician confidence | Immediately after the intervention. | Self-rated confidence in diagnosis and treatment planning measured using a single-item 5-point Likert scale (1 = not confident at all; 5 = extremely confident). |
| Workflow efficiency | During the index diagnostic session. | Time elapsed from image acquisition to final diagnosis and report completion. |
| Satisfaction and usability | At the end of each clinician's participation period, approximately 2 months. | Usability of the AI agent assessed using the System Usability Scale (SUS), a validated 10-item questionnaire scored on a 5-point Likert scale. Each item is scored from 1 (Strongly disagree) to 5 (Strongly agree). Total SUS scores are calculated according to standard scoring procedures and range from 0 to 100, with higher scores indicating better perceived usability. |
Countries
China
Contacts
CONTACTXiaolan Chen
CONTACTDanli Shi, Dr
PRINCIPAL_INVESTIGATORMingguang He
The Hong Kong Polytechnic University
Outcome results
None listed