A customized version of GPT-4 has demonstrated high accuracy and sensitivity in determining patient eligibility for retina-specific clinical trials, suggesting that this technology could significantly streamline the process of qualifying patients for clinical research in ophthalmology, according to a poster presented at ASRS 2024.
To conduct the study, publicly available eligibility criteria from 13 retina-specific clinical trials on clinicaltrials.gov were provided as a knowledge base for the GPT to reference. Researchers created 10 hypothetical patient profiles for each trial, containing information on disease indication, treatment experience, age, visual acuity, phakia status, central subfield thickness, and other study-specific factors that could impact eligibility. The GPT was then prompted to determine eligibility for each example patient using zero-shot prompting.
The preliminary findings were promising. The GPT-4-based system correctly identified the eligibility of 95% (n = 40) of the example patients across 4 different studies. It demonstrated a 100% sensitivity rate and a 90% specificity rate in its assessments.
These results suggest the system is highly effective at correctly identifying eligible patients while maintaining a reasonable accuracy rate in excluding ineligible patients.
Reference
Mitchell C, et al. Applications of GPT-4 in Determining Patient Eligibility for Clinical Trials Involving Retinal Disease. Poster presented at: ASRS 42nd Annual Meeting July 17–20, 2024.
Related Content
