Visual cueing and adaptive learning for teaching thyroid eye disease diagnostic assessment to ophthalmology residents

Background Thyroid eye disease (TED) diagnosis requires residents to interpret subtle clinical and imaging findings for activity and severity assessment. Guided by Cognitive Load Theory, visual cueing may support attention to diagnostically relevant features, whereas adaptive learning may tailor practice to learner performance. This pilot study compared visual cueing, adaptive learning, and traditional static case review for TED diagnostic training. Methods Forty-five ophthalmology residents were randomised equally to static case review (SCR), an adaptive learning system (ALS), or a visual cueing atlas (VCA). Following a 45-minute intervention, residents completed immediate and two-week delayed tests. The primary outcome was immediate overall diagnostic accuracy. Secondary outcomes were Clinical Activity Score (CAS) accuracy, severity-grading accuracy, delayed-test performance (2 weeks post-intervention), and performance decline from immediate to delayed testing. Educational materials were validated by ten expert ophthalmologists. Item-level content validity indices and the scale-level content validity index were calculated from expert ratings. Results All residents completed the study. Expert validation showed excellent content validity for the educational materials, with the Item-level Content Validity Index for all components and the overall Scale-level Content Validity Index both reaching 1.00. At the immediate post-test, VCA achieved higher CAS accuracy than SCR (84.7% vs. 64.7%, p < 0.001). ALS and VCA achieved higher severity-grading accuracy than SCR (83.3% and 78.7% vs. 66.7%; p = 0.004 and p = 0.025, respectively), and both achieved higher overall diagnostic accuracy than SCR (66.0% each vs. 51.3%; p = 0.014 and p = 0.008, respectively). At two weeks, ALS showed higher severity-grading accuracy than SCR (74.0% vs. 56.0%, p < 0.001), while both ALS and VCA maintained higher overall diagnostic accuracy than SCR (54.0% and 52.7% vs. 41.3%; p = 0.014 and p = 0.008, respectively). Conclusion In this pilot study, visual cueing was associated with higher immediate CAS accuracy, whereas adaptive learning was associated with higher severity-grading accuracy on the two-week delayed test, relative to static case review. These approaches may have complementary roles in TED diagnostic education. Larger studies using equivalent instructional exposure, matched assessment forms, and longer follow-up are needed before a staged hybrid approach is recommended.