Oculomics as a filtered window: biological mismatches and a task-specific performance ceiling hypothesis for systemic disease prediction

Introduction Oculomics has emerged as a promising approach for inferring systemic health from retinal and ocular imaging. However, the biological meaning and achievable performance of retinal biomarkers may vary substantially across systemic prediction tasks.Areas covered This Critical Perspective synthesizes evidence from retinal imaging, artificial intelligence, ocular biometry, vascular biology, blood-barrier biology, systemic disease prediction, and clinical prediction methodology. Relevant literature was identified through targeted searches of PubMed and reference-list screening. We propose a ten-domain mismatch framework spanning biological and structural, temporal and contextual, inferential and methodological, and translational factors. We also propose a task-specific performance ceiling hypothesis: retinal-based models may eventually show diminishing returns, with higher ceilings expected when the target systemic process is more directly represented in ocular tissues and lower ceilings when the relationship is biologically indirect.Expert opinion Oculomics should progress from proof-of-concept prediction toward clinically defined applications in which retinal information provides reproducible, incremental, and actionable value. Future studies should move from performance-driven development toward mismatch-aware evaluation, incorporating longitudinal validation, clinically coherent endpoints, external validation, calibration, confounder sensitivity analysis, and biologically grounded interpretability.