Clinical Computer Vision Triage for Diabetic Retinopathy Under Gradual Camera-Sensor Drift: A Continual-Learning Evaluation of Fairness Degradation Across African Screening Cohorts

Gradual drift in camera-sensor response is an under-modelled threat to the clinical safety of computer-vision triage for diabetic retinopathy in African screening programmes, where devices are serviced infrequently and cohorts are heterogeneous. The framework separates covariate shift in image statistics from label-conditional degradation in sensitivity and specificity across cohorts, and defines a fairness-degradation estimand as the cohort-specific change in sensitivity attributable to drift. A variance-components model with between-cluster standard deviation 6.0, residual standard deviation 4.5 and intraclass correlation 0.64 was used to evaluate the design across 2000 replications per cell for cluster counts of 8, 16, 24 and 32 and effect sizes of 3, 5 and 8 percentage points. The design achieved 80% power at 54 clusters for a 5-percentage-point effect. Power rose with cluster count and effect size, and the minimum detectable effect at 80% power fell as clusters increased.