Artificial Intelligence Governance of Clinical Model Withdrawal Notices Under Unmeasured Patient Notification Reach: A Healthcare Political Accountability Estimation

Clinical artificial intelligence models are withdrawn from service with increasing frequency, yet the governance of withdrawal notices rests on an unmeasured quantity: the proportion of affected patients who actually receive and comprehend a notice. the present analysis treats that unmeasured notification reach as a political accountability problem rather than a technical communication problem. It develops a formal account of withdrawal-notice governance in which notification reach enters as a latent parameter that conditions the accountability relationship between deployers, regulators and patients. Because reach cannot be observed directly, the paper characterises the operating characteristics of a before-after-control-impact design that would estimate the effect of a notice regime on accountability outcomes under a declared variance structure. The design assumes a between-cluster standard deviation of 6.0, a residual standard deviation of 4.5 and an intraclass correlation of 0.640, and evaluates power across 2000 replications per design cell for cluster counts of 8, 16, 24 and 32 and effect sizes of 3, 5 and 8 percentage points. The analysis indicates that 54 clusters are required to detect a 5 percentage point effect at 80 per cent power. The political legitimacy of model withdrawal depends less on the notice's content than on the institutional capacity to establish, and publicly defend, a credible lower bound on notification reach.