Causal Estimation of Artificial Intelligence Diagnostic Delay Under Variable Radiology Reporting Queues: A Target-Trial Emulation of Healthcare Escalation Timing for Chest Radiograph Triage

Artificial intelligence triage of chest radiographs is often evaluated by discrimination and calibration, yet the clinically decisive quantity is the delay it introduces or removes from the escalation pathway. That delay is not a property of the model alone: it is produced jointly by model sensitivity, the queue discipline of the reporting service, and the timing rules that govern when a radiograph is escalated to a human reader. the present analysis formalises the escalation pathway as a target trial and derives the before-after-control-impact estimator appropriate to its emulation under staggered site-level adoption. Because no observational cohort is available, the design's operating characteristics are characterised analytically and by a replication-based evaluation under a declared variance structure with between-cluster standard deviation 6.0, residual standard deviation 4.5 and intraclass correlation 0.64. Across 2000 replications per design cell and cluster counts of 8, 16, 24 and 32, the analysis identifies the cluster count at which a 5 percentage-point reduction in time-to-escalation is detectable at 80% power as 54. The contribution is a design-theoretic account of what a causal claim about AI diagnostic delay would require, and a demonstration that queue variability, not model accuracy, is the binding constraint on the estimand's identifiability.