Artificial Intelligence Snakebite Coagulopathy Prediction Under Variable Point-of-Care Coagulometer Calibration: A Conformal Risk-Control Evaluation of Healthcare Antivenom Dosing

Point-of-care coagulometers used in snakebite envenoming are calibrated against local reference plasmas, and the resulting measurement bias propagates into any artificial-intelligence model that maps coagulopathy to antivenom dose. the present analysis treats that propagation as a distributional problem and proposes a conformal risk-control layer that converts a probabilistic coagulopathy classifier into a dosing rule with a finite-sample guarantee on the rate of under-dosing. The guarantee is stated as a bound on the probability that the recommended dose falls below a clinician-specified threshold, conditional on the calibration regime under which the coagulometer operates. A formal derivation establishes that the bound holds uniformly over calibration shifts that lie within a declared total-variation ball, and that the required conformal quantile is inflated by an explicit function of the ball radius and the classifier's calibration error. The operating characteristics of the procedure are characterised under a declared variance structure with intraclass correlation 0.64, between-cluster standard deviation 6.0 and residual standard deviation 4.5; power was estimated across 2000 design cells for cluster counts of 8, 16, 24 and 32 and effect sizes of 3, 5 and 8 percentage points, and 54 clusters were required for 80% power at a 5-percentage-point effect. The contribution is a dosing rule whose safety claim is auditable against the calibration record of the device that produced the input, rather than a claim about classifier accuracy.