Hypothesis: An ancestry-calibrated Bayesian prescribing model that starts with population-specific HLA-B*58:01 pretest risk and updates with eGFR, starting dose, thiazide exposure, and prior rash history will lower false-positive allopurinol avoidance compared with non-quantitative precaution rules, while preserving sensitivity for severe hypersensitivity events.

Why this matters: clinicians often face a tradeoff between under-screening high-risk patients and over-avoiding first-line urate-lowering therapy in lower-risk settings. A calibrated Bayesian framework may handle that tradeoff better than binary precaution logic.

Testable prediction: across external validation cohorts with heterogeneous ancestry structure, the Bayesian model will preserve >=95% sensitivity for adjudicated SCAR/AHS while improving specificity and decision-curve net benefit versus genotype-alone or ancestry-alone precaution strategies.

Suggested design: retrospective-development / prospective-validation study with stratified external validation, reporting calibration slope, calibration-in-the-large, AUC, NRI, and clinical utility under multiple treatment-threshold assumptions.

Falsification: if the Bayesian model does not improve specificity or net benefit at matched high sensitivity, the hypothesis fails.

References: FitzGerald JD et al. Arthritis Rheumatol. 2020;72(6):879-895. DOI:10.1002/art.41247. Goncalo M et al. Br J Dermatol. 2013;169(3):660-665. DOI:10.1111/bjd.12389. Lu CY et al. J Formos Med Assoc. 2019;118(1 Pt 1):206-214. DOI:10.1016/j.jfma.2018.06.006.