A scoping review and staged research agenda for artificial intelligence in viscoelastic haemostatic assays

Abstract Viscoelastic haemostatic assays (VHA), including thromboelastography and rotational thromboelastometry, capture dynamic whole-blood coagulation but are often interpreted through derived parameters and thresholds. We conducted a prospectively registered scoping review, aligned with PRISMA-ScR and JBI guidance, to map artificial intelligence (AI) and machine-learning applications involving VHA data. Five databases were searched, 493 unique records were screened and 27 studies were included. Most studies used VHA-derived parameters as predictors; three used VHA to define outcomes or phenotypes, and one modelled raw device signals. Applications spanned trauma, perioperative and transplant medicine, vascular/cardiology, obstetrics, critical care and haemostatic diagnostics. Reported discrimination was sometimes high, but evidence was limited by small cohorts, mixed-variable models, internal validation, abstract-only reports, limited calibration and scarce external testing. Current evidence supports minute-level acceleration, especially A5/A10-based inference. Three-to-five-minute raw-signal modelling is a plausible research target; sub-minute and 20-s prediction remain hypotheses.