"RMExplorer: A Visual Analytics Approach to Explore the Performance and the Fairness of Disease Risk Models on Population Subgroups"

"    This simulator models the performance and fairness of three atrial fibrillation (AF) risk prediction models \u2014 EHR-AF, CHARGE-AF, and C2HEST \u2014 applied to a synthetic population of patients stratified into five household income subgroups, replicating the core findings of the RMExplorer visual analytics system evaluated on the UK Biobank cohort of 445,329 individuals. The simulator generates synthetic patient populations with realistic demographic and clinical heterogeneity, computes continuous risk scores and 5-year AF probability predictions, simulates AF outcomes at a 1.66% incidence rate, and derives subgroup-level performance metrics (Harrell's C-index and calibration slope) and fairness metrics (statistical parity difference, true positive rate difference, and individual fairness violation rate) for both sex and race as protected attributes. It reproduces key phenomena including lower concordance and worse calibration for lower-income subgroups, consistently lower C2HEST performance, C2HEST individual fairness violation rates near 0.9 versus near 0 for EHR-AF and CHARGE-AF, and the sensitivity of fairness metrics to decision threshold changes from 0.05 to 0.08. Researchers can use this simulator to study how model bias and performance heterogeneity manifest across socioeconomic and demographic subgroups, students can learn about clinical AI fairness evaluation, and health informatics engineers can test visualization and auditing pipelines for risk model assessment tools."