The agent builds a high-dimensional patient embedding using demographics, comorbidity burden, left-ventricular function, coronary anatomy, and acuity. It then retrieves the closest-matching cases from the reference population using nearest-neighbor matching — returning the matched cohort size and the specific similarity criteria so the clinician can see exactly why each patient was selected.

The Digital Twin Agent returns observed 30-day and 1-year survival under each treatment strategy — PCI vs. CABG — for the matched cohort. Results are displayed as risk gauges and Kaplan-Meier survival curves. The 30-day mortality comparison is shown side by side (e.g. PCI 12.1% vs CABG 8.7%) so the heart team has a concrete, patient-specific risk differential before deciding.

The Digital Twin is validated across three large retrospective cohorts covering more than 40,000 coronary treatment decision cases with endpoints on 30-day mortality, 1-year mortality, and major cardiovascular events. Hold-out cohort testing was performed at multiple institutions. Every projection is reported with the matched cohort size — never as a single opaque number without a source.