Prospective In-silico Simulation of the VESALIUS-CV Trial Using Biomedical Knowledge Graph and Real-World Data-Driven AI Modeling
Year: 2026
Type: preprint
Abstract: Cardiovascular outcomes trials are lengthy and costly. In silico trial simulation using real-world data may support earlier decision-making, but evidence generated before trial results are disclosed remains limited. We applied a semi-mechanistic machine learning framework integrating patient-level real-world data with knowledge graph-derived drug representations to simulate the VESALIUS-CV trial of evolocumab versus placebo before its results were public, using a locked model and prespecified trial design aligned with the trial protocol. Patient-level time-to-event predictions were aggregated into virtual trial arms to estimate cumulative incidence, hazard ratios (HR), confidence intervals (CI), and p-values for three-point major adverse cardiovascular events. In retrospective evaluation of the model's patient-level predictive performance, time-dependent areas under the receiver operating characteristic curve ranged from 0.80 to 0.90 across follow-up horizons. In evaluation of retrospective trial-level predictive performance across 22 cardiovascular trials, the model predicted trial meeting the primary outcome with an accuracy of 0.91 (95% CI 0.77-1) and F1 score of 0.95 (95%CI 0.86-1). The prospective simulation predicted a significant reduction in major adverse cardiovascular events with evolocumab (HR 0.78; 95%CI, 0.70 to 0.87) at 54 months. The trial subsequently reported a HR of 0.75 (95% confidence interval, 0.65 to 0.86) at a median follow-up of 55 months, with progressive separation of event curves in both the simulation and the trial. The prospective simulation was concordant with the direction, magnitude, and temporal dynamics of the observed treatment effect, supporting in silico simulation as a tool for early decision-making, trial design, and de-risking in cardiovascular drug development.
Source: medRxiv
Language: English
Subfield: Statistics and Probability
Field: Mathematics
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7204754830