Overview
Digital twins are virtual replicas of individual patients that integrate multimodal clinical data to simulate health trajectories, evaluate interventions, and support personalized medicine. This project develops computational frameworks that combine agent-based modeling, discrete-event simulation, and machine learning to create dynamic, patient-specific models.
The Problem
Healthcare decisions are often based on population-level evidence that may not reflect an individual patient’s unique circumstances. Clinicians need tools that can:
- Predict how a specific patient’s condition will evolve
- Simulate the likely outcomes of different treatment options
- Account for the complex interactions between comorbidities, medications, and social determinants
Our Approach
We build digital twin frameworks using a hybrid simulation-ML architecture:
Agent-Based Modeling (ABM)
- Each patient is modeled as an autonomous agent with individual health states, risk factors, and care trajectories
- Agents interact with healthcare system components (clinics, specialists, interventions) modeled as entities in the simulation
- Population-level dynamics emerge from individual patient behaviors and system constraints
Discrete-Event Simulation (DES)
- Clinical encounters, procedures, and transitions are modeled as discrete events with stochastic timing
- Resource allocation and care pathway bottlenecks are captured to evaluate system-wide impacts
Machine Learning Integration
- Predictive models inform agent state transitions using real-world EHR data
- Transformer-based architectures capture complex temporal dependencies in patient histories
- Risk stratification models are continuously updated with incoming data
Applications
- Fall Risk Stratification — Real-time risk assessment for elderly patients using multimodal data (EHR, sensors, imaging)
- What-If Scenario Analysis — Comparing intervention strategies before implementation
- Population Health Simulation — Modeling disease burden and intervention impact at scale
Technical Capabilities
| Capability | Details |
|---|---|
| Multimodal Data Fusion | EHR, wearable sensors, imaging, social determinants |
| Stochastic Simulation | Monte Carlo methods for uncertainty quantification |
| Real-Time Updates | Online learning from streaming patient data |
| Scalability | Validated on cohorts of 7M+ patients |
This research is supported by NIH-funded projects through the Advancing Healthier Wisconsin Endowment and the CTSI Pilot-BERD program.