Written by: Lisa Moneymaker
Senior Vice President, Strategic Customer Engagement
Medidata Solutions
Background on this Good Clinical Podcast Deep Dive
Earlier in the year, Lisa Moneymaker and Adam Aten joined ACRO’s Good Clinical Podcast (GCP) to discuss ways the clinical research industry should be using insights from the past to better prepare our AI models and other technologies to meet the needs of patients today. Here, Lisa dives deeper into this topic as part of a special series exclusively made for our Informed Content Hub. Read along as Lisa connects themes from her episode to bigger questions and possibilities and the ideas that are shaping the future of AI/ML in clinical research.
Virtual Twins in Clinical Trials: Big Promise and Big Considerations
Imagine testing a protocol on a “virtual patient” before the first participant is screened. That begins the promise of virtual twins in clinical research: computational models of individuals or populations that simulate how participants might respond to an intervention, arm, or dosing regimen. Done right, they can accelerate timelines, reduce control-arm burden, and sharpen decision-making. Done carelessly, they can also encode the same diversity gaps that have plagued trials for decades.
Below, I unpack what twins are, how they’re already changing clinical development, and some concrete steps sponsors and CROs can take to avoid bias at scale.
What is Meant by “Virtual Twins”?
In life sciences, a virtual twin is typically a data-driven model of a real entity (a patient, organ, or process) that mirrors behavior. The concept also emphasizes an interactive loop: not just observing the system, but simulating and optimizing it with real-time feedback. Virtual twins can be seen as a digital counterpart of a patient or population that can underpin in-silico trial pipelines, running simulated participants through a protocol to predict outcomes and refine design before first patient in.
How are Twins Already Moving the Needle?
- Protocol simulation. Virtual twins can pressure-test inclusion/exclusion criteria, visit schedules, and enrollment predictions in advance, reducing amendments and timeline slippage.
- Patient-specific simulation of disease progression. Using genomic, imaging, biomarker, and clinical data, a virtual twin can be created to simulate how a specific patient would respond to interventions – allowing tailoring of therapies, adaptive dosing, and earlier prediction of adverse events.
- Test Data Generation. Virtual twin technology can rapidly create test data sets to validate checks, form builds, and support system testing – dramatically reducing manual efforts and reducing time to deployments.
The FDA has flagged the representativeness problem and has worked – through guidance and legislative mandates – to push sponsors toward diversity action plans that set concrete enrollment goals and tactics.
The Catch: Virtual Twins Inherit Our Data (and Its Inequities)
Virtual twins are only as fair as the data they learn from. Decades of under-representation (by race/ethnicity, sex, age, comorbidity, geography, socioeconomic status) can lead to model miscalibration for the very populations most affected by disease. If historical data skews white, male, and healthier, a twin may:
- Over or underestimate efficacy in underserved groups.
- Miss safety signals that manifest differently by ancestry, sex, or comorbidity.
- Drive biased site and patient selection, reinforcing access gaps.
This isn’t hypothetical. The FDA has flagged the representativeness problem and has worked – through guidance and legislative mandates – to push sponsors toward diversity action plans that set concrete enrollment goals and tactics. In mid-2024, FDA issued draft guidance detailing what such plans should include under the 2022 Food and Drug Omnibus Reform Act; early 2025 brought political turbulence, but HHS leadership publicly committed to finalizing rules to improve diversity.
Practical Playbook: Building Equitable Twins
Start with representative data, by design.
Blend Real-World Data/Real-World Evidence (RWD/RWE) (EHRs, claims, registries), device and wearable streams, and global trial archives to increase coverage of under-represented cohorts. Virtual twins can harness multi-modal data to simulate real participants, but the operative word is inclusive. If your sources don’t reflect your intended treatment population, your twin won’t either.
Keep humans in the loop, especially at decision points.
Use the virtual twin information to inform, not overrule, clinical and statistical judgment. Make effect estimates and uncertainty explainable to clinicians, statisticians, and IRBs. Bidirectional feedback is critical.
Actively de-bias the pipeline.
Apply re-weighting, counterfactual augmentation, and fairness-aware learning when subgroup performance lags. For control arm supplementation, ensure synthetic controls are drawn from demographically matched source data, or correct mismatch with propensity methods that include social determinants and comorbidity burden.
Expand access, not just accuracy.
Virtual twins can reduce patient burden (e.g., smaller controls, fewer visits, an understanding of procedure impact), which should increase participation for those historically excluded by logistics. Align twin-enabled efficiencies with your Diversity Action Plan – use them to open new geographies, support DCT/hybrid models, and offer culturally competent engagement. (As FDA’s evolving guidance makes clear, diversity is a planning problem as much as a data problem.)
The Road Ahead
The science is moving fast. Peer-reviewed overviews of in-silico trials show how virtual twins can both personalize care and complement, or in some contexts replace, traditional trial components. The governance is catching up, with regulators signaling openness provided sponsors demonstrate transparency, validation, and equity.
Vendors are converging on the same north star: credible simulations modeled on clinically relevant data. For sponsors and CROs, the implication is straightforward: you can pilot virtual twins now, but you must build them the way you build regulated systems and partner with vendors who do the same – documented, validated, monitored, and designed for fairness from the start.
Virtual twins can speed trials, reduce burden, and sharpen decisions – but only if they represent the patients you intend to serve. Bias isn’t a bug you can patch at the end, and bias control is a design choice you make at the beginning, with your data, validation criteria, and governance.
If we get this right, virtual twins won’t just simulate a better trial – they’ll change the way we design and run trials entirely.
Listen to today’s clinical research industry leaders share their insights on hot topics and trends like this on ACRO’s Good Clinical Podcast. Subscribe on Spotify, Apple Podcasts, and YouTube.