How Virtual Trials Will Reshape Clinical Development: Digital Twins and In Silico Cohorts at 5 Turning Points | Vol.4

Key Takeaways

• Conventional randomized controlled trials (RCT) remain the standard for new drug development but are the highest-cost, longest (5-7 years, $100-500M) process. “Virtual trials” leverage AI and real-world data to partially replace and complement RCT — a new paradigm.
• Three major technologies: (1) Synthetic Control Arms — generating placebo arms from external data rather than real patients, (2) Digital twins — predictive models for individual patients, (3) Decentralized Clinical Trials (DCT) — enabling patient participation from home rather than hospital visits.
• Regulatory progress: since the FDA’s 21st Century Cures Act (2016), incorporation of real-world evidence (RWE) and synthetic control arms into regulatory approval is accelerating. Unlearn.AI’s digital twins approved as Phase 2/3 synthetic control arms; Pfizer’s external control arm in cardiovascular trials approved; cases continue to multiply.
• Commercial value: synthetic control arms and remote trials can cut Phase 2/3 trial cost by 30-40%, reduce time by 20-30%, accelerate patient recruitment 2-3×. However, traditional RCT is not fully replaced; “hybrid trials” are becoming the standard.

Introduction — Lowering “The Trial Wall” with AI

Phase 2/3 clinical trials are pharma’s largest cost center. A typical Phase 3 trial:

• Patient count: 500-3,000
• Duration: 3-5 years
• Cost: $100-500M
• Failure rate: 30-50% (even at Phase 3)
• Required sites: 50-200 clinical research sites (multi-country, multi-continent)

Major causes of high cost: (a) patient recruitment difficulty (especially rare disease and small cancer subgroups), (b) ethical burden of placebo arms, (c) patient retention (dropout), (d) data collection / monitoring complexity, (e) multi-site / multi-region operation.

Virtual trial technologies are a new paradigm using AI and real-world data to partially solve these challenges. This article covers synthetic control arms, digital twins, remote trials, regulatory environment, and industry dynamics.

Main Body

1. Synthetic Control Arms

The Synthetic Control Arm (SCA) “reconstructs the placebo arm of a randomized trial from past clinical data using AI rather than real patients.”

Typical implementation:

1. Collect past patient data for the same disease (EHR, past trial data, patient registries)
2. AI extracts patients matching the new drug trial’s enrolled patients
3. Matching factors: age, sex, disease stage, biomarkers, comedications
4. Past patient cohort serves as “synthetic control arm” for comparison

Major examples:

• Medidata’s Synthetic Control Arm (Dassault subsidiary): extensively used in oncology trials
• Aetion: real-world evidence analysis platform, supporting SCA construction
• Flatiron Health (Roche): oncology EHR data for SCA generation
• Tempus Labs: integrated molecular + clinical data SCA

SCA advantages:

• No placebo arm needed — resolves ethical concern of untreated arm assignment
• Halved patient recruitment burden
• Shorter trial duration, reduced cost
• Particularly useful in rare diseases (where even active arm patients are hard to enroll)

SCA limitations:

• Depends on past data quality and completeness
• Risk of selection bias from imperfect matching
• Standard treatment evolution (past patients received older standard care)
• Requires careful regulatory evaluation

2. Digital Twin Trial Applications

Volume 3 covered digital twins; trial application uses them as “individual-patient predictive controls.” SCA generates population-level controls; digital twins predict “what would have happened to each individual patient without the new drug.”

Major examples:

• Unlearn.AI neurological digital twins: FDA-approved as Phase 2/3 synthetic control arms for Parkinson’s, ALS, Alzheimer’s
• Owkin (France): oncology digital twins, FL (Federated Learning) for privacy preservation
• Phesi: cardiovascular and metabolic disease digital twins

Individual patient digital twin advantages:

• Compute individual counterfactual for each patient
• Strengthen subgroup analysis (target biomarker positive vs negative)
• Long-term effect prediction (extension beyond trial duration)

3. Decentralized Clinical Trials (DCT)

Decentralized clinical trials (DCT) enable patient participation from home rather than hospital visits. The COVID-19 pandemic (2020-22) accelerated adoption; in 2026, an estimated 30-40% of Phase 1/2 trials and 20-30% of Phase 3 trials have remote elements.

Technical components:

• Telehealth consultation: physician-patient video dialog, symptom assessment
• Home nurse visits: blood draws, tests, medication administration
• Wearable / smartphone data collection: heart rate, activity, sleep, glucose, patient-reported outcomes (PRO)
• e-Consent: electronic consent for trial participation
• ePRO (electronic patient-reported outcomes): patient-recorded symptoms / QOL via smartphone

Major platforms:

• Science 37: DCT specialist, contracts with multiple major pharma
• Medable: DCT platform, used by Pfizer, Novartis, etc.
• Veeva Systems: trial management SaaS, integrated remote elements
• Apple Health Records: wearable-integrated data collection

4. Wearable + Machine Learning Continuous Monitoring

Conventional clinical trials centered on “single-time-point measurements at periodic visits”; wearable devices enable “continuous high-density measurement.”

Application examples:

• Cardiovascular trials: continuous ECG / heart rate via Apple Watch / Fitbit
• Diabetes trials: 24-hour glucose recording via CGM (continuous glucose monitor)
• Neurodegenerative diseases: automatic assessment of gait, tremor, cognitive function
• Oncology: continuous recording of activity, sleep quality, medication adherence

Machine learning extracts previously overlooked early efficacy and adverse event signals from these large data streams.

5. Real-World Evidence (RWE) Regulatory Integration

FDA seriously expanded RWE use in regulatory approval starting with the 21st Century Cures Act (2016):

• 2018: RWE framework released, guidance for RWE use in indication expansion
• 2021: Pfizer Ibrance (CDK4/6 inhibitor) male breast cancer indication expansion approved on RWE alone
• 2022: synthetic control arms regulatorily approved in multiple rare cancer trials
• 2024: FDA released formal AI/ML clinical trial design guidance draft
• 2026: full-scale operation of MIDD and RWE integrated framework

EMA (Europe) and PMDA (Japan) progressing similarly. International regulatory harmonization is in progress.

6. Hybrid Trials — “Complement, Not Replace”

The industry reality is not “AI virtual trials fully replace traditional RCT” but rather hybrid of both as standard.

Typical hybrid model:

• Active treatment arm: traditional RCT in real patients
• Control arm: real patients (half) + synthetic control arm (half)
• Data collection: routine visits + wearable + ePRO
• Subgroup analysis: enhanced with digital twins
• Long-term follow-up: extended with RWE

This hybrid achieves both regulatory approval reliability and operational efficiency.

7. Major Player Map

8. Limitations and Caveats

First, regulatory conservatism. FDA / EMA are advancing AI-driven trials but not to “full replacement of trials with AI.” Major approvals still center on real-patient RCT.

Second, data quality and standardization. Real-world data has missing values, noise, and bias. SCA / digital twin accuracy depends on data quality.

Third, patient privacy and ethics. Continuous wearable / home-test data collection raises patient privacy concerns. Consent process and data management transparency are critical.

Fourth, digital divide. Remote trials presume technology literacy and internet access. Risk of reduced participation by elderly, low-income, rural patients.

Summary

• Virtual trials are a new paradigm complementing traditional RCT with AI + real-world data.
• Three major technologies: synthetic control arms, digital twins, decentralized clinical trials.
• Regulatory progress: FDA RWE framework, Unlearn.AI Phase 2/3 digital twin approval, international regulatory harmonization.
• Commercial value: 30-40% Phase 2/3 cost reduction, 20-30% time reduction, 2-3× patient recruitment acceleration.
• Major players: Medidata, Unlearn.AI, Science 37, Medable, Veeva, internal pharma platforms.
• Limitations: regulatory conservatism, data quality, patient privacy, digital divide. Not full replacement but hybrid.

My Thoughts and Outlook

The article’s core insight: “Clinical trials may be the pharma process most transformed by AI”. Trial operations are pharma’s largest cost center, and AI efficiency gains here have multi-billion-dollar industry impact.
Three structural implications for the global ecosystem. First, the boundary between “trials” and “real-world post-marketing” continues to dissolve. Wearable continuous data, RWE long-term follow-up, post-approval phase 4 surveillance — all feed back into trial design and regulatory updates. The first FDA approval based primarily on continuous wearable data (likely 2027-28) defines a new evidentiary standard. Second, the digital divide becomes a regulatory equity issue. If decentralized trials systematically exclude older, low-income, or rural patients, regulatory bodies (FDA, EMA) will require demographic representativeness validation. AI-augmented trials must demonstrate inclusive participation. Third, “trial as a service” emerges as a market category. Just as cloud computing replaced internal IT infrastructure, “trial-as-a-service” platforms (Medidata, Veeva, Science 37) increasingly replace individual pharma’s bespoke trial machinery. The next 5 years will see major M&A in this space.
2026 is the year AI is rapidly commoditizing knowledge work. Clinical trial design and operation are precisely augmentation territory — AI accelerates patient matching, design optimization, data analysis, but clinicians, regulators, and ethics committees remain essential. Volume 5 (final) will explore pharmacovigilance AI.

Coming Next

Volume 5 (final) covers “Pharmacovigilance and AI — Catching Adverse Events with Machine Learning.” Automatic adverse event signal extraction from medical literature, social media, public databases; large-scale real-time surveillance; automated regulatory reporting — how AI transforms pharma’s safety monitoring operations, with series-wide synthesis.

Edited by the Morningglorysciences team.

Related Articles

• AI Drug Discovery: Front and Back, Vol. 3: Preclinical-to-Clinical Speed
• AI Drug Discovery: Front and Back, Vol. 5: Pharmacovigilance and AI (forthcoming)