The Trial You Never Have to Run: How Simulation Is Rewriting Clinical Development Economics

Picture a regulatory affairs director at a mid-sized oncology sponsor staring at a Phase III protocol draft in late 2025. The statistician has just handed her a sample size calculation: 1,400 patients, two arms, 36-month follow-up. The CRO estimate sitting next to it reads $340 million. She knows the biology is promising. She also knows that 59.5% of drugs entering Phase III never make it to approval. The math, as written, is punishing enough that a single enrollment miscalculation or a futile interim could end the program entirely. What nobody in the room has suggested yet is running the trial virtually first, iterating on its architecture through simulation until the design is hardened against exactly those failure modes before a single patient is enrolled.

That is the core argument of a sweeping 2026 review in Nature Reviews Drug Discovery, which lays out the case for simulation-guided clinical trials as a systematic methodology for accelerated drug development. The paper treats simulation not as a statistical accessory but as a foundational design discipline: model the disease, model the trial, stress-test both, then build the protocol that survives the stress.

The implications for sponsors, CROs, and regulators are structural, not incremental. But the regulatory infrastructure around simulation remains uneven, the industry’s adoption is patchy, and the gap between what the methodology can deliver and what most sponsors actually do with it is wider than it should be in 2026.

Why the Current Design Process Is Failing Sponsors

The conventional trial design workflow runs roughly like this: literature review, protocol draft, sample size calculation based on assumed effect size, one or two sensitivity analyses, submission. The assumed effect size is the load-bearing wall of the entire structure. Get it wrong by 20%, and you are either massively overpowered and burning cash, or underpowered and heading toward a futility call that costs you three years and a program.

The problem is structural. Phase II trials are rarely large enough to anchor a precise Phase III effect size estimate. Model-based simulations suggest that if Phase III success rates improved from 59.5% to between 70% and 80%, overall development success rates could climb to approximately 13.0–15.8%, with corresponding R&D cost reductions of 10–14% per drug. Those numbers should force a reckoning. The industry loses billions annually to underpowered or mis-designed Phase III programs, and the solution is not larger Phase II trials. The solution is better pre-trial modeling.

The Nature Reviews Drug Discovery paper frames simulation as operating at three distinct levels: disease-level modeling (pharmacokinetic and pharmacodynamic, or PK/PD), trial operations modeling (enrollment, dropout, site performance), and statistical design simulation (adaptive rules, interim analyses, type I error control). Most sponsors who use simulation at all stop at the first level. The second two are where the operational leverage lives.

Consider what happens when you simulate trial operations. Enrollment projections built on historical site performance data from similar indications can reveal, before a single contract is signed, that your 18-month enrollment window is optimistic by eight months in three of your five planned geographies. You can redesign the site selection strategy in silico, repeatedly, at a fraction of the cost of discovering the same problem 12 months into an active trial.

The Regulatory Signal Is There. Sponsors Are Not Answering It.

The regulatory infrastructure supporting simulation-guided design has been assembling quietly for years. The FDA issued its final guidance on adaptive designs for drugs and biologics in 2019, which explicitly endorsed pre-specified simulation as the evidentiary basis for adaptive decision rules. The agency also published a guidance on assessing the credibility of computational modeling and simulation for medical device submissions, establishing a risk-informed framework for evaluating whether a simulation model is fit for regulatory purpose. On the European side, the EMA’s Committee for Medicinal Products for Human Use qualified MCP-Mod, the Multiple Comparison Procedure and Modelling approach, as an efficient statistical methodology for model-based Phase II dose-finding studies, an explicit regulatory endorsement of simulation-informed design logic entering the submission record.

Together, these documents signal that regulators are not waiting for industry to figure this out. They have written the framework. What they have not done is mandate simulation as a prerequisite for adaptive designs, which means sponsors can still file complex adaptive protocols without any documented simulation basis. Many do.

Here is the counterintuitive reality that the Nature Reviews paper surfaces, even if indirectly: simulation is not primarily a tool for designing more complex trials. It is a tool for justifying simpler ones. A sponsor who runs 10,000 simulated trials across plausible effect size distributions and dropout scenarios may conclude that a straightforward two-arm fixed design with a pre-planned interim is more robust to their specific uncertainty profile than an elaborate multi-arm platform. Simulation disciplines the instinct toward complexity. That is not a message the adaptive design software vendors advertise.

The PhRMA Working Group on Adaptive Designs, whose survey findings were published in the Journal of Biopharmaceutical Statistics, documented persistent industry tension between statistical elegance in adaptive designs and operational feasibility. Simulation is the tool that should resolve that tension, but the survey captured a gap: sponsors were designing adaptive trials without the simulation infrastructure to validate whether the adaptive rules were operationally executable under realistic enrollment and dropout conditions.

Digital Twins, Platform Trials, and Where the Evidence Is Hardening

The simulation methodology with the most immediate commercial momentum is the digital twin approach to control arm augmentation. In a retrospective analysis of AbbVie’s Alzheimer’s disease Tilavonemab trial, Unlearn.AI’s digital twin platform demonstrated a potential 13% reduction in overall sample size and a 23% reduction in the placebo arm size, through variance reduction rather than through any change in the fundamental statistical hypothesis. The drug, Tilavonemab, was targeting tau protein pathology; the disease area, Alzheimer’s, is one where placebo arm variability is a chronic source of Phase III failure. Shrinking the control arm by nearly a quarter without sacrificing statistical validity is not a marginal efficiency gain. In Alzheimer’s, where enrollment of appropriate patients can run at fewer than two patients per site per year across hundreds of sites, it is a program-saving difference.

The Nature Reviews paper situates digital twins within a broader simulation ecosystem that includes master protocols, which use shared infrastructure across multiple sub-studies, and platform trials, which allow arms to enter and exit based on pre-specified adaptive rules. Both designs benefit disproportionately from upfront simulation because their decision rules are far more numerous and interdependent than those of a standard two-arm trial. A platform trial with four active arms, two interim analyses per arm, and family-wise error rate control across arms has a combinatorial space of possible outcomes that no investigator team can reason through intuitively. You have to simulate it.

There is also a practical point about eligibility criteria that the research literature is beginning to quantify. A 2023 simulation study published in PubMed examined 24 permutations of eligibility criteria for a hepatocellular carcinoma trial involving transarterial radioembolization, varying three disease-nonspecific criteria to assess their effect on primary outcomes. The finding: eligibility criteria that look clinically conservative on paper can silently reduce the trial’s external validity and alter the enrolled population’s baseline risk profile in ways that shift the estimated treatment effect. Sponsors rarely simulate eligibility criteria. The Nature Reviews framework suggests they should treat eligibility as a design variable, not a clinical given, and simulate its downstream effects on efficacy estimation before locking the protocol.

The Operational Playbook Regulators Have Not Written Yet

The FDA’s credibility framework for computational modeling is rigorous on the question of model verification and validation. A simulation model must be shown to reproduce known data, its assumptions must be documented, and its sensitivity to those assumptions must be tested. What the framework does not specify, because it was written for medical devices rather than clinical trials, is how a sponsor should document simulation-based justifications in an IND or NDA submission. That gap creates real uncertainty. A sponsor who has run 50,000 simulated trial replications to justify a 15% adaptive sample size re-estimation rule has no FDA-specified template for presenting that evidence. Different review divisions handle it differently. Some request the simulation code. Some accept a summary table. The inconsistency is not negligible: it makes the regulatory return on simulation investment harder to predict, which suppresses adoption.

The practical recommendation here is direct. For any adaptive design IND where simulation was used to justify interim decision rules, sponsors should request a Type B meeting before submission and present the simulation framework explicitly, including the distributional assumptions, the number of replications, the software platform, and the sensitivity analyses. Do not wait for a review division to ask for it. The FDA’s 2019 adaptive design guidance states that sponsors “should discuss their proposed adaptive design with the FDA before finalizing the protocol,” and simulation documentation is the most substantive input a sponsor can bring to that conversation. Sponsors who treat the pre-IND meeting as a formality and submit simulation results only as a statistical appendix are leaving the most important regulatory alignment opportunity on the table.

The same logic applies to eligibility criteria simulations. If a sponsor has modeled the enrollment and efficacy implications of three alternative eligibility definitions, that analysis belongs in the briefing document for the pre-IND meeting, not buried in a protocol amendment filed after enrollment starts. Regulators cannot respond to information they have not seen, and the operational reality of simulation-guided design requires that the simulation informs the regulatory conversation, not merely the internal statistician’s recommendation.

For CROs, the Nature Reviews framework creates a specific competitive challenge. Sponsors who build simulation capacity in-house can validate CRO enrollment projections against their own operational models, creating a new layer of accountability in the sponsor-CRO relationship. The days when a CRO’s enrollment timeline was accepted on the basis of analogous trial experience and a site feasibility spreadsheet are ending. A sponsor running site-level enrollment simulations calibrated to historical performance data in the specific indication and geography can identify which CRO projections are systematically optimistic before the contract is signed. That is a different negotiating dynamic.

The regulatory affairs director from the opening scene, staring at that $340 million estimate, now has a methodology that lets her interrogate every assumption embedded in that number before committing to it. Run the disease model against the available PK/PD data. Simulate 10,000 trial variants. Find the design that survives plausible effect size pessimism, realistic enrollment drag, and a 15% dropout rate. Then negotiate the protocol and the CRO contract from a position of computational evidence rather than inherited convention.

The 59.5% Phase III failure rate has been the industry’s accepted background risk for long enough that it has started to feel like gravity. Simulation does not repeal gravity. But it tells you, before launch, exactly how far you are likely to fall.

References

1. Nature Reviews Drug Discovery — “Design, simulate, refine: simulation-guided clinical trials for accelerated drug development” (2026)
2. FDA — “Assessing the Credibility of Computational Modeling and Simulation for Medical Device Submissions” (Final Guidance)
3. EMA CHMP — “Qualification Opinion of MCP-Mod as an Efficient Statistical Methodology for Model-Based Design and Analysis of Phase II Dose-Finding Studies”
4. Unlearn.AI — “Real Trials, Real Impact: Q2 at Unlearn” (AbbVie Tilavonemab retrospective analysis)
5. PubMed — “Target Trial Emulation: Simulating Eligibility Criteria Permutations for Hepatocellular Carcinoma Trial” (2023)
6. Intuition Labs — “Adaptive Clinical Trial Design: Phase III Success Rate and R&D Cost Modeling”
7. PhRMA Working Group on Adaptive Designs — “Introduction to the Full White Paper,” Journal of Biopharmaceutical Statistics