Modeling and Simulation in Clinical Trials: A Practical Approach to De-Risking Study Design

By the time a clinical trial begins enrolling, most of its risk has already been locked in. Failure rates remain stubbornly high. Phase II attrition continues to hover around 60 to 70 percent, and roughly half of Phase III trials fail.  

At the same time, today’s clinical trials are demanding more from researchers; more endpoints, more countries, and more data overall.  

In many cases, failures are tied to issues that could have been explored earlier: flawed assumptions, underpowered subgroup analyses, unrealistic enrollment projections, or endpoints that do not fully reflect how a treatment works. 

KerusCloud® was built to address that problem directly. 

Moving Beyond Static Assumptions 

Traditional study design often relies on a small set of fixed assumptions. A single effect size. A single dropout rate. A single enrollment curve. A single power calculation. 

In reality, none of those inputs are fixed. 

Enrollment fluctuates by site and geography. Dropout rates vary across treatment arms. Covariates interact. Endpoints correlate. Subgroups behave differently. Operational factors shift mid-trial. 

KerusCloud® allows these uncertainties to be explored simultaneously in a virtual environment. Rather than running one calculation at a time, statisticians can simulate thousands of scenarios in minutes. Outcomes are visualized in interactive heatmaps that highlight how probability of success changes when design parameters are adjusted. 

This approach surfaces the key drivers of study success early. It also makes trade-offs visible. Increasing sample collection rates, reducing treatment arms, adjusting dose levels, integrating pharmacogenetics prospectively instead of retrospectively, and more. 

Instead of guessing which combination will work best, sponsors can see it. 

Building Virtual Patient Populations 

KerusCloud® simulations are informed by diverse data sources, including historical trials, scientific literature, disease registries, and real-world data. These inputs are converted into synthetic datasets that reflect real patient-level behavior while avoiding privacy constraints. 

Importantly, these synthetic populations are designed to mimic real-world quirks such as missing data patterns and correlated endpoints. That matters because many statistical assumptions look clean on paper but break down under operational reality. 

By modeling those realities upfront, study teams can stress-test designs before the first patient is enrolled. 

Case Example: Enabling Early Approval 

A small biotech developing a treatment for C. difficile faced a familiar challenge. Initial planning suggested the program would require roughly 1,000 patients to support approval. For a resource-constrained sponsor working in a high-mortality population, that plan was not feasible. 

Using KerusCloud®, MMS evaluated alternative development strategies within the sponsor’s constraints. Data from multiple correlated endpoints were processed into synthetic datasets to construct virtual patient populations. Thousands of simulations were run to test “what if” scenarios around endpoints, enrollment rates, and design structure. 

The result was an alternative development plan requiring approximately 180 patients rather than 1,000.  

Regulators reviewed the simulated evidence package and provided scientific advice on how to proceed. The revised strategy reduced time to market by an estimated three to five years and saved more than $20 million in development costs. 

Case Example: Precision Medicine Rescue 

In another case, a Phase IIb Alzheimer’s study with 500 subjects across four treatment arms failed its primary endpoint. The sponsor considered a retrospective pharmacogenetic analysis using banked samples from roughly 60 percent of participants. However, the original study had not been powered prospectively for subgroup detection. 

KerusCloud® was used to compare the retrospective strategy with a smarter prospective design that integrated pharmacogenetics from the start. Virtual patient populations were constructed using published data and literature inputs. Simulations evaluated the probability of detecting a clinically meaningful genetically defined subgroup. 

The original design showed only about a 20 percent chance of success for identifying such a subgroup. By adjusting sample collection rates, reducing treatment arms, and applying dose-response modeling, probability of success increased by as much as 41 percent without increasing overall sample size or cost. 

Designing With Execution in Mind 

Simulation is often viewed as a pre-trial activity. In practice, its value extends further. 

Because KerusCloud® builds virtual populations tied to design assumptions, those populations can serve as a reference throughout trial execution. As real enrollment and outcome data accumulate, teams can assess whether the study is tracking as expected. If enrollment is slower than projected, if dropout rates diverge by arm, or if observed variance differs from assumptions, simulations can be revisited to understand impact on probability of success. 

This creates a feedback loop between design and execution. 

For small and mid-sized sponsors in particular, this matters. Resource constraints leave little room for design inefficiencies. Simulation allows study teams to ask harder questions earlier: 

• Are we enrolling the right patients?
• Are our endpoints meaningful for the target population?
• Is our sample size robust across plausible ranges?
• What happens if enrollment slows?
• Can we reduce patient burden without compromising power? 

It is far less costly and far less risky to answer those questions in a simulated environment than in an active trial. 

A Practical Shift in How Studies Are Planned 

Simulation does not change the fundamental way we work on studies. Sponsors still need robust evidence. Regulators still require credible data. Patients still deserve safe, effective therapies. 

What changes is how uncertainty is handled. 

Instead of treating uncertainty as something discovered after failure, simulation brings it forward into the planning stage. It allows teams to explore adaptive and fixed designs, evaluate operational feasibility, and quantify probability of success under realistic assumptions.  

As trial complexity continues to grow, static design approaches become increasingly fragile. Modeling and simulation provide a more resilient framework. 

The practical outcome is straightforward. Fewer avoidable failures. More efficient use of capital. Faster paths to meaningful answers. 

And for patients waiting on new therapies, that difference matters. 

See how KerusCloud® can transform study planning and execution and learn how simulation is shaping the future of trial design in our on‑demand Technology Use Cases webinar.