Simulation of Clinical Trials

Simulation of clinical trials sits at the design stage of a study, before protocol finalization and before the first participant is screened. Rather than choosing a sample size, dose, or adaptive-design rule by convention or by a single deterministic calculation, trial teams build a quantitative model — typically combining a pharmacokinetic-pharmacodynamic (PK/PD) model of the drug with a statistical model of disease progression and patient variability — and run that model repeatedly under randomly varied assumptions, most commonly using Monte Carlo methods. Each simulated run represents one plausible version of how the trial could unfold; running thousands of these in silico produces a distribution of outcomes (for example, the proportion of simulated trials that reach statistical significance at a proposed sample size) instead of a single estimate.

What simulation is used to decide

Three design decisions are the most common targets of clinical trial simulation:

• Sample size and power estimation. Classical power calculations typically assume a fixed effect size and variance. Simulation instead propagates uncertainty in those inputs — drawn from prior studies, meta-analyses, or PK/PD models — through repeated virtual trials, giving a more realistic estimate of the probability the actual trial will detect a true effect, and how that probability changes as assumptions shift.
• Dose selection. PK/PD simulation predicts the exposure-response relationship across a range of candidate doses before any dose is tested in the target population at scale, informing which doses to carry into later-phase testing and reducing the number of doses that need to be studied directly.
• Adaptive-design planning. For trials that include prospectively planned modifications — dropping an arm, re-estimating sample size, or changing randomization ratios based on accumulating data — simulation is how the trial’s decision rules are tested before the trial runs. Because an adaptive design’s actual statistical properties (its true type I error rate, its power) depend on the interim decision rules in a way that is often not solvable analytically, simulating the design under a range of scenarios is the standard way sponsors demonstrate to regulators that the proposed rules behave as intended. See Adaptive Design in Clinical Trials for how these rules are structured and justified.

Regulatory context: FDA’s Model-Informed Drug Development (MIDD) initiative

Clinical trial simulation is one application within a broader FDA initiative called Model-Informed Drug Development (MIDD): the use of exposure-based, biological, and statistical models derived from preclinical and clinical data to inform drug development and regulatory decisions. FDA’s MIDD Paired Meeting Program — which grew out of a pilot tied to a PDUFA VI performance goal and has continued through the PDUFA VII commitment period — gives sponsors a structured pathway to meet with FDA specifically about proposed modeling and simulation approaches, including trial simulations used to inform trial duration, response-measure selection, and predicted outcomes. For a research administrator or sponsor, the practical implication is that a simulation-informed design is not just an internal planning exercise — it can be the basis of a formal regulatory interaction, and the modeling assumptions behind it should be documented with the same rigor as any other part of the protocol’s statistical analysis plan.

Worked example

A sponsor planning a Phase 2b dose-ranging study has PK/PD data from a completed Phase 1 study. Rather than assume a single effect size for a standard power calculation, the biostatistics team builds a simulation model that draws effect size, variability, and dropout rate from distributions informed by the Phase 1 data and by published data on comparable compounds. Running 5,000 simulated trials at several candidate sample sizes shows that n=180 achieves 90% power across the plausible range of assumptions, while the originally proposed n=120 falls below 80% power in a meaningful fraction of simulated scenarios once dropout is accounted for. The sponsor revises the protocol’s sample size using the simulation output, and documents the model and its assumptions in the statistical analysis plan.

Counter-example

A single deterministic power calculation performed in standard statistical software — entering one assumed effect size, one assumed variance, and solving directly for the sample size that yields 80% power — is not clinical trial simulation, even though it is also a quantitative, prospective design activity. It produces one point estimate rather than a distribution of outcomes across repeated virtual trials, and it does not propagate uncertainty in the input assumptions the way a Monte Carlo simulation does. This distinction matters in practice: a deterministic calculation can understate the true risk that a trial is underpowered when its input assumptions turn out to be optimistic, which is precisely the gap simulation-based approaches are designed to expose.

Where this fits in trial planning

Simulation is one input among several at the design stage. For the broader landscape of design choices it feeds into — parallel-group versus crossover, adaptive versus fixed, single-arm versus controlled — see Clinical Study Design: The Major Types and How They Relate. For how simulated sample-size and power estimates translate into the concrete elements of a protocol, including endpoint definition, randomization, and the statistical analysis plan itself, see Designing a Clinical Trial: Endpoints, Sample Size, Randomization, SAP.