Trial Sample Size Estimation and Study Design Assessment using Monte Carlo Sampling
Claus Andersen and Goran Westerberg
Siena Biotech S.p.A.
Objectives: Construct a simulated clinical trial as close to clinical reality as possible benefitting from available clinical information.
Methods: To simulate a trial in Huntington’s Disease we brought together patient data and trial design to simulate the patient population distributions using Monte Carlo sampling. This started from the patient inclusion criteria and the randomization protocol, covering possible dosing regimens, and balanced as well as unbalanced study designs. Finally the primary end-point measured in terms of Total Functional Capacity (TFC)[1] was statistically assessed using various methods (ANCOVA, t-tests and Mann-Whitney U-test) after 5000 simulated trials.
Results: The trial simulations could faithfully include key elements of the trial design, thus proposing a change of inclusion criteria (excluding patients with TFC=13), and showed that a balanced design (same number of placebo as treated patients) increases power with respect to unbalanced designs. The sample size estimates showed that a 92% power could be obtained with 444 enrolled patients (222 in placebo and treated groups) assuming a 10% drop out rate and a 40% attenuation of decline in TFC. This was optimally assessed by an ANCOVA with alpha=0.05 and baseline TFC as covariate).
Conclusions: Constructing simulations of a trial allows the inclusion of design aspects which can influence the trial outcome just as much as sample size. Considering the investment by all stakeholders in a clinical trial the cost of simulating its key aspects are warranted, and allows design questions to be addressed scientifically and identification of some issues before the trial is run.
References:
[1] Neurology. 1979 Jan;29(1):1-3. Huntington disease: clinical care and evaluation. Shoulson I, Fahn S. PMID: 154626
