Efficient design of an academic dose finding trial with intraindividual comparison to placebo

A Ring1,2, TBS Morris1, GA Johnston3, L Alexander1, D Fairlamb4. 1University of Leicester, Leicester, UK, 2University of the Free State, Bloemfontein, South Africa, 3University Hospitals of Leicester NHS Trust, Leicester, UK, 4Protherax, Bradford, UK

Background: A novel dose finding methodology will be implemented for a new first-in-human trial. The primary objective of the trial is to assess the safety and tolerability of the drug; therefore only 3 doses will be tested. The investigation of preliminary efficacy will be a secondary objective.

Clinical dose finding trials have traditionally been based on efficacy comparisons of each dose against placebo, with each t-test being powered for each dose level. For a trial with 3 doses (5, 10 and 20), a SD of 1 and a maximum effect of 1, an individually powered trial would require 23 subjects per dose group (69 in total) to achieve 90% power. Using Dunnetts test procedure, the sample size would be 20 subjects.

Methods: The R-packages “MCPMod” and “DoseFinding” [1,2] allow to design and analyse dose finding trials by comparing various effect models and investigating their impact on the selection of the dose. For the current trial, we assume that the dose-response follows a sigmoid Emax model, where EC50, Emax and the hill coefficient are unknown. Based on preclinical data, one dose level (10) has been selected to be tested with two surrounding doses. All subjects will receive one active dose and placebo in a randomised fashion, allowing for an intraindividual comparison (which requires a slight extension of the package).

As there is uncertainty with regard to EC50 and Emax, a requirement of the trial design is that it should maximise the power for dose selection over a large range of possible values of EC50 (EC50=2,3,4,8,10).

Results: The power using MCP-mod is universally larger than the power using Dunnett’s test.

With 3 doses, the improvement is between 0.05 and 0.12, with larger differences for higher EC50 values. With a sample size of 13 subjects per dose level (39 subjects in total), the power would be >90% for EC50<5, while the power would still be >80% for EC50<8.

Alternatively, 4 doses could be used (dose levels 7,10,15,20), with 10 subjects per dose level (40 subjects in total), which would achieve >80% power if the true EC50 would be less than 8, but more power than using 3 dose levels with larger values of EC50.

Conclusion: Modern dose finding methodologies allow the design of efficient trials, which require smaller sample sizes than traditional designs while achieving the same power. Moreover, they allow the power of the trial to be to quantified and maximised if there is uncertainty about the maximum effect and which dose is required to produce this effect. It is planned to extend the MCP-Mod package to cover the case of intraindividual comparisons to placebo.

References:

(1) Bornkamp B., Pinheiro J. C., and Bretz, F. (2009). MCPMod: An R Package for the Design and Analysis of Dose-Finding Studies, Journal of Statistical Software, 29(7), 1–23.

(2) Pinheiro, J. C., Bornkamp, B., Glimm, E. and Bretz, F. (2013) Model-based dose finding under model uncertainty using general parametric models, Technical report, http://arxiv.org/abs/1305.0889.