D. Polhamus(1), J. Rogers(1), W. Gillespie(1), J. French(1), M. Gastonguay(1)
(1)Metrum Research Group, Tariffville, CT, USA
Objectives: Model-based drug development is ideally characterized by both comprehensive synthesis of available evidence as well as realistic simulation of future scenarios. To this end, a disease-drug-trial model for Alzheimer's Disease has been developed based on joint modeling of literature meta-data and individual patient data, summarizing available evidence with regard to rates of natural progression, placebo effects, and drug effects for marketed therapeutics [1,2]. Our objective was to facilitate the broad use of this model for the purpose of clinical trial simulation.
Methods: An R package was developed to provide a flexible framework for trial simulation based on the fitted model. The starting point for all package operations is a data matrix representing the posterior samples from the fitted model. Trial simulation proceeds by successively applying three core functions that "recruit", "randomize", and "run" the trial. The "recruit" function is used to generate covariate settings for individual patients, optionally using an included covariate imputation model. The "randomize" function is used to assign treatment sequences. Finally, the "run" function generates longitudinal response data for each individual, based on model-estimated parameters as well as user-specified treatment properties. Additionally, a function is provided to utilize a fitted drop-out model to generate missing data patterns. The robustness of the package design was assessed using several disparate use cases.
Results: The package architecture was sufficiently robust to accommodate all three attempted use cases: a 12 week cross-over design for assessing proof of concept of a symptomatic agent, an 84 week parallel group design for assessing efficacy of a disease modifying agent, and a 91 week delayed-start design for assessing efficacy of a disease modifying agent. Simulation-based estimates of operating characteristics were in approximate agreement with theoretical results in cases where the latter were available (e.g. simulation based estimates of assurance [4] were in approximate agreement with theoretical estimates of power for parallel group designs). The adsim package has also been used to explore trial designs in a real development program [3].
Conclusion: The adsim package enables simulation of a diverse class of clinical trial designs in Alzheimer's Disease, based on a comprehensive synthesis of available evidence.
References:
[1] WR Gillespie, JA Rogers, K Ito, MR Gastonguay. Population dose-response model for ADAS-cog scores in patients with Alzheimer's Disease by meta-analysis of a mixture of summary and individual data. ACoP 2009 Annual Meeting, October 4-7, 2009, Mashantucket, CT (http://2009.go-acop.org/sites/all/assets/webform/acop2009-adascog.pdf).
[2] J Rogers, D Polhamus, K Ito, K Romero, R Qiu, WR Gillespie, B Corrigan. The value of evidence synthesis: Model-based meta-analysis based on the CAMD database, the ADNI AD cohort data, and literature meta-data. ASCPT 2012 Annual Meeting, March 12-17, 2012, Washington DC.
[3] R Qiu, J Rogers, D Polhamus, K Romero, D Stephenson, K Ito, B Corrigan. Clinical trial simulations in Alzheimer's disease: Example applications of a modeling and simulation tool in drug development. ASCPT 2012 Annual Meeting, March 12-17, 2012, Washington DC.
[4] O'Hagan A, Stevens JW, Campbell MJ. Assurance in clinical trial design. Pharm Stat. 2005; 4: 187-201.
Reference: PAGE 21 () Abstr 2580 [www.page-meeting.org/?abstract=2580]
Poster: Study Design