Return to search

Incorporation of expert beliefs in the two-parameter Bayesian logistic dose response model

Logistic regression models are often proposed to describe dose-response relationships in dose-escalation clinical trials to determine the maximum tolerated dose. In a Bayesian setting, the 1-parameter continual reassessment method and the 2-parameter escalation with overdose control designs have been implemented assuming acceptable tolerability thresholds of between 20% and 35%. The literature is sparse on the operating characteristics of 2-parameter Bayesian logistic regression models (BLRM) when sample sizes are small (i.e. <50); response rates <20% or >35% are of interest; and expert beliefs are available for incorporating into prior distributions for model parameters. Motivated by a case study of a new infertility treatment, this thesis describes the operating characteristics of the 2-parameter BLRM in a dose-escalation setting, with small sample sizes, and applied to response rates consistent with both safety and efficacy endpoints i.e. 10% to 90%. When information external to the trial is available from expert beliefs, ways in which those beliefs may be elicited in a structured manner are evaluated. Simulation is used to assess the impact of these prior distributions on trial conclusions. I have demonstrated that elicitation can be performed in a structured manner in both academic and industry settings and I provide specific recommendations for the structured planning and execution of elicitation sessions. Simulations show that there is no single set of priors that always produce unbiased estimates with minimum variance across a range of target response rates, so simulations specific to the planned trial must be conducted. Furthermore, when only discrete doses are available, simulations show that choosing the available dose closest to that recommended by the model is more likely to lead to an unbiased estimate of the dose that attains a pre-specified response rate. Recommendations are provided for how to improve the study design and analysis for the motivating case study.
Date January 2015
CreatorsKinnersley, Nelson Maxwell
ContributorsAshby, Deborah; Wilkins, Martin
PublisherImperial College London
Source SetsEthos UK
Detected LanguageEnglish
TypeElectronic Thesis or Dissertation

Page generated in 0.0021 seconds