A Bayesian Approach to Dose-finding Studies for Cancer Therapies: Producing Personalised Procedures While Incorporating Information from Later Cycles of Therapy.

Bayesian model based approaches for Phase I dose-finding studies are popular procedures to implement due to the efficiency of updating information sequentially after accruing information. Traditionally dose-finding studies for cancer treatments focus on the occurrence of a patient's first dose limiting toxicity in the first cycle of therapy. This thesis develops a Bayesian decision procedure featuring an Interval-Censored Survival model to incorporate information from multiple cycles of therapy. The use of data from multiple cycles of therapy should produce more precise estimates of target doses to recommend for further investigation in later phases of drug development, in a shorter amount of time. An increasingly desired approach in dose-finding procedures is to provide personalised procedures to target therapy to individual tolerances. Features such as allowing intra-patient dose adjustments and incorporating baseline characteristics to investigate the underlying drug tolerance of population subgroups are investigated within the use of the Interval-Censored Survival Decision Procedure (ICSDP). The inclusion of time-varying covariates is also possible when using the ICSDP, which is investigated through including lower grade toxicities as a marker for tolerance. Individual target doses can be estimated, but the analysis of dose limiting toxicities alone provides a population target dose to recommend for further investigation. Results show the ICSDP as an efficient approach to use when observing a patient's first dose limiting toxicity. Target doses are estimated with good precision, comparable to or better than existing designs for dose-finding, and are generally obtained in a shorter amount of time. Multiple target doses can be produced for different subgroups of the population when baseline characteristics are used and intrapatient dose adjustments are possible between cycles. When using intra-patient adjustments based on observation of lower grade toxicities, personalised dose-escalations lead to estimates of individual target doses and a population target dose with good precision in a reduced amount of time