Author ORCID Identifier


Date of Graduation


Document Type

Dissertation (PhD)

Program Affiliation

Biomathematics and Biostatistics

Degree Name

Doctor of Philosophy (PhD)

Advisor/Committee Chair

Yisheng Li

Committee Member

Xuelin Huang

Committee Member

Jing Ning

Committee Member

Ryan Sun

Committee Member

Mark S. Chambers

Committee Member

Hussein A. Tawbi


Bayesian adaptive designs are getting more popular in research and in practice because they are flexible and efficient in evaluating an experimental drug. In oncology, despite the great advances in novel dose-finding designs, the high failure rates of clinical cancer drug development from phase I to III trials call for further improvements on novel designs, in addition to the need to promote and adopt novel designs in practice. Because anticancer agents often have a narrow therapeutic index, an accurate identification of the maximum tolerated dose (MTD) in a phase I trial is crucial for identifying a tolerable and efficacious dose through subsequent confirmatory trials. Since pharmacokinetics/pharmacodynamics (PK/PD) information reflects partially drug exposure and drug-receptor interactions, incorporating this information in phase I trial design may improve the efficiency of the MTD identification. Moreover, the European Medicines Agency also recommends incorporating relevant clinical and non-clinical information, including PK, PD, and toxicology data, in dose-escalation studies.

In this dissertation, we propose novel Bayesian adaptive dose-finding designs with a binary dose-limiting toxicity (DLT) outcome that incorporate PK/PD information. Specifically, we consider extending a recently proposed semi-mechanistic dose-finding (SDF) model framework to different practical settings in phase I cancer trials. The SDF model incorporates dynamic PK, latent PD, and DLT outcome in a unified framework, and employs Bayesian joint modeling of the PK and DLT outcomes. In Chapter 2, we propose an SDF design that allows systematic strength borrowing from data of the other schedule in trials with two treatment schedules. The proposed design employs an appropriate PK model, a generic (latent) PD model and a link function when modeling the dose/regimen-toxicity relationship, where data from different schedules can be pooled via an encompassing PK model under superposition principle. Our simulation studies show the SDF design with data pooling improves identification of the MTD in each schedule. In Chapter 3, we propose an extended SDF design by incorporating measurements for a PD biomarker relevant to the primary DLT in the toxicity-generating pathway from dose to concentration to DLT. We propose joint Bayesian modeling of the PK, PD and DLT data. Our simulation study shows the proposed design outperforms some common dose-finding designs, and it yields improved dose-toxicity curve estimation. In Chapter 4, we propose a new SDF trial design for a combination of two drugs with non-overlapping toxicities, where a factorial type Bliss model is used to model the drug-drug interaction at the toxicity level. Our simulation study shows the proposed design on average outperforms two common dose-finding designs for drug combination trials.

The results in the dissertation are promising, suggesting the usefulness of incorporating relevant PK/PD information of the experimental drug(s) when the underlying PK/PD mechanisms are reasonably understood. Further research is warranted to tackle the practical challenges in implementing our proposed designs.


dose finding, dose-toxicity curve, maximum tolerated dose, pharmacokinetics, pharmacodynamics, Bayesian adaptive trials

Available for download on Friday, April 24, 2026