Precision Dose-Finding Design for Phase I Oncology Trials by Integrating Pharmacology Data

Abstract: Phase I oncology trials aim to identify a safe yet effective dose - often the maximum tolerated dose (MTD) - for subsequent studies. Conventional designs focus on population-level toxicity modeling, with recent attention on leveraging pharmacokinetic (PK) data to improve dose selection. We propose the Precision Dose-Finding (PDF) design, a novel Bayesian phase I framework that integrates individual patient PK profiles into the dose-finding process. By incorporating patient-specific PK parameters (such as volume of distribution and elimination rate), PDF models toxicity risk at the individual level, in contrast to traditional methods that ignore inter-patient variability. The trial is structured in two stages: an initial training stage to update model parameters using cohort-based dose escalation, and a subsequent test stage in which doses for new patients are chosen based on each patient's own PK-predicted toxicity probability. This two-stage approach enables truly personalized dose assignment while maintaining rigorous safety oversight. Extensive simulation studies demonstrate the feasibility of PDF and suggest that it provides improved safety and dosing precision relative to the continual reassessment method (CRM). The PDF design thus offers a refined dose-finding strategy that tailors the MTD to individual patients, aligning phase I trials with the ideals of precision medicine.

• The paper introduces a novel Bayesian design that integrates patient pharmacokinetic data to model individual toxicity risks.
• It employs a two-stage process combining cohort-based dose escalation with personalized dosing based on predicted toxicity probabilities.
• Simulation studies show improved maximum tolerated dose selection and enhanced patient safety compared to conventional methods.

Precision Dose-Finding Design for Phase I Oncology Trials

The paper "Precision Dose-Finding Design for Phase I Oncology Trials by Integrating Pharmacology Data" (2509.05120) introduces a novel methodological framework for Phase I oncology trials that aims to refine dose-finding by incorporating pharmacokinetic (PK) data. This approach directly addresses inter-patient variability by using individual patient data to model toxicity risk, thereby advancing towards truly personalized medicine in clinical trial design.

Overview of the Precision Dose-Finding Design

The paper presents the Precision Dose-Finding (PDF) design, a Bayesian Phase I trial methodology that integrates patient-specific PK data into the dose-finding process. The PDF design consists of two stages: a training stage for model parameter updates using cohort-based dose escalation, followed by a test stage where doses are tailored to individual patients based on PK-predicted toxicity probabilities.

The framework assumes a one-compartment PK model characterized by individual parameters such as the volume of distribution ($V_i$) and the elimination rate constant ($k_i$). The toxicity probability for each patient is calculated using these personalized PK metrics, distinguishing the PDF design from traditional population-level methods.

Figure 1: Dose assignment for the last patient treated in the precision dose-finding stage, illustrating personalized dose allocation with respect to individual toxicity profiles.

Methodological Details and Computational Approach

Bayesian Probability Model

The model utilizes the area under the concentration-time curve (AUC) as a covariate, which is derived from the patient-specific PK parameters. The toxicity probability $p_i$ for patient $i$ is expressed as:

$$p_i = logit^{-1}(\beta_0 + \beta_1 \times \log \frac{d_i}{V_i k_i}),$$

where $d_i$ is the administered dose, and $\beta_0$, $\beta_1$ are model parameters estimated using MCMC simulations. This individualized approach facilitates prediction of toxicity risk through posterior samples, maintaining robust inference throughout the trial.

Stage I: Cohort-Based Dose Escalation

The initial stage employs cohort-based dose escalation, where decisions are based on predictive toxicity probabilities derived from cumulative patient data. The approach adheres to traditional safety and escalation rules to balance dose exploration and patient safety effectively.

Stage II: Precision Dose-Finding

In the test stage, each new patient receives a dose tailored to his/her predicted toxicity profile, enhancing dosing precision. Selection of individual doses is informed by the posterior summaries of $V_i$ and $k_i$, ensuring personalized treatment optimization.

Figure 2: Dose assignments and DLT outcomes of six trials in scenario 1, showcasing dose escalation and precision dose-finding phases.

Simulation Study

Extensive simulation studies corroborate the PDF design's efficacy, demonstrating improved safety and dosing precision compared to the CRM. The simulations exhibit the strength of personalized dosing in reducing overdose risk and enhancing MTD identification accuracy, particularly in scenarios with high variability in patient responses.

Results

Results indicate that PDF efficiently allocates patients to appropriate dose levels, often achieving superior selection of MTDs with lower risk of toxicities. While the PDF approach aligns closely with CRM in identifying correct dose levels, it demonstrates a notable advantage in patient-specific safety profiles.

Figure 3: Examples of three volumes of distribution $V_i$ with varying elimination rate constants $k_i$ and their corresponding predicted MTD values.

Discussion and Future Directions

The PDF design represents a significant step towards incorporating precision medicine principles into early-phase oncology trials. By leveraging individual PK data, this method provides a more accurate and personalized strategy for dose assignment, paving the way for innovations in clinical pharmacology and therapeutic customization.

Future research may focus on extending the PDF model to include more complex PK/PD relationships and dynamic dose adaptation strategies. Additionally, further investigation into the integration of machine learning approaches for real-time patient data analysis could enhance prediction accuracy and model generalizability.

Conclusion

The proposed PDF design effectively integrates individual PK data into the dose-finding process, offering a refined strategy for Phase I trials. With its Bayesian framework, the method supports personalized dosing while maintaining rigorous safety oversight, aligning early-phase studies with the ideals of precision oncology.