[Paper Review] Bayesian dose-regimen assessment in early phase oncology incorporating pharmacokinetics and pharmacodynamics

Emma Gerard, Sarah Zohar+4 more|arXiv (Cornell University)|Nov 23, 2020

Statistical Methods in Clinical Trials28 references22 citations

TL;DR

This paper proposes DRtox, a Bayesian dose-regimen assessment method that integrates pharmacokinetic/pharmacodynamic (PK/PD) modeling to improve early-phase oncology trial design by estimating the maximum tolerated dose-regimen (MTD-regimen). By modeling cytokine response as a PD biomarker and using hierarchical or logistic regression, DRtox outperforms traditional methods in correctly identifying the MTD-regimen, especially under non-monotonic toxicity profiles, and enables recommendation of untested regimens for expansion cohorts.

ABSTRACT

Phase I dose-finding trials in oncology seek to find the maximum tolerated dose (MTD) of a drug under a specific schedule. Evaluating drug-schedules aims at improving treatment safety while maintaining efficacy. However, while we can reasonably assume that toxicity increases with the dose for cytotoxic drugs, the relationship between toxicity and multiple schedules remains elusive. We proposed a Bayesian dose-regimen assessment method (DRtox) using pharmacokinetics/pharmacodynamics (PK/PD) information to estimate the maximum tolerated dose-regimen (MTD-regimen), at the end of the dose-escalation stage of a trial to be recommended for the next phase. We modeled the binary toxicity via a PD endpoint and estimated the dose-regimen toxicity relationship through the integration of a dose-regimen PD model and a PD toxicity model. For the dose-regimen PD model, we considered nonlinear mixed-effects models, and for the PD toxicity model, we proposed the following two Bayesian approaches: a logistic model and a hierarchical model. We evaluated the operating characteristics of the DRtox through simulation studies under various scenarios. The results showed that our method outperforms traditional model-based designs demonstrating a higher percentage of correctly selecting the MTD-regimen. Moreover, the inclusion of PK/PD information in the DRtox helped provide more precise estimates for the entire dose-regimen toxicity curve; therefore the DRtox may recommend alternative untested regimens for expansion cohorts. The DRtox should be applied at the end of the dose-escalation stage of an ongoing trial for patients with relapsed or refractory acute myeloid leukemia (NCT03594955) once all toxicity and PK/PD data are collected.

Motivation & Objective

• To address the limitation of traditional dose-finding designs that treat complex dose-regimens as single doses, failing to account for intra-patient dose escalation effects.
• To improve the accuracy of maximum tolerated dose-regimen (MTD-regimen) selection in phase I oncology trials by incorporating PK/PD data.
• To model the non-monotonic relationship between dose-regimens and toxicity using a PD biomarker (cytokine response) and Bayesian hierarchical or logistic regression.
• To enable the recommendation of untested, potentially safer or more effective dose-regimens for expansion cohorts based on predicted toxicity probabilities.
• To provide a robust, data-driven method applicable post-dose-escalation using all collected toxicity and PK/PD data from ongoing trials.

Proposed method

• DRtox models the relationship between dose-regimens and toxicity using a nonlinear mixed-effects PK/PD model to estimate drug exposure and PD response over time.
• Toxicity is modeled as a binary outcome (e.g., CRS) linked to the maximum PD biomarker (e.g., peak cytokine levels), with two Bayesian approaches: logistic regression and hierarchical modeling.
• The hierarchical model enforces the constraint that toxicity must occur at the peak of the PD response, improving biological plausibility.
• The method integrates actual administered doses (not just planned regimens), allowing for handling of missing doses or incomplete PD data via PK/PD modeling.
• It uses informative priors based on prior knowledge of PK/PD profiles to improve estimation precision, especially with limited sample sizes.
• The approach is applied post-dose-escalation, using all collected toxicity and PK/PD data to estimate the full dose-regimen toxicity curve and recommend the MTD-regimen.

Experimental results

Research questions

• RQ1 Can integrating PK/PD data into a Bayesian dose-regimen assessment improve the accuracy of MTD-regimen selection compared to traditional model-based or algorithm-based designs?
• RQ2 How does the inclusion of PD biomarker profiles (e.g., peak cytokine levels) enhance the estimation of toxicity across multiple dose-regimens?
• RQ3 To what extent does the hierarchical model’s constraint—tying toxicity to the peak PD response—improve performance and biological relevance?
• RQ4 Can DRtox recommend untested, alternative dose-regimens for expansion cohorts with toxicity probabilities close to the target?
• RQ5 How robust is DRtox to model misspecification, missing data, or variability in PK/PD and toxicity profiles?

Key findings

• DRtox outperformed traditional model-based designs in correctly selecting the MTD-regimen, with a higher percentage of correct selections across all simulation scenarios.
• The inclusion of PK/PD information significantly improved the precision of the estimated dose-regimen toxicity curve, especially in non-monotonic toxicity profiles.
• The hierarchical-DRtox model demonstrated robustness, with only 2% of simulations failing due to undefined models, primarily when PK/PD estimation was inaccurate.
• The method successfully recommended alternative untested regimens for expansion cohorts based on predicted toxicity probabilities, even when not part of the original panel.
• Performance was enhanced when informative priors were based on reliable prior data, but poor-quality priors could reduce accuracy, highlighting the importance of data quality.
• The method effectively handled missing data: missing doses were addressed via actual regimen modeling, censored PK/PD data were treated as censored, and missing toxicity data were replaced in simulations.