Early-phase oncology trials are increasingly exploring combination therapies to improve outcomes for patients with complex cancers. While these approaches hold promise, they also introduce new methodological challenges, particularly when identifying the optimal dose combination that balances efficacy with acceptable toxicity.
In this blog, we explore how recent methodological research [1] from our statistical experts, Matt George and Ian Wadsworth, in collaboration with Pavel Mozgunov of the MRC Biostatistics Unit in Cambridge, applies to real-world early-phase oncology trials, with a focus on improving dose-finding flexibility, precision, and patient safety.
The Practical Challenge: What if No Combinations are Desirable?
In early-phase combination trials, sponsors typically begin with a predefined grid of dose combinations. The goal is to identify the maximum tolerated dose combination, one that aligns closely with the target toxicity level. However, a challenge can arise here. What if none of the predefined combinations are actually close to the desired toxicity target?
When this happens, the trial may:
- Drift toward overly conservative, subtherapeutic recommendations
- Escalate toward combinations that exceed acceptable toxicity thresholds
Historically, teams have addressed this issue through ad hoc adjustments, such as manually adding new dose combinations mid-trial. While pragmatic, these approaches can introduce operational complexity, regulatory uncertainty, and potential bias if not guided by a clear statistical framework.
This creates a need for a more systematic and defensible way to adapt the dose grid during the trial.
A Smarter Solution: Contour-Based Dose Insertion
The proposed approach developed by our experts introduces a formal method for inserting new dose combinations mid-trial using a contour-based framework.
Applied to an existing dose-finding design, at a high level, the method works by:
- Identifying the contour that partitions combinations truly above and below the target toxicity based on cumulative data
- Triggering a new dose insertion if we identify this contour with a degree of certainty, as this suggests no existing combination is close to the target toxicity
- Determining which (and how many) dose levels to insert
This method asks whether the current dose grid allows the trial to realistically identify a combination near the desired toxicity. If the answer is no, the design adaptively inserts new, more informative dose combinations to explore, in a statistically principled way.
This creates a data-driven mechanism for adaptation rather than relying on reactive or ad hoc decisions.
Why This Matters: Efficiency, Precision, and Safety
Improved Precision in Identifying Combinations at the Target Toxicity
By introducing new combinations, the design can increase the likelihood of identifying dose combinations that align with the target toxicity level. Simulation results show a higher probability of selecting combinations close to the target toxicity.
Maintained Patient Safety
This added flexibility does not increase the probability of recommending overly toxic or subtherapeutic combinations. The insertion is only triggered when there is evidence supporting that the current grid is insufficient, preserving the core safety principles of model-informed dose finding.
Greater Trial Efficiency
When the initial dose grid is suboptimal, traditional designs may spend valuable cohorts exploring combinations that are far from the target toxicity. Mid-trial insertions focus learning where it matters most, improving decision efficiency and potentially reducing the time needed to identify a suitable combination for later phases.
Flexibility Across Modern Dose-Finding Designs
A key strength of this approach is its broad applicability. While evaluated using the PIPE design and two-dimensional Bayesian logistic regression model (BLRM), the method itself is not tied to any single model.
Instead, it can be applied to:
- Model-based designs
- Model-assisted designs
- Other adaptive dose-finding frameworks that estimate toxicity surfaces
This flexibility allows sponsors to integrate the insertion strategy within their existing methodological toolkit rather than requiring a complete redesign of their dose-escalation approach.
Enabling More Informed Adaptive Trials
As combination therapies become more prevalent, the risk of selecting initial dosing grids with no adequate candidates for later phases may become increasingly apparent. Methods that enable controlled, statistically grounded adaptation offer a way to preserve rigor while enhancing learning within the trial.
By introducing a contour-guided rule for inserting new dose combinations, this approach provides:
- A transparent decision framework
- Better alignment with target toxicity
- Maintained safety and regulatory defensibility
Ultimately, it supports more informed and efficient early-phase development, helping sponsors move forward with greater confidence in the dose combinations they advance.
This work reflects the growing importance of sophisticated adaptive and model-informed methods in early-phase oncology development. At Phastar, our statisticians specialize in complex adaptive dose-finding strategies designed to improve decision making while safeguarding patient safety.
References
- George, M., Wadsworth, I., & Mozgunov, P. (2026). A novel method for inserting dose levels mid-trial in early-phase oncology combination studies. Statistics in Medicine. Advance online publication. https://doi.org/10.1002/sim.70417
