Improving Maximum Tolerated Dose Selection in Model-Assisted Designs for Phase I Trials Through Bayesian...

Area of Science:

• Clinical Trials
• Biostatistics
• Pharmacology

Background:

• Model-assisted designs are popular for identifying the maximum tolerated dose (MTD) due to accuracy and simplicity.
• Current methods using isotonic regression (PAVA) can yield unstable estimates in small sample sizes, common in Phase I oncology trials.
• This instability can reduce the accuracy of MTD selection.

Purpose of the Study:

• To propose a novel MTD identification strategy for model-assisted designs using a Bayesian dose-response model.
• To improve the stability and accuracy of dose-limiting toxicity (DLT) probability estimation.
• To enhance the overall efficiency of Phase I oncology trials.

Main Methods:

• A Bayesian dose-response model was developed to estimate DLT probabilities while enforcing monotonicity (toxicity increases with dose).
• The model leverages information across dose levels for stable estimation, particularly in small sample settings.
• Different link functions (logit, log-log, complementary log-log) were explored to assess their impact on MTD selection accuracy.

Main Results:

• The proposed Bayesian approach demonstrated improved MTD selection accuracy compared to conventional methods.
• Accuracy improvements reached over 10% in specific scenarios and averaged approximately 6%.
• The method provides more stable DLT probability estimates, especially crucial for small sample sizes.

Conclusions:

• The novel Bayesian dose-response model offers a more accurate and stable method for MTD identification in Phase I oncology trials.
• This approach enhances the efficiency of early-phase clinical trials by improving MTD selection.
• The findings support the adoption of this strategy for optimizing trial design and patient safety.