[Paper Review] Time-to-Event Model-Assisted Designs to Accelerate Phase I Clinical Trials

Ruitao Lin, Ying Yuan|arXiv (Cornell University)|Jul 23, 2018

Statistical Methods in Clinical Trials25 references4 citations

TL;DR

This paper proposes TITE-keyboard, a model-assisted phase I clinical trial design that simplifies decision-making under fast accrual or late-onset toxicity by tabulating dose escalation/de-escalation rules in advance. It uses a novel likelihood approximation to handle pending toxicity data, achieving performance comparable to complex model-based methods like TITE-CRM while being far easier to implement than existing alternatives.

ABSTRACT

Two useful strategies to speed up drug development are to increase the patient accrual rate and use novel adaptive designs. Unfortunately, these two strategies often conflict when the evaluation of the outcome cannot keep pace with the patient accrual rate and thus the interim data cannot be observed in time to make adaptive decisions. A similar logistic difficulty arises when the outcome is of late onset. Based on a novel formulation and approximation of the likelihood of the observed data, we propose a general methodology for model-assisted designs to handle toxicity data that are pending due to fast accrual or late-onset toxicity, and facilitate seamless decision making in phase I dose-finding trials. The dose escalation/de-escalation rules of the proposed time-to-event model-assisted designs can be tabulated before the trial begins, which greatly simplifies trial conduct in practice compared to that under existing methods. We show that the proposed designs have desirable finite and large-sample properties and yield performance that is superior to that of more complicated model-based designs. We provide user-friendly software for implementing the designs.

Motivation & Objective

• To address the challenge of delayed toxicity assessment and fast patient accrual in phase I trials, which hinder real-time adaptive decision-making.
• To develop a model-assisted design that maintains the simplicity of algorithm-based methods while matching the performance of complex model-based designs.
• To enable transparent, pre-tabulated dose-finding rules that do not require repeated model fitting during trial execution.
• To improve practical adoption of advanced designs by reducing implementation burden without sacrificing statistical efficiency.

Proposed method

• Proposes a novel likelihood approximation to handle pending toxicity data, enabling seamless integration of incomplete follow-up into decision rules.
• Adapts the time-to-event continual reassessment method (TITE-CRM) framework to allow for pre-tabulated dose assignment decisions.
• Uses a simplified, tabular decision rule based on observed and pending toxicity outcomes, eliminating real-time model fitting.
• Incorporates prior information on toxicity timing through weighting schemes (e.g., uniform or piecewise uniform), improving robustness.
• Employs a conservative rule to prevent dose escalation unless sufficient follow-up data are available, enhancing safety.
• Derives theoretical properties such as monotonicity, coherence, and consistency to ensure logical and statistically sound dose progression.

Experimental results

Research questions

• RQ1 Can a model-assisted design be developed that maintains high performance while being simple and transparent for clinical use?
• RQ2 How can pending toxicity data from fast accrual or late-onset events be handled without requiring complex real-time computation?
• RQ3 Does a tabulated decision rule based on likelihood approximation outperform existing algorithm-based designs like R6 in terms of accuracy and safety?
• RQ4 How robust is the proposed design to variations in toxicity timing, cohort size, and accrual rate?

Key findings

• The TITE-keyboard design achieves operating characteristics comparable to the more complex TITE-CRM, with superior ease of implementation.
• Simulation results show TITE-keyboard outperforms the R6 design in both accuracy of identifying the maximum tolerated dose and safety in patient allocation.
• The design is robust across 12 diverse simulation configurations, including varying toxicity onset patterns, cohort sizes, and accrual rates.
• TITE-keyboard demonstrates lower risk of overdosing patients and poorer allocation than the TITE-mTPI design, making it preferable for practical use.
• The performance of TITE-keyboard is minimally sensitive to the choice of weighting scheme for pending toxicity data, enhancing its practical robustness.