A regression based phase I clinical trial for late-onset toxicities without clinician elicitation

Abstract

An extension of the isotonic regression based phase I clinical trial design is presented that incorporates partial follow-up times into estimation of the raw toxicity probabilities. This phase I clinical trial design, called the TITE-IR design, drastically decreases average trial duration by allowing patients to be treated immediately after being enrolled in a phase I clinical trial. The TITE-IR design does not require specification of a prior skeleton of toxicity probabilities like the continual reassessment method, has an additional trial parameter for controlling aggressiveness of dose escalation, and has an easily understood formula for estimating toxicity probabilities. An R statistical software package is described in detail in the appendix for simulating and implementing the design. A simulation study shows that the TITE-IR design outperforms the 3 + 3 design in terms of selecting the true maximum tolerated dose and results in shorter trial times, without a large loss in efficiency, compared to the isotonic regression design and Storer's up-and-down design D. These properties make the TITE-IR design a more appealing option to clinicians than the two most commonly used 3 + 3 designs and the isotonic regression design with larger follow-up windows for toxicity.

Keywords: Clinical trials; Dose-finding; Isotonic regression; Phase I; Time-to-event.

Fig. 1

Simulation Study: Dose selection probabilities in each scenario. Prob Toxicity denotes the true toxicity probability for each dose and * denotes the dose toxicity probability closest to the target of 1/3 without exceeding 1/3.