|
Abstract:
|
The traditional Phase 1 design is the 3+3 Up-Down design where cohorts of 3 patients are accrued. 2 or more dose-limiting toxicities (DLTs) in the first 6 require de-escalation and 0 DLTs out of 3 or 1 DLT out of 6 patients permits escalation if possible. Newer designs require decisions outside the comfort of the traditional 3+3 decision grid. To help address this issue, we reverse engineered the logic of the 3+3 design based on two beta priors and two decision thresholds (six parameters, a1, b1, a2, b2 and T1, T2). The two thresholds were for escalating or de-escalating. We sampled this six dimensional space to find sets of parameters that matched the 3+3 rules. These rules were applied to test cases. For example, with 2 DLTs out of 7 patients 57% of the constructs suggested "Stay", while 43% suggested de-escalating. However, all de-escalating constructs had a DLT rate threshold less than 1/6. As a result, if 1/6 is an acceptable DLT rate, all such constructs result in a "Stay" decision. Such analysis can help guide physicians with non-traditional decisions for dose escalation, de-escalation or continuing to accrue at the current dose.
|
