Dose-Escalation Trial Protocols that Extend Naturally to Admit Titration

Abstract

Dose-escalation trials in oncology drug development still today typically aim to identify 1-size-fits-all dose recommendations, as arbitrary quantiles of the toxicity thresholds evident in patient samples. In the late 1990s efforts to individualize dosing emerged fleetingly in the oncology trial methods literature, but these have gained little traction due to a nexus of conceptual, technical, commercial, and regulatory barriers. To reduce the activation energy needed for transforming current 1-size-fits-all dose-escalation trial designs to the dose-titration designs required for patient-centered dose individualization, we demonstrate a categorical formulation of dose-escalation protocols that extends readily to allow gradual introduction of dose titration. Central to this formulation is a symmetric monoidal preorder on the accessible states of dose-escalation trials, embodying pharmacologic intuitions regarding dose-monotonicity of drug toxicity and ethical intuitions relating to the therapeutic intent of such trials. A trial protocol that assigns doses to enrolling participants consistently with these intuitions is then a monotone map from this preorder to the sequence of doses being trialed. We illustrate this formulation by reference to the ubiquitous 3+3 dose-escalation design, which despite its many widely discussed flaws remains familiar to oncology trialists and moreover has available an executable specification in Prolog. Remarkably, examined in light of our preorder the 3+3 protocol discloses a new flaw not previously described: a non-monotone dose recommendation. The right Kan extension approximates this protocol from the side of safety, dissolving its triplet cohorts to allow incremental enrollment, and rectifying said non-monotonicity. It also facilitates accelerated enrollment while toxicity assessments remain pending, as well as discretionary dose titration.