Pavel Mozgunov; Andrew Hall; Lizzi Pitt

This session is a collection of presentations considering approaches to dose finding in early phase trials.

Practical Implementation of the Partial Ordering Continual Reassessment Method in a Phase I Combination-Schedule Dose-Finding Trial  - Pavel Mozgunov 
There is a growing medical interest in combining several agents and optimising their dosing schedules in a single trial. Evaluating doses of several drugs and their scheduling in a single Phase I trial simultaneously posses a number challenges, and specialised methods to tackle these are required. While several suitable designs were developed and proposed in the literature, the uptake of these methods is slow and implementation examples of such advanced methods are still sparse.  
 
In this presentation, we will share our recent experience of developing and implementing a modified model-based Partial Ordering Continual Reassessment Method (POCRM) Design for 3-dimensional dose-finding in a Phase I oncology clinical trial in patients with advanced solid tumours. In the trial, doses of two agents and the dosing schedule of one of them can be escalated. We will provide a step-by-step overview on how the POCRM design was implemented and communicated to the trial team. We will demonstrate a novel approach to specify the design parameters that is more intuitive for communication and will demonstrate a number of developed visualisation tools to demonstrate the statistical properties of the design. This included both performance in a comprehensive simulation study and in individual scenarios. The proposed design went through evaluations of health authorities and was successfully used to aid the decision-making in the ongoing trial.  
 
Decision making under uncertainty in PI-II dose finding trials in Oncology - Andrew Hall 
There is increased interest in dose finding methods in oncology using both toxicity and efficacy endpoints with targeted therapies. A phase I trial design proceeds in stages with a decision as which dose to give the next group of patients made after every stage. Bayesian decision theoretical approaches have previously been found to be in theory ethically and scientifically sound. In practice however, it is challenging to specify a utility function capturing clinical preferences while maintaining good operating characteristics sensitive to specification.  
 
Outcomes from treatments are not deterministic; utilities are a measure of preference when facing an uncertain outcome. We consider situations where preferences/utilities are defined with respect to outcome probabilities. In doing so clinicians can account for individual patient risk while meeting wider trial objectives, i.e. identifying a recommended phase II dose. We argue attitudes to risk in this setting follow heuristics from prospect theory. Namely they are framed from the perspective of a reference point, with a risk averse attitude for perceived gains, and risk seeking for losses. Additionally, with loss aversion it is ethical to avoid losses more so than to pursue gains. The bivariate utility is formed by inspecting utility independence axioms to describe the payoff between two separate utility functions for efficacy and toxicity.  
 
I will explain why heuristics from prospect theory to structure utilities around outcome probabilities are justified in dose finding trials, and show this leads to consistent and in some scenarios improved operating characteristics over designs specifying value rather than utility functions.  
 
Building the bridge from PhD to practice: optimising phase I trials using estimand-style formulation - Lizzi Pitt 
We have developed a framework to obtain optimal dose escalation schemes for phase I trials. The emphasis is on fully specifying the aims of the trial up front: if you tell us what you want the trial to do, we can find the optimal dose escalation scheme for your specific trial. We achieve this using dynamic programming. We have considered trials with one binary safety endpoint with a variety of aims, as well as trials that also include a binary efficacy endpoint.  
 
This research was conducted as a PhD project and focussed on a first in human trial with a particular generic structure. How do we translate this theory into practice and apply the methodology to a real trial, with a different structure?  
 
We shall present reflections on some case studies of real phase I trials with different structures and covering different therapeutic areas. Spoiler alert! The key is in the problem formulation. This framework encourages early discussion on what makes the trial a success, what quantity the trial seeks to estimate and how the information will be used in phase II. This brings the flavour of estimands to phase I trials and creates trial designs that are fit for purpose, facilitate decision making and enable us to learn more about the treatment earlier.