Bjoern Bornkamp, Statistical Methodologist at Novartis and Jose Pinheiro, Senior Director, Johnson & Johnson provided their insights on adaptive designs for dose finding in Cytel’s latest webinar. Finding the right dose in Phase 2 gives a potential new therapy its best chance to demonstrate efficacy during Phase 3. A well-executed dose-ranging trial therefore has the potential to alter the course of the entire clinical development program. This webinar demonstrates how adaptive and Bayesian techniques can be implemented for optimal dose-finding.
This two-part blog series will provide a summary of the webinar. In this first part, get key highlights from the presentation by Jose Pinheiro on the need to conduct dose finding Phase 2 studies, dose selection comparisons and the use of MCP-Mod for dose finding. Click the button to access the webinar recording and download the presentation slides
Motivation to conduct dose finding phase 2 studies
Improper understanding of dose response, resulting in inadequate dose selection for Phase 3, remains a critical problem in drug development. One common reason why dose selection is suboptimal in Phase 3 is that the Phase 2 studies, intended to be exploratory in nature, are generally designed as mini Phase 3 studies, using pairwise comparisons to control for dose selection. Also, only few doses are evaluated, often with similar efficacy and covering relatively small dose range. As a result, poor dose selection leads to delays and even denial in regulatory approval, and possibly prevents patients’ access to effective drugs.
Dose finding Phase 2 studies are conducted with two main goals. One is to establish the proof-of-concept, that is to identify evidence of dose response (DR) signal on efficacy. Once we know that the drug generates a clinical response, the next aim is to the decide which dose(s) needs to be brought forward to a confirmatory stage. This is a critical question that is often not properly addressed. There are different ways in which we can specify a target dose, for example, minimum effective dose (MED) which is the smallest dose that leads to a clinically relevant effect, or ED90 which is the dose that produces 90% of maximum possible improvement over placebo.
Dose Selection Comparisons
There are the two broad classes of methodologies and designs that are used for these studies:
Pairwise comparisons between doses and placebo (accounting for multiplicity) with dose selection driven by hypothesis testing;
Modeling of DR relationship(s), with dose selection driven by model-based estimation (e.g., dose corresponding to minimum clinically relevant effect.)
Pairwise comparisons approach is more widely used as there is hope that if successful, it can be one of the confirmatory studies that can be used for submission. The focus is on control of multiplicity as there is little incentive to consider more than two or three active doses. The sample size determination is based on the probability of having at least one pairwise comparison be statistically significant.
The Model-Informed Dose Selection method on the other hand, focuses on estimation of doses with certain characteristics. The design of the study should account for estimation of dose response and target dose. The sample size does not need to increase with number of doses as it does in pairwise comparisons. You can make more efficient use of trial data by modeling response as a (continuous) function of dose (exposure), and it can be used with both efficacy and safety endpoints. However, the caveat is that it typically requires pre-specification of the dose-response curve’s shape which is often not known. Also, this method is generally not accepted as pivotal study.
MCP-Mod for dose finding
MCP-Mod is a unified dose finding approach that is designed to enable testing for the presence of a dose-response signal. It combines MCP and modeling for model-informed dose selection (and DR estimation) under model uncertainty. Once we have the data we test for a dose-response signal using contrast tests. If we can identify that, we move on to model selection out of the set of significant models. The next step is to estimate the dose-response and target dose, based on the selected model(s).
This method can handle most common types of responses that we use in clinical trials, such as, continuous, binary, count, time-to-event, etc. One of the advantages of this approach is that it can be implemented with software like DoseFinding R package, East MCPMod module (Cytel), ADDPLAN DF and SAS Macro (also PROC MCPMod). There are also SAS implementations available for this method. MCP-Mod received a qualification opinion from the European Medicines Agency as an efficient statistical methodology for model-based design and analysis of phase 2 dose finding studies under model uncertainty. A similar fit-for-purpose determination of MCP-Mod was received from the FDA in 2016.
The next part of this blog series will focus on Adaptive Designs for Dose Finding presented by Bjoern Bornkamp. Click the button to access the on demand webinar.
This webinar is a part of Cytel’s “Introduction to Complex Innovative Trial Design” webinar series. It aims to introduce clinical fellows, early career biostatisticians, and others interested in clinical research, to some of the more commonly used complex innovative trial designs. Register to attend the upcoming webinar with Nitin Patel, Co-founder of Cytel, on designing clinical trials from a program perspective.