Taking on an Adaptive Dose Response Model

By Allison Proffitt 
October 31, 2013 | Earlier this month, the European Medicines Agency (EMA) issued a qualification opinion on a statistical and modeling methodology for determining dose in clinical development. 
The Multiple Comparison Procedure – Modelling (MCP-Mod) assists with both the design and analysis of dose-finding studies. In its draft opinion (EMA is accepting public comments on the opinion through November 24), the Agency’s Committee for Human Medicinal Products (CHMP) considers that this methodology could promote better design and analysis of trials that can incorporate a wider dose range and an increased number of dose levels. 
Andy Grieve agrees. Grieve is SVP, Clinical Trial Methodology for Aptiv Solutions and has been a statistician for over 40 years. He is a proponent of the methodology and has developed a toolset for Aptiv Solutions to facilitate dose selection according to the MCP-Mod. 
Determining the right dose for drugs can be tricky. At low doses potential drugs don’t have much effect, then there is usually a steeply-ascending portion of the curve, Grieve says. Finally the effect tapers off as a biological target is saturated, for example. “We understand that in biology we get these non-linear dose response models,” Grieve says, but modeling the non-linear curves—and designing trials around them—can be costly and time consuming. 
The EMA calls estimating dose-response and selecting the appropriate dose for Phase III trials one of the most difficult challenges of the drug development process. Currently half of Phase III trials fail, in part because of sub-optimal and inefficient practices for dose selection in Phase II. 
Current approaches for estimating the target dose are also prone to uncertainty, which can result in the use of an inappropriate statistical model that over- or under-estimates the true effective dose.
“We would typically take two or three doses of a drug in Phase II and compare it to the control,” Grieve says. “And it’s very difficult if you only have two or three doses to truly estimate the dose response curve. So you need to start to think about increasing the number of doses.” 
Wider ranges of doses are more likely to yield positive outcomes than very narrow ranges, Grieve explains. “But if you then remain wedded to a statistical process of hypothesis testing in which you compare each individual dose back to the control, you end up with trials that are much too big, that can’t be run.” 
The alternative, Grieves says, is to reduce the number of patients in individual dose groups, but increase the dose groups. 
“The better way is to start off with multiple doses, randomize patients to control and a range of multiple doses. Start to learn about the shape of the dose response curve, and then preferentially randomize patients to those doses which are of interest to you. If you see that by 50 mg, the dose response curve is starting to ascend, you might want to restrict yourself to doses above 50 mg. If you started with doses at 5, 10, and 25 mg, you’ll either down-weight the number of patients randomized to those doses or drop them completely.” 
The MCP-Mod approach is unique in that it defines several plausible candidate dose-response models, tests them for significance, and then identifies the most appropriate statistical approach to model dose-response and estimate the target dose. 
In the extensive simulations reviewed for the qualification opinion, the EMA found that the MCP-Mod methodology appears to be better than the common pairwise ANOVA-based approach in terms of bias and absolute error. 
“ANOVA approaches work well if the true dose response model is linear, but many drugs do not behave this way,” says Grieve. “This is a significant shift for the industry. The EMA opinion challenges the way most dose finding studies are performed, essentially saying that the number of doses and the dose range selected are sub-optimal. This leads to poor selection of the effective dose at Phase II, which causes significant problems in subsequent clinical trials.” 
MCP-Mod is a core component of Aptiv Solutions’ ADDPLAN DF software. “Up until now, the only availability of this software was in some freeware, but it wasn’t fully validated to regulatory standards,” explains Grieve. “Until we made available our software, sponsors would have to validate their own versions of that freeware. Now we have out there in the marketplace, software that meant to be used to design and analyze dose-finding studies fully with this methodology.” 
ADDPLAN DF is the only validated commercial software available for this methodology, according to the company. ADDPLAN DF is a fully-validated design and analysis tool based on MCP-Mod and covers dose finding designs for establishing Proof of Concept and determining the optimum target dose to select for Phase III trials.
“The original methodology was developed by statisticians at Novartis in Switzerland [and was published in 2005], and we’ve been working with them on the development of the software package,” Grieve explains. He says, “only a handful of sponsors” have used the methodology, but he hopes that the EMA’s opinion will expand its use. 
Aptiv continues to work with Novartis to develop new methodologies. The fully validated version launched in September. Two additional top 10 pharma customers are now using the software as founding members of the ADDPLAN DF Consortium the company launched earlier this year. 


Add Comment

Text Only 2000 character limit

Page 1 of 1

For reprints and/or copyright permission, please contact Angela Parsons, 781.972.5467.