A modeling project was undertaken to improve efficiency and informativeness of Phase II trial strategies, and to optimize
the treatment regimen. First, drug and disease models were developed to quantify expectations and uncertainties about efficacy
and side effects. This formed the basis for a penalized dose-response model, developed to provide a utility criterion for
choosing best dose. (See "Dose Benefit Model".)
A model of various dosing strategies was developed. Simulations assessed each design's performance compared to fixed-dose
dose-ranging. The criteria for trial quality were: closest estimate of best dose (Dbest) and minimum effective dose (MED),
as measured using the penalized dose function (utility index). The simulations indicated an advantage to particular designs
because the uncertainty in safety and efficacy was understood and a utility index provided a single, quantitative measure
of that best dose.
Case Study: Identifying No-Gos Evaluating a potential next-in-class drug against successful predecessors requires close comparison of risk-benefit profiles.
A CUI is ideal for the job.
A new selective estrogen receptor modulator (SERM) was in early human trials. Concern had risen over a possible side effect,
endometrial hypertrophy, which could lead to cancer if unchecked. Further trials posed not only financial risk but possible
health risks for hundreds of women if this side effect proved troublesome. To explore whether any dose of the new candidate
could be expected to perform acceptably against the popular first-in-class drug, Evista (raloxifene), Aventis Pharmaceuticals
turned to modeling and simulation.
Dose-response models for efficacy versus side effects were developed from early trials of the new SERM and prior information
on related compounds. An initial version of the CUI was elicited from project team members. They identified 10 critical attributes
of the product profile. For each attribute, possible attribute levels and their clinical value were defined using a preference
ratio. The attributes were ranked and their importance weighted. The team reviewed the attributes to ensure that they described
all important and relevant clinical issues, captured the range of outcomes for each attribute, and reflected the expected
clinical value of possible outcomes.
Simulation produced a distribution of likely patient outcomes for each attribute. Each distribution was parsed into categories,
each of which had a known likelihood of occurrence. Every unique combination of attribute levels yielded one possible CUI
score. The probabilities of the attribute levels for a given CUI score yielded the likelihood of that score being the "true"
product CUI. The combined set of possible CUI scores and their likelihoods were characterized as a probability distribution,
describing the expected CUI values for the new SERM and its primary competitor.
Simulations revealed that the new SERM, as reflected in its CUI distribution, was expected to perform worse than the established
competition at all doses. Sensitivity analysis made it clear that the drug could have performed well except for the key side
effect; an identical drug without the risk of endometrial proliferation would surpass the competition. The upcoming trial
would have been quite expensive and those funds could be used more effectively on other compounds. Consequently, it was decided
to progress a more promising back-up compound that did not appear to have this liability in preclinical testing.
In early drug development, a CUI makes optimal use of internal expertise and minimizes the cost of focusing development on
the best product profile—not the most efficacious one. By using comparisons to current competitors and new entries under development,
drugs that have no differentiation can be weeded out earlier from a crowded marketplace, providing more resources for R&D.
It can substantially lower the cost of delivering a beneficial drug to the marketplace.
Additional benefit is gained by the use of modeling and simulation analyses in Phases II and III. By investing in these more
intensive and adaptive early phase analyses, developers can spot poor drug candidates earlier and improve positioning of those
going into Phase III. The use of a CUI can optimize the trade-offs in speed, cost, and learning during these phases, leading
to quicker wins. The end result is a label that supports the best therapeutic use of a new therapy, and offers the best patient
benefit. Thus, the CUI is a valuable communication, decision making and optimization tool.
Bob Korsan (email@example.com
or (650) 314-3843) is director of decision services, and Kevin Dykstra is a senior scientist, both for Pharsight. Dr. William Pullman is senior vice president and global head of clinical discovery and human pharmacology for Aventis Pharmaceuticals.
The authors would like to acknowledge Shawne Neeper and T. J. Carrothers (both of Pharsight) for their efforts in developing