Aug 15, 2005 | It’s no secret that drug development costs are painfully high. At Pfizer there are often 75 to 150 new drug candidates in clinical development at any time, with out-of-pocket trial costs for initial proof-of-concept studies averaging $24 million and costs for large-scale efficacy tests often reaching $84 million. Quickly identifying and ending projects that are unlikely to succeed is a high priority.

Category: Clinical Research & Trials Title: Strategic Application of PK/PD Modeling, Simulation, and DMX to Aid Decision Making Company: Pfizer Global R&D Partner: Pharsight Corp.

In 2003, Pfizer Global Research & Development began using sophisticated modeling technology and collaboration software from Pharsight Corp. to speed its race to a go-no-go decision on a cholesterol-lowering drug project. Ultimately, the use of modeling data helped Pfizer kill the project sooner, saving $1.4 to $2.8 million in costs and four to six months in delay.

“Pharmacological modeling and biosimulation have proven quite useful in preclinical and clinical development, and their use is continuing to grow at Pfizer,” says David Hermann, former director of pharmacometrics, Pfizer Global R&D, and one of the project’s leaders. (Hermann left Pfizer in July, along with Daniel L. Hartman and Peter Van Ess, to join the drug development team at Iceland’s deCODE genetics.) One result, according to the Pfizer entry, is that investigators involved in this particular project have indicated they are likely to use a similar approach on future projects.

	COLLABORATIVE TEAM WORKFLOWS USING DRUG MODEL EXPLORER (DMX): This illustrated workflow provides a context for the interplay between drug model building and simulation, visualization of simulated product attributes, and uncertainty using DMX, and DMX-supported communication processes with project teams, modeling experts, and senior decision-makers.
	CLICK FOR LARGER VIEW >>

Modeling does seem to be gaining traction. “Last year, Pharsight delivered over 100 projects to companies ranging from top-three pharma to biotechs and emerging small molecule companies,” says Bill Gillespie, VP, drug development consulting services, Pharsight. “Modeling and simulation is especially useful where the drug candidate will face intense competition within its therapeutic class.”

The Pfizer experience is instructive. The company was seeking to learn both if its new gemcabene-statin combination therapy was effective and if it would be sufficiently competitive with existing ezetimibe-statin combination therapies to justify further investment. Preliminary safety and efficacy tests showed evidence of benefit without unacceptable side effects, but the therapy had not yet been tested in its target population.

Pfizer decided to run a single, small Phase II trial to measure gemcabene’s performance and also to undertake a more comprehensive model-based analysis to evaluate gemcabene-statin combination therapy versus the competition.

Historical Models
The past two decades have seen a significant expansion in the application of mathematical modeling and simulation techniques for quantifying clinical drug effects. Drug exposure-response analyses have been routinely applied to single-trial data. Pfizer’s team extended these models to fit data spanning multiple trials. A literature search provided competitor data from 21 randomized trials, covering more than 9,000 patients, through May 2003. The Pfizer models combined these data with internal Pfizer study reports and publicly available data from FDA submissions to generate a single, comprehensive map of the treatment landscape.

It took the equivalent of two full-time staff, working for approximately four months, to construct the first model, but only a week was needed to incorporate Pfizer’s recent trial data. “The model provided a more precise estimate of the dose-response behavior for gemcabene, with narrower uncertainty bounds than information from a single trial could provide. It also allowed data-driven comparisons among treatments that had not been compared in a head-to-head trial,” Pfizer noted in its entry.

To explore the simulation results, Pfizer used Drug Model Explorer (DMX), a software tool that had just been introduced by Pharsight. DMX is designed to make modeling and simulation more accessible to nonmodeler team members by supporting interactive visualization and communication of model-based product profiles. DMX includes a Microsoft Windows-based interface to allow team members to view and query these results interactively, and publishing tools share models.

Results from the new Phase II trial results arrived in July 2003. The data showed that gemcabene, alone or in combination with one statin, worked better than placebo at lowering cholesterol. It also gave an estimate of how well the treatments worked in patients with high cholesterol.

Within one week, the modeling team integrated the new study data with the dose-response models, generated new simulations, and republished the updated database to DMX for team exploration. The full modeling results, backed by data from multiple trials and displayed in DMX, made it clear to the team that gemcabene combination therapy would not provide superior cholesterol-lowering benefit versus ezetimibe combinations.