“Save a Disease” – perhaps not the greatest of titles – is the working internal designation for an FDA prototype program to create a library of drug/disease models that could eventually be used by industry to simulate and design clinical trials. But don’t start looking for an Internet repository of FDA models just yet.
“We need to have dialogue inside the agency on what people are comfortable with. It’s not like we can just create a website and start piling data there,” says Bob Powell, one of the project’s backers and director of FDA’s division of pharmacometrics inside the Office of Clinical Pharmacology and Biopharmaceutics (OCPB).
Click here for a free
Bio-IT World briefing
on Systems Biology
The FDA is the largest storehouse of clinical trial data in the world and perhaps the largest storehouse of information on drug/disease interactions. The prospect of mining this data, much of which is proprietary information submitted by sponsors, is tantalizing and was mentioned in FDA’s 2004 Critical Path Initiative.
Powell’s proposal wouldn’t give out proprietary data, but would make available models extracted from FDA’s massive data cache, and permit drug companies to use the models to simulate and design new trials. “We’re currently working on two diseases to prototype with,” says Powell, declining to identify them as yet.
He envisions the models – both mechanistic and empirically derived – to be a kind of knowledge repository of drug/disease interactions for particular disease. “Say someone wants to do a trial in Parkinson’s disease, he could look at a library of information extracted from prior experience. It wouldn’t be the data but it would look like a kind of quantitative summary.”
Modeling is not new at FDA. For the last five to seven years, “people here have started using modeling and simulation for reviewing NDAs. That’s been primarily done in cardio-renal, oncology, and CNS, amongst other diseases," says Powell. “I wouldn’t say there’s a uniform view [of modeling’s value] in the FDA, but there are people throughout the organization, including some senior people like Janet Woodcock, with a lot of hope for what can be done.”
So far, modeling and simulation technology has had more impact on what gets onto the label than on go/no-go approval decisions. Powell cites dosing regimes and post-market trial commitments as two areas in which simulation has resulted in changes from sponsor recommendations.
Recently, an internal study of 42 NDAs submitted between 2000 and 2004 for cardio-renal, neuropharmacology, and oncology products revealed that pharmacometrics (the agency’s umbrella term for modeling and simulation) was involved in approval issues on 26 of the NDAs and involved on label-related issues on 37 of the NDAs.
“We started to expand the use of modeling about a year and a half ago with what we call ‘end of Phase IIA’ meetings,” says Powell. The goal was to help avoid Phase III failures – the costliest of all drug development stages – and also to improve FDA productivity. “Equivocal results, things that are marginal, can eat up a lot of FDA time,” for trials that ultimately fail, he says.
Now, FDA may ask if a sponsor wishes to meet after the first Phase II trial, which is when more is known about the drug, but the sponsor hasn’t solidified Phase III plans. FDA asks for the trial data, a summary of the project, and the anticipated next protocol. “We’ll do either mechanistic or empirical modeling, and we simulate the next trial,” Powell says.
The actual meeting isn’t binding. “It’s really a consultation,” says Powell, and FDA gives the sponsor “everything we do,” including all the code for the simulation. And FDA encourages the sponsor to extend the simulation as the project moves forward.
“The other thing we do, because we have data from other drug trials, is we’ll take the model before we apply it to the sponsor’s trial and run it against other data to see if it predicts what was found elsewhere. We don’t share that with the sponsor that we’re dealing with,” says Powell.
The agency is “completely opportunistic” in obtaining modeling technology. For an HIV model, it combined the work of a prominent New York AIDS clinic, taken from the literature, with work from Los Alamos National Laboratory. That same model has since been used by Abbott, Pfizer, and a number of others, says Powell, whose own experience includes work at Big Pharma as well as with model-maker Pharsight. He also cites work by Entelos on FDA efforts to develop a hepato toxicity model.