Predictive Modeling Needs Drug-Trial Data

Opinion: Predictive modeling can be a force for the common good, and it requires vast volumes of data-some of it proprietary-in order to be effective.

Hospitals would like to peer into the future, and the Centers for Disease Control and Prevention is making these glimpses possible, in a sense. It has developed software programs that help hospitals and other health care agencies predict how many staff, beds and supplies a hospital will need during a flu pandemic.

Though the programs were designed for U.S. health planners, a senior health economist at the CDC reports that several other countries have used the software in planning their response to an influenza pandemic.

/zimages/3/28571.gifCopies of the programs, called FluAid and FluSurge, are available for free from the Department of Health and Human Services National Vaccine Office.

This software is a straightforward example of how predictive modeling can be a force for the common good. The FDA is also moving into the prediction realm for the benefit of public health, but its route will be convoluted.

Effective predictive modeling requires vast volumes of data, and government Web sites are brimming over with public health data. In drug development, however, data are proprietary and sensitive and so can be accessed only by government mandate. And often the government mandate is to keep the data secret.

Earlier this year, the FDA announced a new effort, called the Critical Path Initiative, to promote predictive technologies in drug development. The announcement was a bold move for a rather stodgy agency, but it hopes to tackle the growing cost and static productivity of drug development.

Despite all of the new technologies and the rocketing price of drug development, a drug entering clinical trials today has no better chance of getting to the market than did a drug entering trials more than a decade ago.

The FDA feels it can improve the situation because of the vast volume of confidential data drug developers bring before the agency. Some of these include nonhuman data indicating that the drug would help more than it hurts.

Different drug companies depend on different measurements for such predictions: One company may examine a suite of 15 genes to gauge a particular risk; another might put its faith in a panel of seven proteins. But of course, the only way to know for sure if these indirect, fortune-telling measures are accurate is to test the drug in people and see if the predictions hold true.

In its quest, the FDA must perform a juggling act with its hands tied. The agency may well know whether the suite of genes or the panel of proteins is better at predicting patients responses, but it also must protect companies intellectual property. And it also absolutely must protect patient safety. And its supposed to help make the overall drug development process more efficient, too. (Patient safety trumps the others when attempts to avoid a clash fail.)

The solution? The FDA isnt quite sure, but by articulating the problem as a failure of science—rather than a failure that can be met by tweaks in the regulatory process—the agency has made a huge step in the right direction. It calls for more computational biology and in silico modeling, but these efforts require large quantities of high-quality data, which drug companies are loathe to reveal.

/zimages/3/28571.gifClick here to read more about the Critical Path Initiative.

For the Critical Path Initiative, the FDA is collaborating with the industry and actively soliciting its input. But other moves are under way to force drug companies to publicly disclose data from certain late-stage clinical trials publicly.

GlaxoSmithKline and Forest Pharmaceuticals have both been accused of fraud for failing to disclose results showing that their antidepressants were ineffective while publicizing other results showing that they worked.

Around the same time, a group of medical-journal editors threatened not to publish results of clinical trials unless the trial sponsor had ensured both favorable and unfavorable results could be disclosed. The American Medical Association called for the creation of a national database in which drug companies would have to submit clinical-trial results for marketed products.

Patients and physicians want to see trial results to be more confident that they are using the most appropriate drug currently available. And the biomathematicians and biostatisticians who create and use software to model clinical trials say a database of clinical-trial results could be used to predict the effects of combinations of drugs, optimize dosing, and help prove the safety and efficacy of experimental drugs faster.

/zimages/3/28571.gifClick here to read more about these possibilities.

Last week, three Democratic senators asked the FDA and National Institutes of Health what resources—financial and legal—they would need to create a national database of clinical trials, and the World Health Organization said it was investigating plans to create a worldwide version.

Of course, the requirement for publicly disclosed data would only be a boon if drug companies continue to collect that data in clinical trials. Disclosure mandates could encourage drug companies to learn only the bare minimum necessary to get a drug approved for market. After all, drug companies have paid top dollar to test their drug against those of competitors and are not keen to reveal results that favor a competitor.

To truly foster the power of prediction, the government will need not only a mandate to make data public, but also the means to encourage companies to collect data in the first place.

/zimages/3/28571.gifCheck out eWEEK.coms Health Care Center at for the latest news, views and analysis of technologys impact on health care.