New Generation Clinical Trials Could Save Time And Money, Improve Patient Care

"We need to rethink how we design and conduct clinical trials in the United States," says Berry. "Our current system has served us well for the past 50 years, but the demands of 21st century medicine are beginning to put a strain on the current system, and we believe we have something to relieve that strain."

Berry outlines his approach to conducting clinical trials in the January 2006 issue of Nature Reviews Drug Discovery. In the article, he advocates turning the statistical method used to evaluate new drugs on its head. He states that the statistical method used nearly exclusively to design and monitor clinical trials today, a method called frequentist or Neyman-Pearson (for the statisticians who advocated its use), is so narrowly focused and rigorous in its requirements that it limits innovation and learning.

His solution, which he has advocated for more than 30 years, is to adopt a system called the Bayesian method, a statistical approach he says is more in line with how science works. He sites examples of Bayesian approaches being used routinely in physics, geology and other sciences. And he is putting his approach to the test at M. D. Anderson, where more than 100 cancer-related phase I and II clinical trials are being planned or carried out using the Bayesian approach. The main difference between the Bayesian approach and the frequentist approach to clinical trials has to do with how each method deals with uncertainty, an inescapable component of any clinical trial. Unlike frequentist methods, explains Berry, Bayesian methods assign anything unknown a probability using information from previous experiments. In other words, Bayesian methods make use of the results of previous experiments, whereas frequentist approaches assume we have no prior results.

"Using the Bayesian approach, it is natural to do continuous updating as information accrues," says Berry. "This characteristic makes it possible for us to build adaptive designs in clinical trials."

He argues that the Bayesian approach is better for doctors, patients who participate in clinical trials and for patients who are waiting for new treatments to become available.

"Doctors want to be able to design trials to look at multiple potential treatment combinations and use biomarkers to determine who is responding to what medication," says Berry. "At the end of the day, when they enroll the last patient in the study they want to be able to treat that patient optimally depending on the patient's disease characteristics. Using a Bayesian approach, the trial design exploits the results as the trial is ongoing and adapts based on these interim results. That kind of thing is an anathema in the standard approach."

However, Berry argues, such flexibility is crucial to clinical trials in the 21st century. "The advances of the 20th century have taught researchers that cancer is a diverse disease, and what works to treat one person's disease may not work for another," he says. "In order to have the kind of personalized medicine the 21st century will demand, it will be necessary to be more flexible in how we evaluate potential new treatments."

Of course, the most important factor in whether the Bayesian approach will gain acceptance in clinical trials reporting is whether the U. S. Food and Drug Administration will accept Bayesian approaches in making determination of safety and efficacy of new treatments. Berry says progress is being made both at pharmaceutical companies and at the FDA in bringing regulators up to speed on the Bayesian approach. "Our biggest challenge is to convince the regulators that we are not throwing the baby out with the bathwater by using a Bayesian approach," says Berry. "It is rigorous and we are not losing science by using it." For example, the FDA has approved the Bristol-Myers Squibb drug Pravigard Pac for prevention of secondary cardiac events based on data evaluated using the Bayesian approach.

In addition, Berry says, it is possible to reduce the exposure of patients in trials to ineffective therapy using the Bayesian approach.

For example, in adaptive clinical trials, if interim results indicate that patients with a certain genetic makeup respond better to a specific treatment, it is possible to recruit more of those patients to that arm of the study without compromising the overall conclusions. Moreover, using the Bayesian approach may make it possible to reduce the number of patients required for a trial by as much as 30 percent, thereby reducing the risk to patients and the cost and time required to develop therapeutic strategies.