|
|
|
|
|
|
|
|
|
|
|||||||
 |
 |
 |
 |
|||||||||||
|
Biostatistically Speakingby David Galloway In biomedical research today, biostatistics is much more than static number crunching. At M. D. Anderson Cancer Center, biostatisticians are central to the analysis of complex biological problems and the design of the clinical trials of tomorrow. On a molecular level, cancer comes in a staggering number of forms. “The mutations required to create a cancerous tumor can occur in any of the large number of genes that regulate cell growth, programmed cell death, and DNA repair,” said Bradley M. Broom, Ph.D., an associate professor in the Department of Biostatistics and Applied Mathematics at M. D. Anderson Cancer Center. “Consequently, the number of possible combinations and permutations of cancer-causing mutations is enormous.” Examining all of those billions of possibilities requires considerably more computing power than a desktop computer can provide. To deal with the need for this type of advanced computation in cancer research, Dr. Broom and colleagues at M. D. Anderson have formed the Gulf Coast Center for Computational Cancer Research, a joint project with Rice University. Researchers in the center have access to a terascale cluster on the Rice University campus, which is a group of interconnected computers capable of executing one trillion floating-point operations per second. “So, say there’s a computer program that commonly takes a day to run, but our researchers want to test the feasibility of making it run in a few minutes so it could be used in a clinical situation. We can do that using the terascale cluster,” Dr. Broom said. “For instance, we have an application that provides four-dimensional magnetic resonance imaging scans of the lung. Currently, it’s too slow to be clinically useful, so we’re working to make it run faster,” he said.
The need for computing power grows exponentially, because each new variable added to a problem can more than double the amount of computer time needed to solve the problem. “So if you’re going to do something with five more variables, you may need 30 times as much computing power,” Dr. Broom said. Searching through databases of billions of genetic mutations is a daunting task, even for this kind of supercomputer. “So just searching faster is not going to help,” Dr. Broom said. “We need to be able to search ‘cleverer’ to identify the right signposts to look at as we’re searching this space to find something that is optimal or very good in a reasonable amount of time.” “And it has to involve the biologists who understand much more than we do about the way these things work and where the signposts might lead,” said Donald Berry, Ph.D., professor and chair of the Department of Biostatistics and Applied Mathematics. “So it really has to be a collaborative effort.” With the help of modern computers, M. D. Anderson’s biostatistics faculty and analysts collaborate with physicians and research scientists throughout the institution to provide advanced statistical analysis, mathematical modeling, and database development. Recently, such joint efforts have led to new statistical designs for clinical trials. “We’re building trials that, well, they’re not your mother’s clinical trials,” Dr. Berry said. “For instance, we’ve developed new trial designs using adaptive randomization. Instead of waiting until the end of the study to analyze results, we look at the data after partial enrollment and adjust the randomization algorithm accordingly, assigning a higher percentage of patients to the arm with the higher response rate. It’s a trial that adapts as we acquire more data.” In adaptive randomization, the beginning of the process looks just like it does in conventional randomization: equal numbers of patients are randomly assigned to each study group. But the similarity ends there. In particular, in the classic model, equal numbers of patients continue to be assigned to each study group throughout accrual, and the results are not analyzed until all patients have been treated. Adaptive randomization, on the other hand, adjusts the randomization probabilities to reflect the interim results of the trial. For example, if the response rate appears to be twice as high in one study group, twice as many patients will be assigned to that group as enrollment continues. Elihu Estey, M.D., a clinical investigator and professor in the Department of Leukemia, sees this approach to clinical trial design as corresponding more closely with the needs of practicing physicians and patients, while allowing for scientific accuracy. “Adaptive randomization allows patients to benefit from the knowledge we’re gaining as we get it, rather than years down the road,” he said. Dr. Estey has collaborated with colleagues in Biostatistics to develop new trial designs that compare several experimental drugs and monitor multiple outcomes in one study.
According to Dr. Berry, somewhere between 50 and 100 clinical trials, mostly phase I and phase II trials, are using adaptive randomization. Most of the clinical trials at M. D. Anderson are phase I and II trials, he said, which means the researchers have more latitude to try such innovative approaches. When it comes to phase III trials, though, the U.S. Food and Drug Administration is less flexible. However, Dr. Berry said the agency has recently approved the design of a phase III clinical trial using adaptive randomization. “This is really quite an achievement that they’ve gone along with this,” he said. Adaptive randomization is related to the field of Bayesian statistics. Dr. Berry explained the difference between classical and Bayesian statistics with the common statistical model of a series of coin tosses. “You toss a coin a hundred times. If you get 60 heads, you say the null hypothesis is that the coin is a fair coin and gives heads half the time. The P value is the probability of observing 60 heads or more,” Dr. Berry said. “That doesn’t address the question of ‘what is the probability that it’s a fair coin?’ The Bayesian approach addresses that specifically and, as data accumulate, allows you to update what that probability is.” Despite the use of supercomputers and complicated mathematics, the ultimate goal of these researchers is to accelerate the pace of progress in identifying effective new treatments for cancer. Eventually, Dr. Berry pointed out, all of these ideas that turn out to be successful will work their way outward from the laboratory to the clinic. “It’s then that the work we’re doing will really pay off,” he said.For more information on this topic or for questions about M. D. Anderson’s treatments, programs, or services, call askMDAnderson at (877) MDA-6789. Other articles in OncoLog, February/March 2005 issue:
Home/Current Issue | Previous Issues | Articles by Topic | Patient Education ©2008 The University of Texas M. D. Anderson Cancer Center |
|||||||||||
TOP