Improved method for genetic data analysis
BY AMY ADAMS
Statistics isn't what normally comes to mind when people think of cancer research, but a new statistical tool developed at the School of Medicine could smooth out some of the fits and starts that have plagued the effort to understand and treat the disease.
The tool is at the heart of a new study that divides kidney tumors into subtypes depending on which of thousands of genes are turned on or off. The idea behind this and related studies of other types of cancer published over the past five years is that doctors can use the information to decide the most appropriate treatment strategy for each patient. Targeting the treatment to a patient's specific cancer means quicker treatment and fewer side effects. Sounds good, right?
The problem is that in many cases the subtypes that turn up in one analysis are absent in follow-up studies, rendering the work clinically irrelevant. The new way of analysis, published online Dec. 5 in the Public Library of Science-Medicine, should minimize these setbacks and help turn cancer research into cancer cures.
Robert Tibshirani, PhD, professor of health research and policy and of statistics, said part of the problem lies in how scientists analyze the data. "Many people have applied old statistical tools to new data, and they don't necessarily work," he said. The studies in question, called microarray studies, generate veritable haystacks of data. Most researchers search for genetic needles in those haystacks. Sometimes they find the needles, but sometimes they accidentally mistake hay for a needle, confusing the entire field.
Tibshirani and his colleagues got into a debate in the pages of the New England Journal of Medicine in March over one such study that Tibshirani said used flawed statistical methods, generating interesting but false conclusions. He said his new tool would help eliminate such misleading studies and let cancer researchers focus on more relevant data. He said the new tool looks for larger groups of genes—equivalent to searching for pitchforks rather than needles—that are more likely to turn up again in future studies. These groups represent biological pathways that are characteristically active or inactive within tumors.
In the first test of the tool, he and his colleagues James Brooks, MD, associate professor of urology, and research associate Hongjuan Zhao, PhD, studied the most common form of kidney cancer, called renal cell carcinoma. This cancer kills nearly 95,000 people worldwide each year.
Brooks and Zhao analyzed the genes that were turned on or off in kidney tumors removed by their collaborators at Umea University in Sweden. Using Tibshirani's approach they found 259 interesting genes. Whether these genes were being actively used by a particular cancer could reveal whether that cancer was likely to spread aggressively and need more vigorous treatment.
"Picking out who has a more or less aggressive cancer can help us decide how to follow that patient once we treat them," Brooks said. For example, somebody who has a less aggressive cancer may need fewer CAT scans after surgery. That would save the patient from needing to come to the hospital regularly for tests, prevent any problems from excessive exposure to radiation during the CAT scans and save money. Likewise, the information could help identify patients who need very aggressive treatment beyond surgery alone.
Eventually the group wants to narrow those 259 genes to a smaller subset that can accurately distinguish between cancers. Tibshirani said that he could only have developed this statistical tool at Stanford, where he can work closely with colleagues like Brooks and Zhao who are actively using microarrays to study cancer. "I think that our collaboration is an excellent example of Stanford's strength in translational medicine," he said. Tibshirani, whose expertise is in biostatistics, said he needs input from cancer surgeons and researchers to tweak his statistical programs effectively.
The new tool is now available to other researchers carrying out similar analyses. Tibshirani hopes it catches on with researchers in the field and helps prevent some of the misleading studies that have been published in the past.
