Recent advances in genetic and molecular biology laboratory techniques heralded the promise of better biomarkers for disease diagnosis and treatment decisions. But translation of these research techniques to clinical use has been slow. Now that millions of potential new markers for cancer and other diseases are being discovered, someone needs to determine which have the potential for significant public good.
VIDI assistant member Dr. Holly Janes wants to use new statistical methods to separate the wheat from the chaff, and help investigators focus their efforts on developing the best biomarkers. People are developing better markers in many disease areas, Janes said. “The question is, how do you quantify how much better those markers are?”
PSA screening for prostate cancer in men is a classic example of a marker that could be improved upon. PSA levels are widely used to test for prostate cancer in men 50 and older, but many factors other than cancer may cause elevated PSA levels, such as an enlarged or inflamed prostate.
While disease markers can be used for diagnoses, disease risk prediction, and prognosis, Janes is specifically interested in markers that predict an individual’s response to a given treatment. She has just submitted an R01 grant to study a new biomarker panel, called Oncotype DX, in women with a certain type of breast cancer, estrogen-receptor-positive (ER+) breast cancer, to inform their course of treatment. Many of these women are currently treated with chemotherapy in addition to hormone therapy such as tamoxifen, but it is likely that not all women benefit from the chemotherapy. The panel, which is already in clinical use, looks at tumor expression levels of a variety of genes to identify those women who would benefit from chemotherapy in addition to tamoxifen, and allow other women to avoid the toxicity associated with chemotherapy. Janes’ statistical acumen may come in useful in analyzing the cost effectiveness of new markers such as this panel. Many newly developed markers, such as panels of gene expression, carry a hefty price tag. “It’s important not only to ask is marker A better than marker B, but is it enough better than marker B to warrant the additional cost?” Janes said.
The marker panel was evaluated in the context of a clinical trial, where women with ER+ breast cancer were randomized into groups receiving either tamoxifen or tamoxifen and chemotherapy. The study organizers then examined the women’s expression levels of the marker panel and asked whether different marker levels were associated with different benefits of chemotherapy.
The problem with this type of analysis, Janes said, is that it doesn’t actually reveal much about the marker. “You could get significant results that are not actually clinically useful,” she said. For example, you could have very few women with marker values that suggest chemotherapy can be avoided, even with a large interaction. In this setting, the vast majority of women would not benefit from having the marker measured.
So Janes, along with Center collaborators Drs. Margaret Pepe and Garnet Anderson, and Dr. William Barlow of Seattle’s Cancer Research and Biostatistics, wants to use three statistical measures to improve marker analysis. The first measure will summarize treatment effects for each marker value, which will help clinicians make treatment decisions once markers are measured. This measure will also determine the proportion of people who would benefit from having this marker measured, Janes said. In the case of women with ER+ breast cancer, it would reveal how many women could avoid having unnecessary and potentially toxic chemotherapy.
The second measure would quantify how accurately the marker could distinguish between patients who would benefit from one treatment over another and those who wouldn’t. A good marker would be able to clearly assign the two groups of people with minimum ambiguity.
The third measure looks at the overall population impact of using this marker. “What’s the impact on survival or recurrence-free survival if instead of treating everyone, you use the marker to make treatment decisions?” Janes said.
These statistical methods are useful for cancer markers since so many chemotherapy drugs are toxic and therefore avoiding unnecessary treatment is paramount. However, Janes sees her work as having broader applicability to many other diseases. For example, treatment selection markers might be useful in resource poor areas to focus treatment strategies on those most likely to benefit.
Since coming to the Center in 2007, Janes has also been working with the HIV Vaccine Trials Network (HVTN) on statistical analysis of vaccine trials. She, along with many collaborators in the HVTN, have been analyzing data from the Step study, an HIV vaccine trial that was stopped in 2007 as it failed to protect against infection. Janes has focused on assessing the vaccine’s impact on participants who became HIV infected during the trial. So far, she said, there doesn’t seem to be any difference between participants who received the vaccine and those who received the placebo in disease progression.
“One of the best parts of being a statistician is that we are generalists who can work in a variety of fields,” Janes said. Her background reflects this – she got her PhD in biostatistics at the University of Washington, and then worked on statistical methods for air pollution epidemiology during a postdoctoral fellowship at Johns Hopkins University before returning to Seattle.
“This is really a unique place for statisticians,” Janes said. “The Hutch and UW are pioneers in biostatistics.”