More-Accurate Cancer Treatment
A computer model can predict which drugs a cancer patient will respond to best.
Two patients with the same type of cancer may have very different reactions to identical chemotherapy treatments because their cancers have different genetic causes. Doctors currently can’t predict which patients will have a good long-term response to a drug and which patients’ cancer will come back.
Researchers at the University of Virginia have developed a computer model that addresses this problem by using genetic analysis of tumors to predict which chemotherapy regimen will do patients the most good. The model has initially been tested in bladder and breast cancer, but it seems to have predictive power in all cancer types. The researchers are integrating the model into clinical trials to test whether it can be used to select the patients most likely to respond to a particular drug. They hope it will eventually be used routinely to help doctors quickly find the best treatment for cancer patients.
In a test of the model in two large trials of drugs for breast cancer, the model predicted with 85 percent accuracy which patients would and would not respond to treatment.
Dan Theodorescu, director of the Paul Mellon Prostate Cancer Institute, at the University of Virginia, believes that his group’s model is the first of its kind. It was developed using a database of information on about 60 human-cancer-cell lines maintained by the National Cancer Institute. This database includes information about how these cancer cells respond to more than 100,000 compounds, including all anticancer therapies approved by the Food and Drug Administration. The model also draws on gene-expression analysis of these 60 cancer types–that is, what genes they have and how they are using them.
Theodorescu says that his model uses gene-expression analysis as a kind of Rosetta stone for translating drug sensitivity in the 60 initial cell lines into a prediction of drug sensitivity in whatever tumor researchers are studying. He suggests that the model will also have predictive power for the many other cancer types not included in the initial group, including lung and pancreatic cancer and lymphomas.
Theodorescu and his collaborators also showed that the model could predict how patients would respond to cancer drugs. He is planning studies and clinical trials with researchers and centers around the country to validate the model across all major human cancers.
“This is a very innovative approach,” says W. Marston Linehan, chief of urologic oncology at the National Cancer Institute’s Center for Cancer Research.
“Scientists have been trying to apply [genetic analysis] to cancer treatment, but it hasn’t panned out the way we’d like,” says Cheryl Lee, associate professor of urology at the University of Michigan Comprehensive Cancer Center. “What [Theodorescu] is doing with [his model] is very novel because it uses computer modeling not only to discern tissue types, but [he] literally tries to match patients with chemotherapeutics.” Lee is part of a Michigan group that is collaborating with Theodorescu on a bladder-cancer clinical trial incorporating the model.
Most patients respond to chemotherapy in the short term, but, particularly in the case of bladder cancer, patients who respond over the long term to a given drug are rare, and the cancer often recurs. Lee hopes that the model will help match patients with drugs that will work over the long term.
The gene chips used for the studies, made by Affymetrix, now cost only $260. Theodorescu says that this price is “trivial compared to imaging,” such as the magnetic resonance imaging and PET scans routinely used in clinical trials. And he says that the price could come down if researchers used Affymetrix chips customized for particular cancer types.
Most hospital laboratories don’t have the resources to perform and interpret gene-expression analysis. But Theodorescu hopes it will become a routine part of patient care in the next five or ten years, complementing the under-the-microscope tissue analysis currently used to characterize cancers.
Become an Insider to get the story behind the story — and before anyone else.