By Mark D. Uehling
August 13, 2002 | Breast cancer can be a grim family heirloom, a predisposition to which is passed from one generation to the next. Although usually a sporadic disease, inherited mutations in two key genes, BRCA1 and BRCA2, present in seven percent of women, raise the lifetime risk of developing breast cancer to as high as 85 percent.
Finding those mutations is not yet a standard workup in most hospital labs. So one pressing question for doctors is how to decide which patients should get an analysis of their DNA. Now a Dallas surgeon, David Euhus of the University of Texas, is using IT.
Euhus recently demonstrated an unusual combination of his medical and computer programming knowledge in two papers for the Journal of the National Cancer Institute. In one study, Euhus and his colleagues predicted 148 different patients’ risk for BRCA mutations using either experienced genetic counselors at large cancer centers, or a computer program called BRCAPRO. In statistical terms, the program was twice as specific about finding risk as the human counselors.
“BRCAPRO was more willing to say ‘there is no mutation’ than the counselor was,” says Euhus. “BRCAPRO had overall discrimination that was better. If you’re not an experienced genetic counselor, the program is a great assistant, a great mentor. It’s pretty accurate.”
There’s no reason for genetic counselors to fear for their jobs: They can use the software, after all, and the computer can’t perform other tasks such as helping patients understand the limitations and risks of any medical test or procedure. Nor is the use of BRCAPRO limited to academic environs: The software is used at Euhus’ own patient clinics and in other hospitals around the country, and for diseases other than breast cancer. “You use the computer to help you screen through families. You like to be able to tell one patient, ‘You know what? We should look at your genes.’ ”
Originally developed at Duke University, BRCAPRO was a bit ragged around the edges at first, Euhus concedes. It ran in DOS. He fixed that, writing a Windows BRCAPRO interface called CancerGene, which deposits the patient data into a relational database. Roughly 1,000 major cancer centers in 40 countries have now licensed his program, which is free.
Doctors like to show their patients graphs based on various diagnostic events or medication changes. The program generates a simple line showing a patient’s risk at various ages. “When patients can see their risk graphically,” says Euhus, Òthey seem to understand more clearly where they stand in relation to others. It’s a very powerful counseling tool: ‘Here’s what happens if you stay on these hormones for another 10 years.’ ”
Like many physician-researchers in the breast cancer field, Euhus is also working to understand the origins of the vast majority of breast cancer cases in which there are no BRCA1 or BRCA2 mutations. In such patients, there is a strong suspicion of other gene complexes that may be going awry. Using a battery of genetic markers to evaluate that hypothesis, Euhus found well-defined sections of the genome that were deleted in breast cancer patients. “We had pinned these markers down as hotspots for losing DNA in breast cancer,” says Euhus. “It’s kind of a barometer of the integrity of the genome.”
LOH and Behold
In the other paper, Euhus reported data from 30 women showing that so-called loss of heterozygosity (LOH)—the loss of one copy of a gene or chromosomal region—may be causing cancer. Says Euhus: “Our study was looking for small mutations that should be there.”
To test the accuracy of looking for LOH, Euhus compared it to a well-tested predictor of breast cancer risk known as the Gail model. The Gail model is a statistical algorithm that can expertly assign a woman to category of risk for breast cancer—low, medium, or high—using a variety of biographical facts. But doctors would like to be able to tell a woman she will or will not get cancer, and the Gail model is not always able to do that.
“We need a more individualized test, which we hope the LOH will be,” says Euhus.
Intriguingly, there was a high correlation between the well-tested Gail model and the LOH approach, suggesting that they are both measuring the same biological process. The conclusion of Euhus’ article is especially eye-catching: “Molecular analysis of benign breast epithelial cells may provide an approach for accurate, individualized breast cancer risk assessment and may provide clues to the earliest steps in breast carcinogenesis.”
In other words: Informatics-based approaches to breast cancer patients could ultimately prove more targeted, more precise in estimating the risk of developing the disease than the traditional analysis of stained cells under the microscope, which remains an entirely subjective art.