The narrator in the Lemony Snicket children’s books reminds us, in perfectly straightforward terms, just how bad it can be to make assumptions based on incomplete evidence.
“Assumptions are dangerous things to make,” the narrator notes, “and like all dangerous things to make — bombs, for instance, or strawberry shortcake — if you make even the tiniest mistake, you can find yourself in terrible trouble.”
In the world of healthcare, this rule holds especially true. So when trying to determine how to best interpret lab results when a genetic mutation of unknown pathogenicity is discovered — gene mutations that are commonly called “variants of unknown significance,” or VUS — assumptions should be avoided at all costs.
Unfortunately, that rule has not always been observed. Take the following example: A woman goes to her doctor asking about a genetic test for breast cancer, noting that her sister was recently diagnosed. So doctors run a genetic test on her and her affected sister to see if they can spot a mutation in one of the two genes associated with breast cancer risk (BRCA1 and BRCA2).
The results come back from the lab, and a mutation is noted in the woman — one that matches a mutation in her affected sister. This genetic mutation could be classified one of three ways: as a mutation that is known not to cause breast cancer; as one that they know does cause breast cancer; or as a VUS, meaning they don’t know if it does or doesn’t.
If it’s a VUS, another classification takes place, in which the lab (and, in turn, the patient’s doctors) tries to establish a likelihood that this mutation augurs a high probability of breast cancer. It is at this point where, too often, for many understandable reasons, the laboratory and healthcare officials have leaned toward suggesting that the family history plus the appearance of an unknown mutation suggests a high likelihood of developing the disease — which has led to patients preemptively removing their breasts and/or ovaries who later discover the mutation was shown not to indicate a high likelihood of breast cancer.
“This has happened,” says Heidi Rehm, a laboratory director, molecular geneticist, and associate professor of pathology at Harvard. “Traditionally, labs and researchers have often over-assumed pathogenic effects, even when insufficient evidence existed. It’s now clear that many of these variants may not be disease-causing. We just haven’t been able to prove it, but we’re getting better about recognizing the uncertainty, and patients are getting better advice as a result.”
The way in which researchers judge whether a variant is likely to be pathogenic or benign often involves many steps. Elana Silver is principal consultant at the health and education consultancy firm Laurelton Research, currently working with genome interpretation software company Omicia. She says the first step is to use special software or a large database of genetic data to determine the effect of the variant. A “loss-of-function mutation” (i.e., a mutation that causes reduced or abolished protein function) in a known disease-associated gene is often presumed to be pathogenic. Silver says that for other types of mutations, researchers often turn to medical literature and case reports to gather evidence, using mathematical models that combine some or all of the available data to give a score indicating the probability a variant is pathogenic.
Rehm is one of many specialists the world over trying to both acknowledge and properly relay the limits of what we currently know regarding variants of unknown significance, while at the same time trying to increase our understanding of their effects through better data-gathering and analysis.
Their efforts begin with educating patients as to exactly what a VUS is and why it’s understandable that the effects of so many mutations are not yet known. “We start with the fact that the human genome has a lot of variation, and variants explain what makes us all different,” Rehm says. “Some explain disease, and some have no effect on our phenotype or traits. So the first step is trying to figure out which variants actually cause disease.”
Silver says these ambiguous results are still very common, although they are becoming less so as data-sharing practices become more accepted. To see how much uncertainty is still involved, consider these numbers: For BRCA1, almost 40 percent of identified variants are labeled variants of unknown significance; for BRCA2, that number is 49 percent. “These numbers tell us about the proportion of variants that are not understood,” Silver says, “but don’t really tell us how often a VUS will occur in a clinic setting.”
Over time, she says, the VUS rate has declined, with recent data from several laboratories indicating a VUS rate for BRCA1 and BRCA2 of 2.9 percent and 4.4 percent. “However, the VUS rate is dependent on the laboratory’s testing experience,” Silver says, “and as such is expected to change over time as more laboratories begin to offer BRCA1 and BRCA2 testing.”
The story of our genes is often of the choose-your-own-ending variety, where a mutation may or may not cause a disease. Patients are fairly used to going to the doctor and getting answers that aren’t clear: Doctor, my arm hurts. Take some ibuprofen — let’s see if it goes away. But patients often equate genetic information with certitude. As well, the discussions often concern serious diseases. For example: An expectant mother of an advanced age might have a family history that causes doctors to conduct a prenatal chromosomal microarray analysis early in the pregnancy, a test that can detect chromosomal abnormalities in the fetus. A lesion may be identified, one that can’t be explained, but one that could portend serious abnormalities. Given the time urgency and the uncertain nature of the VUS, the patient’s fear and frustration is heightened. Again, no one wants to make decisions on incomplete information — but sometimes that’s all a patient has.
This is why most labs label their reports using a five-point scale, in which they classify a variant in relation to how likely it is to cause disease: pathogenic, likely pathogenic, uncertain significance, likely benign, and benign. This can help patients have more information when health decisions must be made.
The main challenge with this system in practice, Rehm says, is that variant classification is subjective. The classification of variants differs not only between labs — one lab classifying a variant “likely benign” and another labeling it “likely pathogenic,” for example — but also over time. A variant once thought to be likely pathogenic can, over time and with observations in more people, be found to be benign.
“We’ve had physicians burned by this who put too much confidence in less certain variant classification,” Rehm says. “Genetic test results used to only be used to determine the cause of disease. Now that people are using it to determine how to treat a patient or who else in the family is at risk, the consequences of being wrong are more serious.”
The goal, then: increase the accuracy of variant classification. To that end, the medical community is helping to grow a free database called ClinVar, which catalogs the relationships between human variation and disease.
ClinVar is like a central storage unit for everyone who has data on a genetic variant. That way, someone combing the information can see that different labs have different classifications of the same variant — this is an indication that more research must be done before doctors can make a determination regarding said variant. As well, 10 labs in agreement can be interpreted with reasonable faith that the greater number deserves more weight when making clinical decisions. “As doctors and even patients look up their variant in the database, they can immediately see if there is a consensus or problem with the interpretation of their lab,” Rehm says. Adding family history or phenotype information helps contextualize the results, helping doctors predict disease outcomes for patients.
Sharing data on genetic variants in databases like ClinVar is key to being able to accurately assess the pathogenicity of variants. Without sharing data in a transparent way, patients are at a higher risk for misinterpreted results. That’s why the community is asking all laboratories to share their data. One laboratory that has not chosen to share its data is Myriad Genetics, which used to corner the market on BRCA1 and BRCA2 testing and is trying to keep that market share by arguing that it has the best dataset to interpret variants.
The Sharing Clinical Reports Project, part of the Free the Data movement, has taken the request for data sharing directly to patients, asking anyone with a BRCA1 or BRCA2 test result to share their findings openly in ClinVar.
The ClinVar database is an attempt to consolidate information about all variants around the world. That’s why Rehm is also working with groups to share the data between countries. Rehm says standardizing this process across the world will help improve available information.
In fact, Rehm recently had a meeting of an international group of breast cancer researchers, diagnosticians, and clinicians, organized by the Global Alliance for Genomics and Health, the intent of which was to get the members to share all their data and to use a common approach to classifying variants. She hopes it’s the first step toward helping doctors not only make better assumptions, but also the first step toward helping them eventually not have to make them at all.