Translating Basic Research Tools Into Diagnostics

Because we like to snark around here, let’s snark first:

Some of the same people who react with horror when parents, who know they’re being irresponsible, still demand an antibiotic to treat a childhood ear infection are the same people who also believe that patients, when presented with potentially life-changing, frightening and complex population genomics data, will behave calmly and rationally.

I have some doubts.

Anyway, onto the less snarky bit. Michael Eisen writes (boldface mine):

If genetics were simple and our understanding of it were complete, companies could provide accurate reports that say “based on your genotype, your age and personal history, you have a 7.42% chance of developing ovarian cancer in the next 10 years”. However, we are far, far, far away from this. We have an incomplete catalog of human genetic variation; known genetic variation can explain only a small fraction of the heritable component of most phenotypes of interest; we have a poor understanding of how different genetic variants interact to affect disease risk or other phenotypes; and we have essentially no capacity to incorporate environmental effects into predictive models. In many cases current, incomplete, data may point to someone having an elevated risk of some disease, when they really have a lower than average risk. And, to top it all off, there are very few cases where knowing your risk status or other phenotype points to genotype-specific actions (with the BRCA status referred to in the FDA letter a notable exception).

The data are, at this point in time, very very messy. I don’t think anyone disagrees with that. The question is what to do about that. One the one side you have people who argue that the data are so messy, of so little practical value, and so prone to misinterpretation by a population poorly trained in modern genetics that we should not allow the information to be disseminated.

A key problem is that 23andMe is, at its core, a for-profit privatized GWAS (genome-wide association study) company. As Matthew Herper noted Tuesday, they take patients’ genomic data along with their health histories (‘metadata’) and look for associations, with the goal of identifying genes that are linked with disease. This is a powerful research tool, one that is widely used. But with rare exceptions (BRCA and breast cancer; Huntington’s chorea), its power as a diagnostic tool–what does it mean for an individual patient–can be pretty weak.

There is another problem: by definition, some people will be at an elevated risk for any given condition. Even if that additional risk is minimal (e.g., an increase from one percent to 1.5%), it’s still terrifying, if for no other reason that these estimates could be incorrect. There are people who when they hear words like Alzheimer’s, cancer, MD decide that they want all the tests now. “Well, it’s a minimal risk, probably nothing…” “No, I want all the tests now!” (again, consider ear infections and antibiotics)

This isn’t conjecture. The NY Times just published an article (with serendipitous timing) on genetic screening for breast cancer in Israel, with the end result of some women having pre-emptive mastectomies. Dr. Jen Gunter describes her own clinical experience with blood clotting (boldface mine):

Say you are a young woman and you get tested via 23andMe and you are told based on your genetic profile that you carry a genetic mutation that may increase your risk for blood clots. You have no personal or family history of blood clots so you had no clinical indication for testing, yet now you have these results. What next?

Certain genetic mutations do confer an increased risk of abnormal clotting, some significantly, although not everyone with the mutation that you tested positive for will have a blood clot. In fact, most won’t and so universal screening isn’t recommended. Many investigators have looked at this specific mutation and wondered, should we screen women with no history (personal or family) for this mutation? After all, women are more likely to take a medication with estrogen (birth control pills) and they get pregnant, both of which increase the risk of blood clotting, which is potentially fatal. However, after a lot of studies, obsessive reviews, and hashing it out at meetings the experts seem to agree. Mass screening, testing women with no family or personal history, for this specific mutation is likely to cause more harm than good, so we don’t recommend it.

But now you have the result and you tested positive. Now you have the result that if you act upon it might cause more harm than good. You just don’t know. What do you do? Do you go with the belief that your risk might be higher? Do you ignore the results? It’s a conundrum.

If you’re a young woman and you go with the results that means not taking the birth control pill with estrogen. Maybe that’s not such a biggie, after all there are IUDs. But what if for some reason you don’t want an IUD? What if you get pregnant because you choose a non-IUD form of contraception with a higher failure rate than the pill?

And then what do you do in pregnancy? Do you take blood thinners during pregnancy or just after delivery or do we give them to you at all? What if you get blood thinners because your doctor is worried that if you get a blood clot he or she will get sued because some women, regardless of genetics, will have a blood clot. What if you get blood thinners and have a catastrophic bleed during your delivery and lose your uterus or even worse, die? If you truly had an increased risk, then studies tell us that with blood thinners the benefits outweigh the risks, but what if you didn’t really have an increased risk to begin with? Then you have assumed all the risks of blood thinners with no potential benefit.

There are also reports of unnecessary tests for Alzheimer’s.

This isn’t being a Luddite or anti-genomics (I do genomics for a living, by the way). I’m not opposed to people having control over their genomic information (or other medical information). But there are real issues surrounding how a statistical method–even a powerful one–can be turned into a reliable clinical diagnostic. The FDA would be derelict in its duty if it did not regulate this area.

Regarding the particulars of the 23andMe situation, I think they’ll be able to work out a compromise that will involve providing links to genetic counsellors and scaling back the diagnostic claims, especially in the advertising material* (if history is any guide, the latter is probably what angers the FDA the most–using overblown claims to drum up business for its GWAS work).

And because we like the snark:


*Below the fold are some screenshots from their website as of Nov. 27, ~11:30am EST:




This entry was posted in Genomics, Public Health. Bookmark the permalink.

1 Response to Translating Basic Research Tools Into Diagnostics

  1. Min says:

    “Even if that additional risk is minimal (e.g., an increase from one percent to 1.5%), it’s still terrifying, if for no other reason that these estimates could be incorrect.”

    Unfortunately, such an increase is typically reported as “increases your risk by 50%!”

Comments are closed.