1) If we all had read the relevant literature (because we all have such copious amounts of free time), we could have saved ourselves the bother. While this PLoSBiology article is about whole genome sequencing (‘WGS’), it’s still relevant (boldface mine):
Moreover, the downstream costs of a diagnostic intervention can far outweigh the upfront costs of the initial test. This is especially true for tests that generate a large amount of information, and potentially large amounts of ambiguous information as well as false positives and incidental findings. The downstream resource and health consequences of ambiguous results are substantial and can include clinical follow-up, additional tests, and also unnecessary surveillance and interventions—as is seen with other technologies, such as has happened, for example, with the introduction of prostate specific antigen (PSA) testing. In clinical practice, there is rarely such a thing as a “low cost” test; the “low cost high value” WGS may be rarer still.
…we need to bear in mind that its uses in research do not necessarily imply equivalent utility in the clinic. Utility in a clinical setting depends on many—and very different—factors, and must take into account not only such performance characteristics as sensitivity, specificity, and positive and negative predictive value, but also demonstration of beneficial impact of using the test on patients’ health, or on health services delivery. Failure to do so can trigger overt harm to patients in addition to excessive cost to the health care system.
It is clear that genomic sequencing will prove to be a useful diagnostic approach in specific situations…
In contrast to these successes, there are few data and little compelling support to suggest that WGS of individuals with common diseases will result in clinically actionable information, or that whatever benefits are accrued might outweigh the burdens of, for example, false positive results or the follow-up investigation of ambiguous results. Common diseases that, by definition, affect the greatest number of individuals, have a relatively low genetic component, placing an inherent ceiling on the usefulness of genomic information to meaningfully inform individuals regarding these disorders.
There’s more to diagnosis than just information.
2) Acceptable error rates for a research tool, especially one that is primarily used to generate hypotheses (i.e., what genes might be involved in Crohn’s disease?), can be trouble when applied to diagnostics. Lior Pachter points out that many 23andMe customers will have at least one medically relevant error–which, at best, leads to additional tests to rule out the problem (see point #1). Also, how heritability and epistasis affect diagnosis hasn’t really been discussed much on the intertoobz, but it’s important.
3) While people probably won’t be getting double mastectomies based on 23andMe (though they might get unnecessary tests–point #1), I’m getting really pissed off about the warfarin/blood clotting mocking (here’s a really bratty example). Any woman who either uses hormonal contraception to prevent pregnancy or who is (or could be) pregnant–and thus at risk for a difficult pregnancy–will have to factor in this information. In this situation, imprecision can kill. Which is precisely what the FDA is supposed to prevent, as of 1976.
4) The thing that hasn’t been discussed: why is this a for-profit, private (non-sharing) venture? This is a mature technology–it’s built around Illumina’s SNP chip, so, beyond the human-friendly bioinformatics wrapper, it’s not clear what 23andMe brings to the rest of us. If having a humongous database with genetic and patient data is really that critical, then why keep it private (within the boundaries of HIPAA)? I find it odd that some of the staunch defenders of open access, well, everything seem to find little fault with locking away these data for private gain–which, as Matthew Herper noted when this scandal first broke, is 23andMe’s business model. If you want to claim 23andMe is doing this better than a non-profit consortium would (especially the gene finding part), how would you know?
5) All of that said, there are legitimate diagnostic applications for sequencing (e.g., cancer, rare diseases), so I hope 23andMe’s arrogance in dealing with the FDA doesn’t poison the rest of the field.