23andMe’s health reporting problem

23andMe, the direct-to-consumer genetic testing company, launched ten years ago this month. But despite the hype around genetics, then and now, 23andMe has so far failed to deliver on its promise.

That’s largely due to a crackdown four years ago by the US Food and Drug Administration (FDA).

The FDA deemed that 23andMe did not have enough scientific evidence to back up the disease risk probabilities in its health reports. The FDA’s main concern was that customers might take unnecessary or harmful medical action based on the results.

That caused 23andMe to pull the health reports from its offering, leaving only the raw DNA data and ancestry reports.

Over the past four years, a small amount of progress has been made. In April, 23andMe came to an agreement with the FDA to offer limited health reports to US customers. When I say limited, I mean only ten diseases are allowed to be profiled. Although that includes two big ones: Parkinson’s disease and late-onset Alzheimer’s disease.

Unfortunately, we in New Zealand don’t even have access to the limited report. It may have something to do with our Fair Trading Act.

According to a 2016 research paper by Victoria University law student Helena Jochem, “if companies offer the delivery of a genetic risk report for a certain disease without a valid link between disorder and gene mutation, they violate s11 and s13(b) of the Fair Trading Act 1986.”

It may also be that we’re not a big enough customer base to warrant the effort of 23andMe getting approval from our regulators.

In any case, arguably we’re not missing out on much. I usually don’t back regulators who stamp down on technology, like the FDA did, but in this case I agree with them. I myself have experienced 23andMe‚Äôs dubious data.

I signed up to 23andMe in 2012, so I was able to obtain my health report at that time. I was especially curious to find out if I had a genetic predisposition to type 1 diabetes. That’s because I had been previously diagnosed with that disease, ironically on very same day 23andMe launched: 19 November 2007.

As I described in my 2014 book, Trackers: How Technology is helping us Monitor & Improve our Health, 23andMe gave me a 4.1% risk of getting type 1 diabetes during the course of my life. That was about four times the average risk. But still a low risk (if I could live one hundred lifetimes, on average I’d only get it four times).

However, after doing more research for my book, I began to doubt the accuracy of the 4.1% figure I was given.

23andMe’s health reports, when they were available, gave you statistical probabilities for a number of diseases. But how those probabilities were calculated was sometimes opaque. When I dug into my 4.1% risk factor for type 1 diabetes, I found that 23andMe had made unexplained assumptions about the SNPs it used (SNPs are genetic markers and there are roughly ten million of them). In short, they didn’t provide me with enough evidence for the 4.1% figure.

The bigger problem, which still remains today, is that factors other than genes – such as environment and lifestyle – can play a large part in whether or not you’ll get a disease. In other words, your DNA on its own is a poor predictor. While 23andMe’s report did mention this, it still claimed that genetics was more important. But that’s debatable.

The kicker was that even if I’d discovered my risk factor prior to getting type 1 diabetes, it would’ve been useless knowledge. Nobody knows what causes the disease, so it cannot be prevented.

The ten disease risks the FDA has now allowed 23andMe to include in its health report in the US are all backed by strong evidence. But it may take many more years for the company to get sufficient evidence-based data for other diseases, such as type 1 diabetes. So I’m not expecting 23andMe to offer a full health report any time soon.

In the meantime, the company has pivoted into drug development. With its 2 million customers, 23andMe has a lot of DNA data to work with. It’s also gathered an estimated 600 million phenotypic data points, meaning self-reported data from questionnaires.

As explained in a Nature.com article last month, 23andMe customers “have each answered an average of 300 questions on a huge array of traits, including their medical histories.” There is some debate about the accuracy of self-reported data, but even so using it to help find new drug solutions seems like a great mission.

According to the same article, the gene-testing market is expected to grow threefold over the next five years. So the market for personal genomics is still bullish, despite 23andMe’s health reporting woes.

23andMe has a lot to prove though. I remain skeptical about how useful genetic data is in predicting disease – not to mention preventing it. That said, 23andMe is an innovative company in the self-tracking movement and its goals are admirable. So I’m glad it exists and continues to be well funded.

I’m also glad the FDA held 23andMe to account. It’ll make the company work harder to deliver accurate, useful health reports.