In the comments to that post, industry observer David Hamilton pointed me to an article he’s written for BNET on the implications of 23andMe’s price plunge, which (like everything he writes) is well worth a read.
Hamilton has long argued, quite compellingly, that the true business model of personal genomics companies relies only indirectly on selling genome scans to consumers:
…personal-genomics companies don’t intend to make money by selling the tests. Instead, their business generally depends on amassing a giant anonymized database of customer genetic information that can be mined for research studies by academic researchers or drug companies.
And true to form, the new price of 23andMe’s chips can’t possibly leave much room for a profit margin; instead, Hamilton argues that 23andMe’s price cut should be viewed as a kind of loss leader, designed to bring in the huge numbers of customers required to find new associations between genetic markers and interesting human traits.
But there’s a serious problem with this business model: it requires accurate collection of trait data from consumers, and it’s uncertain how well 23andMe can do this using online surveys:
But it’s never been clear exactly how 23andMe planned to extract phenotypic information from its customers. It’s even harder to see how this data would prove useful in research, since any disclosures would be entirely self-reported by a self-selected fraction of all 23andMe customers — two classic sources of research bias.
(The strategy underlying the voluntary survey-based trait collection arm of 23andMe – cutely called 23andWe – is laid out in this post on 23andMe’s blog, The Spittoon; for a sample of the survey questions asked, check out this post on BUZZYEAH.)
The second problem for 23andMe is that the plummeting cost of genome scan chips also makes it easier for their major competitors to engage in large-scale genetic association studies for any trait you care to name. And their major competitors aren’t actually deCODEme or Navigenics – they are academic research institutes and international organisations like the Wellcome Trust Case Control Consortium, which is currently engaged in detailed genome analysis of patients suffering from around 30 different common diseases.
It will be difficult for 23andMe to use its standard customer base to compete with these efforts, partly due to simple mathematics. Many “common” diseases have a frequency of around 1% in the population; so even if 23andMe manages to recruit 100,000 customers they’ll still only have around 1,000 customers suffering from any given disease (assuming their customers have similar disease frequencies to the general population; see below). That’s barely enough for even an exploratory genome scan study – and even modern genome-wide association studies using ten times that sample size have struggled to find genetic variants explaining more than 10% of the risk of many common diseases.
Then there are the problems Hamilton alludes to above, relating to the notorious problems with using self-reported survey data to assess health. One of the stronger predictors for success in a genome-wide association study is rigorous clinical inclusion criteria – in other words, making damn sure that all the patients in your group suffer from the same disease, and (somewhat less importantly) that your controls don’t suffer from the disease. It would be very difficult for 23andMe to be highly rigorous if the only tool at their disposal is an online survey.
Once reputable medical institutes get into the genomics-research game in a big way, research scientists aren’t going to pay out significant fractions of their grants to get access to a commercial database when noncommercial databases — quite likely of higher quality — are more readily accessible. Drug companies might be more willing to pony up in the short term, but only until they can access better and publicly available genetic-variation databases.
These are good points, which I’m sure give the researchers at 23andWe the occasional sleepless night. However, I think it’s far too premature to call the death of the industry yet.
Firstly, personal genomics companies don’t have to rely solely on their current (largely healthy) customer base to provide the disease samples they need to generate new associations. Instead, they can actively recruit customers suffering from specific diseases by targeting disease support groups, as 23andMe did back in May in collaboration with the Parkinson’s Institute. The ability to recruit participants by itself doesn’t give these companies an edge over academic researchers (who also have strong networks for gathering patients), but they can offer something to patients that no academic organisation currently can: they can give patients additional information about their entire genome, including risk of other diseases, ancestry information, and the genetic basis of other traits like eye and hair colour, in a format that is user-friendly and easy to digest. That’s a potentially powerful recruiting tool, if wielded carefully.
It’s also quite likely that personal genomics customers would be enriched for disease patients even in the absence of specific targeting, simply because such patients would likely be naturally drawn to the possibility of learning more about the genetic basis of their disorder. Still, the lack of solid clinical ascertainment would be a major issue with using customers as research subjects, so perhaps the targeted approach through support organisations (with good clinical databases) is the only reliable long-term strategy.
Secondly, even if it proves impossible to compete with academic researchers in the medical genomics space, 23andMe and deCODEme (but not Navigenics) do have a fairly good fall-back position: they can generate information about things that fascinate the public, but are difficult to get funding for in the public arena. I’m talking, of course, about both genealogy/ancestry testing and the genetics of normal variation (e.g. pigmentation, handedness, attached versus detached earlobes, that sort of thing).
Genetic genealogy and ancestry testing are huge markets, and 23andMe already seems to do them better than most other companies in the space – certainly their interface is slick, and their investment in this whole-genome analysis of almost 1,000 genetically diverse humans performed by Stanford University researchers demonstrates their commitment to this area. In addition, 23andMe’s customer base will give it substantial power to uncover the genetic determinants of many variable traits that would seem trivial to a grant committee, but that customers would pay good money to explore – especially once scans become cheap enough to run on multiple family members. (Imagine the attraction of tracing your family’s “big nose gene” back to your maternal grandmother.)
Of course, this isn’t information that big pharma is likely to want to pay for, so a forced shift to pure customer service would deprive the industry of their projected long-term pay-off (as Hamilton has argued). I suspect that personal genomics can be made viable even without the promise of a novel medical database for sale, but such an outcome would certainly result in a fairly major contraction in the personal genomics industry.
I do expect to see 23andMe persist even if the industry as a whole is thrown into crisis. The company has already proven itself to be both flexible and imaginative, successfully differentiating itself from its market competitors (to the point that it virtually owns the “personal genomics” brand in the public mind) and negotiating the regulatory hurdles laid down by the California Health Department back in June. It also has strong ties with Google, not only personally (23andMe co-founder Anne Wojcicki is married to Google co-founder Sergey Brin) and financially (Google has invested at least $3.9 million in the start-up) but also ideologically, in its explicit focus on “data democratization”. The deep pockets and technical expertise of Google certainly can’t hurt 23andMe’s long-term prospects.
For 23andMe’s competitors the outcome is less rosy. deCODEme’s parent company, Iceland’s deCODE Genetics, is by all accounts in fairly dire financial straits; there may not be much left in the coffers to bail out the personal genomics service if the market goes awry. However, the undermining of the medical genomics market would be most disastrous for Navigenics, who from the start have unambiguously committed themselves to providing their customers only with medically actionable information, and dismissed their competitors’ genetic ancestry and normal variation genetics as gimmicky and frivolous. It will be very interesting to see how they respond if frivolity turns out to be the only way to actually make money in the personal genomics arena.