Personalized Genetics: Too Soon, Too Little?

9 02 2009

ResearchBlogging.orgPersonalized Medicine is the concept that managing patient’s health should be based on the individual patient’s specific characteristics instead of on the standards of care. Often the term ‘personalized medicine’ is restricted to the use of information about a patient’s genotype or gene expression profile to further tailor medical care to an individual’s needs (see [1])

This so called Personalized Genetics is a beautiful concept. Suppose you could predict people’s risk for a certain disease and be able to prevent it by encouraging positive lifestyle changes and/or start a tailor made therapy, suppose you could predict which patients would respond to an intervention and which people should avoid certain medications. Wouldn’t that be wonderful and much better than treating everybody the same way only to benefit a few?

Research like the human genome project and recent advances in genomics research have boosted progress in the discovery of susceptibility genes and fueled expectations about opportunities of genetic profiling for personalizing.

But are the high expectations justified?

For personalized genetics to be (clinically) effective it must fulfill the following requirements (based on [2]):

  1. Clear and strong association of the gene (expression) variant with the susceptibility to a disease or the outcome of a treatment
  2. Improved prediction compared to other risk factors, including traditional risk factors and clinical judgment…
  3. determined in good quality studies with a sufficient number of events (if the events are rare you cannot accurately predict the outcome)
    (1-3 make up the predictive performance)
  4. The availability of effective interventions or effective alternatives
  5. Cost-effectiveness


According to an editorial in the January issue of the Annals of Internal Medicine “the promise of personalized genetic prediction may be exaggerated and premature” [2]. This is especially true for many complex diseases, where 1 variant alone is unlikely to make the difference.

The editorial is written by John Ioannidis, who is a professor at the University of Ioannina School of Medicine in Greece and has an adjunct appointment at Tufts University School of Medicine in Boston. His research focuses on meta-analysis and evidence-based medicine with special emphasis on research methodology. Ioannidis is a brilliant researcher, epidemiologist and inspiring lecturer (I have attended a lecture of him once at a Cochrane Colloquium). Therefore I would urge everyone interested in personalize genetics to read his editorial.

Here I will give a summary of the editorial entitled “Personalized genetic prediction: too limited, too expensive, or too soon” [2].The editorial summarizes two publications in the same issue of the journal [3,4] and gives an overview of the literature.

Ioannidis stresses that recent studies into the predictive performance of common genetic traits have several shortcomings, including an often weak design with few events (*3)*, incomplete comparison with traditional risk factors (*2) and exaggerated prediction of effects because of the models used (*1).

To date, the genotypic information does not substantially improve the prediction of future cardiovascular disease (CVD), prostate cancer and type 2 diabetes beyond traditional risk factors. In the case of age-related macular degeneration, genetic information does increase the ability to predict progression to the disease. However the predictive power to individualize risks remains relatively weak.

Indeed, a recent paper published in PLOS [5] reinforces that a strong association between single nucleotide polymorphisms (SNPs) and a multifactorial disease like age-related macular degeneration, diabetes type 2, CVD and Crohn disease may be very valuable for establishing etiological hypotheses, but do not guarantee effective discrimination between cases and controls and are therefore of little clinical value yet. For further details with regard to the methods used to determine clinical validity of genetic testing you are encouraged to read the entire (free) paper [5].

Likewise, the study of Paynter et al [3] reviewed by Ioannidis, shows that genetic variation in chromosome 9p21.3 (rs10757274) was strongly and consistently associated with incident CVD in a cohort of white women, but did not improve on the discrimination or classification of predicted risk achieved with traditional risk factors, high-sensitivity C-reactive protein, and family history of premature myocardial infarction. Thus “knowing a patient’s rs10757274 genotype would not help a clinician make better preventive or therapeutic decisions to reduce future risk for heart disease”.
This holds also true for many other potentially causal single SNPs: they have a relatively small effect on their own. Complex diseases are probably the result of numerous gene-gene and gene-environment interactions, which may differ from one population to the other and only explain a small proportion of the trait variance.

Even improved prediction (*1-3) does not necessarily make a predictive test useful. The prevalence of the disease is also an important determinant, i.e. people with high risk gene variants for a rare disease may have a significant higher-than average risk, but still a negligible probability of developing the disease.

Clinical utility of the genetic prediction also depends on the availability of effective interventions (*4) and the cost effectiveness (*5). Another paper in the same Ann. Intern. Med. issue [4] shows that although CYP2C9 and VKORC1 strongly predict the chance of bleeding as a side effect of warfarin treatment, genotype-guided dosing appeared not to be cost-effective for patients requiring initiation of warfarin therapy. Piquant detail: The FDA has approved this kind of genetic testing, although there is no good evidence that such genotyping does in fact reduce the risk of hemorrhage in everyday clinical practice. Such knowledge would require large well designed RCT’s.

Ioannidis emphasizes that despite the poor evidence, genetic testing and commercial use (direct to consumer genetic testing) have already begun and are here to stay. He proposes several safeguards, including transparent and thorough reporting, unbiased continuous synthesis and grading of the evidence and alerting the public that most genetic tests have not yet been shown to be clinically useful. He concludes the editorial as follows:

Helping patients and physicians to decide when to do genetic tests will be a tough task because neither knows much about the rapidly emerging field of genomics. We need to learn more about what our genome can tell us and, more important, what it cannot tell us.

* refers to list, points 1-5


2: Ioannidis JP. (2009). Personalized genetic prediction: too limited, too expensive, or too soon? Ann Intern Med, 150 (2), 139-141 DOI: 19153414 {=wrong DOI researchblogs click here to be linked to PubMed)
3: Paynter NP, Chasman DI, Buring JE, Shiffman D, Cook NR, Ridker PM. Cardiovascular disease risk prediction with and without knowledge of genetic variation at chromosome 9p21.3. Ann Intern Med. 2009 Jan 20;150(2):65-72.
4: Eckman MH, Rosand J, Greenberg SM, Gage BF. Cost-effectiveness of using pharmacogenetic information in warfarin dosing for patients with nonvalvular atrial fibrillation. Ann Intern Med. 2009 Jan 20;150(2):73-83.
5: Jakobsdottir J, Gorin MB, Conley YP, Ferrell RE, Weeks DE. Interpretation of genetic association studies: markers with replicated highly
significant odds ratios may be poor classifiers. PLoS Genet. 2009 Feb;5(2):e1000337. Epub 2009 Feb 6 (free full text).

6: Janssens AC, van Duijn CM. Genome-based prediction of common diseases: advances and prospects. Hum Mol Genet. 2008 Oct 15;17(R2):R166-73. Review.

You might also want to read:

23andme 23notme not yet (post 2008/09/29/)


23andMe: 23notMe, not yet

29 09 2008

23andme cheeper

The company 23andMe was in the news thrice this month:

  1. cutting the price of its service by more than a half
  2. organizing a celebrity spit party
  3. the husband of the 23andMe co-founder Anne Wojcicki, better known as Google co-Founder Sergey Brin, revealed he is at risk for Parkinson’s Disease, as determined by….23andMe.

Coincidence or part of a strategic plan?

23andMe is a ‘direct to consumer genetic testing’ company that as 23andMe puts it: “democratizes personal genetics”. The lowering of the service price from $999 to $399 brings personalized genomics within the range of many.

What do you get for those $399? A spit kit, you do your thing, send the tube to a certified lab, which analyzes your saliva for more than a half-million points (called SNPs) scattered across the 23 pairs of chromosomes you have (hence 23andMe), as well as your mitochondrial DNA. 23andMe shows the digital data and gives you information on certain traits and diseases. 23andMe also gives information on your ancestry and compares your DNA to your relative’s and friend’s-genes, if you want to share that knowledge with them. With your genes in their database you help 23andMe to perform more research for new discoveries, a program called 23andWe. In fact once you sign up you cannot refuse the use your (anonymous) DNA for this purpose.

The main question is: what purpose does this serve (besides as a potential for yielding income)?

According to 23andMe the main purpose is ‘for research’, ‘for education’ and ‘for fun’: “It’s fun to learn about your own genome”.

In this light, we should probably see the recent event 23andMe organized: a spit party where a few hundred people were lured away from the catwalks during the Fashion Week in New York City. On the sound track of “a whole lot of love” celebrities were spitting their DNA-containing saliva in a tube (see here and here). According Guy Kawasaki, who report on it on his blog (see here),

“even Goldie Hawn and Kurt Russell were there providing their spit, but their handlers wouldn’t let me take a picture. I found this ironical: Giving DNA was okay but not a picture.”

The aim for which Sergey Brin let 23andMe test his DNA was less funny. As Sergey (whos mother has Parkinson) explains in his brand new blog:

(…..) Nonetheless it is clear that I have a markedly higher chance of developing Parkinson’s in my lifetime than the average person. In fact, it is somewhere between 20% to 80% depending on the study and how you measure. At the same time, research into LRRK2 looks intriguing (both for LRRK2 carriers and potentially for others).

Thus this shows a 3rd aim: diagnostic?!
Formally 23andMe denies there is a diagnostic purpose (in part, surely, because the company doesn’t want to antagonize the FDA, which strictly regulates diagnostic testing for disease). However, 23andme does give information on your risk profile for certain diseases, including Parkinson.

In addition, 23andMe encourages the formation of networks of people sharing the same traits.

“If you want to have a community around psoriasis,” Ms Wojcicki said, “we’d like to be able to allow you to form a psoriasis-specific community.” (see New York Times article)

Psoriasis-specific community when you only have the genes that may enhance the risk of getting psoriasis??

That sounds like condemning you to a psoriasis patient already?!

Then lets discuss the following burning question: how well does 23andMe predict that you will get the disease?

Even the LRRK2-gene data of Mr. Brin aren’t that conclusive. A marked higher chance of 20% to 80% is often misconceived as meaning that Sergey’s chance of getting Parkison is 20-80%, or “he will almost get the disease for sure”. As explained by the Gene Sherpa in his excellent post on this subject (see here) it only means that the LRRK2-mutation increases the normal chance of Americans/Europeans getting Parkinson from 2-5% to 4-10% at the most (the chance is less than doubled). Furthermore LRRK2 isn’t the most crucial gene for getting Parkinson.

23andMe has chosen to relate personal health info only to common diseases and common genes. Thus whether you have an enhanced or lowered risk for breast cancer (normal 1 out of 8 women) is determined by 2 (not very predictive) SNPs associated with Breast Cancer, but not by determining BRCA1/2 mutations that are highly predictive for breast cancer, but rare in the entire (western) population .

Although 24andMe explicitly mentions that the tests are for non-diagnostic purposes, it is hard to imagine that people will see it otherwise. But:

  • Most genes are only weakly predisposing
  • Often multiple genes are working in concert in a difficult to predict way (seldom one gene-one disease)
  • The environment and chance also play an important role.

Thus the value of these fun predictions is low, but how does it affect people that think they are prone to having a disease? For some it might be reason to adjust their lifestyle (but then, what is the chance you really change “your destiny”), others may get fixed on their presumptive future disease, confused, or depressed. It is not without reason that genetic screening is usually restricted to people with high risks, when a disease can be predicted accurately (without too many false positives and negatives), something can be done about it (prevention or treatment), and only as part of a genetic consultation by professionals.

Sources; further reading