Tuesday, August 17, 2010

Francis Collins, personalized genomic medicine, and the nature of probabilistic risk

Probability and the weather
Personalized forecasts
People regularly complain about the weather forecast in a way that misunderstands probability.  It's a sensitive subject for Ken, as a one-time meteorologist in a long-ago former life.  If the forecast is for a 30% chance of showers, and it stays dry, it's often viewed as a bad forecast.  You should have played golf after all.  If it rains, it's often viewed as a bad forecast.  Dammit, they ruined your golf outing, by driving you prematurely into the 19th hole bar!

Neither conclusion is right.

One can only really tell whether the forecast was right the next day, by totaling up the area that received rain, or by averaging the fraction of each area that received rain at any given time, or something like that.

The key point is that this is a 'personalized' forecast in the same way that genomic medicine, and Direct to Consumer (DTC) genetic risk estimation are personalized.  For a given genotype, or given golf course, whether it actually rains or not, or whether you get the disease in question, is almost irrelevant to whether the forecast was a good one.

Probability and risk of disease
Personalized genomics
We said on Saturday that Francis Collins reactions to his genotypic diagnosis of being at elevated risk for diabetes were, at his diabetes-free status at age 60, irrelevant to the genotype-testing result.  If the population average risk of diabetes is, say, 10%, that means 10% of the population will get the disease at some point in their lives.  If Francis' genotype-based relative risk is elevated by, say, 20% relative to  the risk of the average Joe, his absolute risk is raised to 12% (and we're exaggerating risks here to make the point, since most of the genotype-based  risks of common diseases add considerably less than 20% of the average lifetime relative risk, and most lifetime risks are less than 10%).

But in fact he doesn't have diabetes.  That says almost nothing about the accuracy or usefulness of his risk estimate.  12% risk means that 88% of people with the same genotype don't get diabetes, and his status is wholly consistent with that (or with his being at the average risk), and says nothing about whether the risk estimate was accurate.  The only risk estimate that should affect his behavior (he reports slimming down etc. in response to the genotype data), is the 'conditional' risk of getting diabetes at some future age, for a white male who has already survived until age 60.  That is probably an unknown value, for many reasons, such as lack of enough data.   Indeed even if the genetic risk estimate is accurate, a person with his genotype could, if diabetes-free at age 60, actually be at lower than average risk from his age onward.  Why?  Because the bulk of those who were born with the same genotype might already have gotten diabetes at a younger age: earlier onset is a typical way that elevated-risk genotypes work.

The issues are subtle, and just like weather forecasts, easy to misunderstand.  Another reason people should shun DTC services, and just adopt health lifestyles to the extent they can.

More profoundly, if Dr Collins is at our hypothetical 20%  relative risk, that means that the average person with his detected genotype is at 20% increased risk, but does not mean that everyone with that genotype is!  For example, those with the genotype who eat too much McFastfood may be at a 90% risk of diabetes, hugely elevated over the general population risk perhaps, while vegans with the same genotype could even be at lower risk than the population average.  We rarely know enough about interacting or confounding variables, like other genes or lifestyles, relative to the one we know about (the tested genotype) to say more than what's average for that genotype.  But we have every reason to believe that not everyone with the genotype is at its average risk.

When risk differences are less than huge in absolute terms, personalized medicine cannot judge the relevance of these issues except at best in a here-and-now population context, such as by a case-controls study, and these exposure contexts are always changing.  That's why it's important to understand what probabilistic predictions mean, when it comes both to your golfing decisions, and to your life.

In fact for most diseases if you want to know much, much more about your risk than any genotype service can tell you,  just look at the health history of your relatives. That's been known since Darwin's time.  And unlike DTC services, that information's free!  No doctor bills, no unnecessary testing, no specific genes need to be identified.  If we learn of specific gene-based therapies, and you're at high familial risk, then it would be worthwhile to have a proper genetic counseling service do the test.

A Snake Oil factor, but it cuts both ways
The DTC business is cowboy capitalism at this stage, and selling this kind of snake oil to people totally unable to understand the actual meaning of the data is an ethical as well as policy issue, and of course there is disagreement about where the ethical lines are where they should be drawn.  Even most doctors--even most geneticists--are helpless to understand the nature, accuracy, or stability of these risks.  Of course there is always a lot of selling going on, as a glance at Parade or an airline magazine will clearly show (shrink your prostate!  get rid of age wrinkles! have a 20 year-old body at age 70!).  As P T Barnum said, there's a sucker sitting in every airplane seat.  But this is why regulation is important, in our view, in something so closely involved with life, death, and health-care costs.

We've suggested in our posts that the FDA should treat this kind of DTC 'advice' as practicing medicine, and license it only as a part of genetic counseling where it could in principle be properly constrained and regulated to stick closer to truths that we can generally agree on and that consumers (and counselors and physicians) can make reliable sense of.

However we must admit that this is partly a political stance by us.  If the entire business were just shifted to medical clinics, the same testing would probably take place, with the same level of understanding.  After all, the issues are complex even for those with sophisticated knowledge.

In fact, transferring the business (doubtlessly to be supplied, probably on a larger scale, by the same companies) and perhaps establishing it part of what would be lobbied into routine medical exams, the genomic testing could be much more costly to the health-care system, and more lucrative for the companies in the future, and hence perhaps even more problematic than it is now.  Still, we think regulation is better than caveat emptor.

Bottom line?
So the simple conclusion for us is that it's best to follow advice we've only known since Hippocrates a mere 2400 years ago (moderation in all things), and skip both that extra dessert and the personalized DTC genotyping.


occamseraser said...

I apologize if you've already posted and commented on this from Venter:

Ken Weiss said...

Hadn't seen it but it reveals the issue that when there is a lot of money as well as credit involved, egos will be, too.

'The' human genome sequence was a technological coup, but a mixed one involving different and partly overlapping approaches. But it was not a conceptual coup at all. 'The' sequence is not found even in a single person (it's a mix, and in any case each of us has two copies), nor was it (or is it) 100% complete. Not only is there lots of variation but some of it is very hard to sequence, etc.

Craig Ventor is entrepreneurial as well as combative, and perhaps he could be greedy too (I don't know about that). But is a big-thinker, very smart and capable.

Meanwhile, Sulston and Collins and in a sense those who aided their cause in the battle with Ventor, come from institutions that also had to protect their vested interests.