Friday, September 2, 2011

Who's at risk and why?

The problem of determining causation is one of our ongoing themes here on MT, so, with that in mind, given what is known about the causes of heart disease, why is this man at risk of heart disease?  The article on the CNN website begins:
If you're not overweight, eat pretty well and exercise now and then, you might think you're in good heart health. But doctors say you don't have to look like a heart attack waiting to happen to be one.
Tom Bare, 54, is a case in point. The high school science teacher was thin, active and ate well, but still needed open-heart surgery this spring to bypass blocked coronary arteries.
The risk factors for heart disease are well-known -- obesity, smoking, high cholesterol, inactivity, diabetes, family history.  It's primarily a lifestyle disease. But Tom Bare was doing everything right, was even taking statins to lower his cholesterol (which they did, from 300 to 125), but because of his family history (probably a mix of genetic background and decades of environmental exposures), he was worried.  And it turned out he had good reason to be, given the plaque on his arteries.  So he underwent bypass surgery, and is now hoping he can avoid the heart attack that once seemed to be inevitable.

Bill Clinton has gone even further.  If heart disease is a lifestyle disease, it's clear why he had it.  He loved junk food and was overweight.  After surgery for heart disease, he has now drastically changed his diet.  He calls himself a vegan, and eats no meat, dairy, eggs or added oil.  He has lost 20 pounds and says he is healthier than he has ever been.  And is hoping he has dramatically reduced his risk of dying of heart disease. Indeed, Clinton is now following the "heart attack-proof" diet, if you believe the man selling it. 

Two men, one with the classic risk factors, the other doing everything currently considered to be protective.  Why are they both at risk?

With most common chronic diseases, causation is worked out at the population level, and then applied to individuals who will look like the average at-risk person determined in aggregate not at all, spot-on, or somewhere in between (as both Bare and Clinton, neither of whom had all the major risk factors).  Someone -- epidemiologists, clinicians translating epidemiological data to clinical practice -- will have to make a judgement call as to when a patient looks enough like study subjects with the disease, that is, has enough risk factors, to warrant intervention.

These people could be at genetic risk, but not involving known genes.  But small effects at hundreds of genes could be contributing in ways we can specifically identify but that might be reflected in the family history.  Or not.  This is the problem with genetically based prediction, and the same applies to environmental exposures, surely not all of which are identified, measured, or even accurately measurable.

You can calculate your own risk of heart disease using one of many heart attack risk calculators on the web.  Here's one -- enter a few numbers and you'll get your probability of having a heart attack in the next 10 years.

But what if the calculator tells you your risk is 30%?  The site explains that this means that 30 of 100 people with your level of risk will have a heart attack in the next 10 years.  So, does that mean that all 100 of you share the same level of risk, and bad luck or unidentified factors will push 30 of you over the edge, or that 70 of you are not in fact at risk, while the unlucky 30 are at 100% risk?  And if 70 of you aren't at risk, why are you being treated as though you are?

It's understandable that clinicians will treat anyone with some risk as though they are at 100% risk, and suggest the patient do whatever it takes to bring down that risk, but it skirts the question of how to reliably predict who will in fact have a heart attack and who will not -- something that in fact can not be known, if people change their behavior when told they are at risk, and if luck or unidentified factors are in fact at play.  From a public health point of view, that's perfectly ok, but it does make understanding causation rather murky. 

2 comments:

John R. Vokey said...

As the philosopher Judea Pearl has made clear, causality isn't simple, obvious, or commonsensical. And, as a linguistic term, is often used in incommensurate ways. It *CAN* be defined, as Pearl does in his opus, Causality, but that definition does not apparently coincide with what many (most?) apparently mean when they use the term (if they mean anything coherent at all), including many (most?) scientists. Pearl attributes that incoherence to the users of the term, not the concept itself. And, he may well be right in that.

I think Anne's points here reflect that confusion quite clearly. First, risk has nothing directly to do with causation (even in Pearl's sense): it refers merely to observed correlation between category membership and outcome with unknown causal structure. So, what does a 30% risk mean? It means a low correlation between the "risk factors" and the outcome. That's all. It is often taken to mean something causal, but without any real justification. I suspect most people, especially in medicine, treat these risk percentages as *propensities*: All those with the 30% "at risk" category membership have a greater propensity for the outcome than those with, say, a 10% "at risk" category.

The reasoning is something like this: imagine two coins: one has a .5 probability of coming up heads and the other a .8 probability. The second has a greater propensity for heads, or so it seems. The problem is that the imagining here is artificial. We can easily construct a physical model of a .5 coin (or, al least something close to it), but a .8 coin is not so easy: we now hove to posit specific *causal* forces to produce the .8 that were not seemingly needed in the simple 50/50 coin. So, they don't just differ in propensity (if at all), but rather in causal structure. Or so it seems...

Anne Buchanan said...

I agree, causality very often _isn't_ simple, obvious, or commonsensical. And yet people, from those whose aim it is to determine cause -- of anything, from disease to the London riots or the 9/11 attacks -- to those whose aim it is to interpret what it means to them (will I get cancer/heart disease/dementia?) often prefer to treat it as simple and obvious. This is good for companies selling, say, genetic risk estimates but bad for science and understanding.

It's interesting about that 30% risk. Using a number of heart attack risk estimators online, I tried hard to kick my risk up higher, giving myself extreme values for cholesterol or weight, e.g., but I couldn't get it much higher. And, as you say, 30% is actually pretty low. But, most people would consider a 60 year old male smoker, weighing 400 pounds, with a total cholesterol of 300 to be at very high risk. And in fact _some_ people with that profile are in fact at 100% risk. Who? 30% of the people? Why isn't everyone?