Unhealthful News 78 - Controlling for other variables, answering one question well, but not another

Today's New York Times reported on a study that is a nice addition to my recent examination of confounding.  I have to admit that I am just running with the description in the news story rather than reviewing the original study, because it is such a good example based on what was reported that I hesitant to learn more and mess up the story.

There has long been a conventional wisdom that carrying a bit of extra weight in an "apple shape" (around your midriff) puts you at higher risk of heart attack than carrying the weight in a "pear shape" (in your hips and butt).  Euphemisms aside, this claim had implications that are both diagnostic (be more worried if your extra weight is around your middle) and advisory (work harder to lose the weight if it is around your middle because you have more to gain by getting rid of it).  The new study pooled the data from 58 studies (yes, it was one of those synthetic meta-analyses that I have pointed out are often terribly flawed, but as I said, I like the story as told) and found that after you take into account the effects of other causes of heart attack (the news article mentions blood pressure, cholesterol, diabetes and smoking, and implies that there are others), the pear-vs.-apple distinction has no further explanatory power.  That is, by looking at your blood pressure, etc., you can assess your heart attack risk, and having done that, you learn nothing more by looking at where you carry your extra weight.

This is a potentially useful thing to know (assuming it is right, and there are a lot of ways it could have gone wrong).  It is also one legitimate way to analyze data for asking a specific question:  Does the apple factor have extra diagnostic power beyond what we already know?   Notice how the study is described:

Conventional risk factors like blood pressure, cholesterol, diabetes and smoking were accurate predictors of a heart attack or stroke, but additional information about weight or body shape (ascertained by measuring waist circumference or waist-to-hip ratio) did not improve the ability to predict risk.

It is important to not mistake this for controlling for confounding.  This is one of the places where epidemiology gets really complicated – far too complicated for most of the people who do it.  If we were asking the question "does apple shape cause heart attack risk?", we would want to control for confounders.  But as I have pointed out this week, this does not mean just controlling for every variable you might have, because some of them are not confounders.  To take the simplest case, it could be that having fat around your middle does more to cause higher blood pressure or unhealthy cholesterol levels than the same amount of weight elsewhere.  Thus, naively controlling for blood pressure or cholesterol is not proper controlling for confounders, because they are on the causal pathway from exposure to disease, and thus controlling for them hides the real effect of the exposure.  (As an aside, it really gets messy here because inevitably at least some of the blood pressure differences between the exposed (apple shaped) and unexposed groups represents confounding, but some might be on the causal pathway or otherwise not a confounder.  Thus it becomes necessary to figure out how to control for the right part of the effect of blood pressure, something that is quite challenging.)

So, it is critical to not make the mistake, based on information like this study that controls for everything that is a confounder or not, of claiming that a particular exposure causes, or in this case, does not cause an outcome.  That would be incorrect, and doing so would reflect a failure to understand how to do epidemiology.  The study, as described, lets us know whether there is any diagnostic value in the apple shape, given complete other information, but it does not actually answer the advisory question, "should I make an extra effort to lose this weight because it is around my middle and that causes higher risk?"

As I have tried to drill into the head of every student I have ever had:  The quality of an answer depends rather crucially on what question you are asking.

Regular readers will guess where I am going this this.  Apparently the study researchers did not understand the limits of what they had found:

 “Whatever your shape is doesn’t really matter,” said Dr. Emanuele Di Angelantonio, a lecturer at the University of Cambridge and a member of the Emerging Risk Factors Collaboration, which carried out the study.  He emphasized that being overweight or obese is one of the main modifiable risk factors for cardiovascular disease, and is often an early sign of future risk. But he said, “Whatever form of obesity or overweight you have is all the same.”

But this is not what the study actually found.  All it found was that other variables screened any effect of where you carry your fat ("screen" being a term that is sometimes used to describe the logic, "if you know these other things then you learn nothing more from knowing this" or in an intervention study, "if you fix these other values then you block -- i.e., screen off -- the effect of the one").  That does not mean that shape was not causing bad outcomes, just that it was doing so via pathways that we measured and thus picked up the effect of in the other variables.

Actually controlling for confounding, in order to figure out the effects of where you carry your fat, is much more difficult.  The fact that various studies in the past have suggested that the apple-shaped are at higher risk means that we should still consider it possible, and those with that shape should still be more worried about losing the extra weight (and, thus, those with pear weight can remain a bit less worried).  A study that controls for obvious pathways via which being fat around the middle rather than in the butt (I am sick of the euphemisms) might cause risk does not actually contradict that claim.  It cannot. 

If these mistaken claims of the study's authors, via credulous news reporters, become conventional wisdom (keep in mind that most people evaluating the claims do not know as much as you learned about proper controlling for confounding reading UN this week), the official word will to be not to worry about your fat middle.  But this will only apply if you have a complete catalog of all the other screening information and its implications, based on medical testing from your physician and parsing that information by your consulting epidemiologist (what, you don't have one of those?).  The public health advice, however, should not change because as far as we know, losing that fat middle is still more important than losing fat hips.


[A note on New York Times references:  As those of you who are avid internet news and blog readers probably already know, the NYT just announced that it is going to a semi-paywall model, wherein you have a limited number of article accesses a month unless you pay to subscribe, or link in from twitter or a few other sources that do not include this blog host as far as I understand.  In consideration of this, I will stop including links to NYT stories unless citing that particular version of the story is crucial to my analysis, which it usually is not, and will make sure to note when a link is to the NYT so you do not click and spend one of your monthly allotment without knowing you are doing so.]