Allocating health outcomes to risk factors, part 1
By Austin Frakt
In 2017, Nancy Krieger, Professor of Social Epidemiology at the Harvard T.H. Chan School of Public Health, published a truly insightful paper in the American Journal of Public Health in which she raised several conceptual problems with allocating health outcomes to contributions from risk factors.
This is the first of three posts that will unpack some of the content of her article. It is co-authored by Austin Frakt and Sherry Glied.
There have been many efforts to attribute the extent to which a health outcome (like death) is caused by broad factors. For example, some widely cited figures derive from work done by the Centers for Disease Control and Prevention in the in the mid-1970s that concluded,
[R]isk factors due to lifestyle contributed approximately 50 percent to United States premature mortality in 1977, followed by environmental factors and human biological factors with 20 percent each, and finally inadequacies in the health care system with only 10 percent.
You’ll notice these sum to 100%. Moreover, not a word is said about how the factors interact to contribute to deaths. These are significant, and related, problems.
Krieger’s paper* is titled “Health Equity and the Fallacy of Treating Causes of Population Health as if They Sum to 100%.” That’s a big clue about one of her main points: If we identify risk factors for a health condition and measure the extent to which each factor is associated with it, we should not expect them to sum to 100%. If we force them to sum to 100%, we’ve made a mistake.
To understand what’s wrong with allocating contributions to health outcomes across mutually exclusive risk factors and forcing their contributions to sum to 100%, consider a simple, hypothetical example. Imagine a world composed of exactly half men and half women. Exactly half the men have red hair, half have brown. Same for women.
Life was blissful in this world until a deadly disease we’ll call “Z” broke out. Z was so devastating to families and communities, there was a public outcry for research and intervention.
Within a few years, Z researchers discovered that being male or having red hair each made individuals equivalently susceptible to Z; no other risk factors could be identified. Though some public health experts referenced some mathematics in support of the claim, it’s intuitively appealing to attribute contracting Z equally to sex and hair color: 50% due to sex and 50% due to hair color. This information was widely disseminated and accepted as gospel. As we will see, it was misleading.
Because this world had a very responsive public health policy apparatus, substantial investments were made in both sex and hair color alteration. Appealing to scientific evidence available at the time, the world leader stated, “Since half of Z is due to sex and the other half to hair color, if you want to address 100% of the problem, you have to do both.” And so it was.
Subsequently, problems with the allocation of Z risk factors began to emerge. One clinical trial showed that if you genetically change all men to women, 100% (not 50%) of Z is eliminated. (Yes, we’re imagining a world in which this is possible.) Another showed the same thing from genetically changing all redheads to brown-haired. (Yes, this too is possible in this hypothetical world.)
If each risk factor is responsible for only half the problem, how can changing just one or the other address it entirely? The answer, which emerged slowly from the scientific community, is that Z is actually associated with being both male and redheaded. Neither one alone is sufficient for contracting the disease; changing just one will prevent it. In a sense, both are 100% responsible. If we’re compelled to add them up, we might say the risk factor contributions sum to 200%.
Unfortunately, this knowledge was obtained only after a great deal of resources had been spent in trying to address both sex and hair color. Sex in particular was problematic to change and had significant consequences for population growth and economies. It would have been sufficient to only have addressed hair color, at far less economic and social cost.
Though this story is hypothetical, there are some real-world examples with similar characteristics. For many diseases, a lot of things have to go wrong simultaneously to cause a problem. To indicate that just one or another is only partly responsible misses this crucial point.
Consider, for example, death due to tobacco-related lung cancer. Genetic susceptibility plays a role, as not all smokers who are otherwise similar get lung cancer.
Though necessary, genetic susceptibility is not sufficient to acquire tobacco-related lung cancer, of course. There needs to be a behavior — exposure to cigarettes. Then, if one gets lung cancer, the health system plays a role. Medical treatment needs to fail (either to be accessed, delivered properly, or to be effective) for one to die from it. If any one of those things is missing — susceptibility, exposure, health care imperfection — there can be no death from tobacco-related lung cancer.
Most dangers in life are like this. Multiple points of failure are required for bad outcomes. If this weren’t so, we’d all be dead. One might say, “Successes are alike, but every failure is different in its own way.”
Being able to attack and address a problem from multiple angles has huge implications for policy. Sometimes the most efficient and effective way to attack the problem is to look for the underlying genetic susceptibility, sometimes it is to change behavior, sometimes it is to address medical care. Sometimes, a problem is most efficiently addressed through a multi-front attack.
But efficiency is not equity. The choice about how to address a problem also affects the distribution of winners and losers. For whom will behavioral change be most feasible? Who can best access advances in medical care? And so forth.
The bottom line is that though a list of mutually exclusive risk factors that sum to 100% can be very useful for some purposes, it is important to recognize that it is not appropriate for all purposes. At the very least, it is an incomplete guide as to what is the most efficient or equitable way to address a problem.
* There is an erratum associated with her paper that corrects errors in numerical examples. The main points of the paper are unaffected, however. The paper also received a critique to which Krieger responded.