Moving Beyond the Cause Constraint: A Public Health of Consequence, May 2018

Both of us have been involved in decades’ worth of population health science writing that has, at times, sidestepped the issue of whether a condition, factor, or circumstance of interest was indeed a “cause,” sometimes nudged by coauthors, other times by editors, others by reviewers, and still others by our own timidity, as we sought to avoid unnecessary argument.

CAUSE AND CONSEQUENCE

It is this timidity that an excellent and provocative commentary in this month’s AJPH by Hernàn (p. 616 and p. 625) seeks to dispel. Hernàn argues that we should use the word “cause” when we mean cause and persuasively and most provocatively outline the dangers that emerge when we avoid the word “cause.” Several other articles in this issue (see Begg and March [p. 620], Ahern, Chiolero [p. 622], Glymour and Hamad [p. 623], and Jones and Schooling [p. 624]) elaborate on this.

When we focused on promoting a public health of consequence,¹ it was likely unintentional that our language and thinking returned to the notion of cause² in public health; to actually move the needle on population health, we have to move our thinking, our research, and our actions from describing disparities to understanding the causes of them. Reflecting on the arguments introduced by Hernàn, we wonder why “cause” has become such a challenging word to use in our science. We offer here three potential explanations to aid the understanding of what we do and how we do it and, hopefully, to keep our thinking focused on the causes and not find contentment with descriptions and associations.

CAUSE RELUCTANCE

First, our concern with using “cause” emerges from an overabundance of caution about the limitations of any particular statistical approach to tell us that one cause is the active ingredient that influences the outcome, all other factors being held equal. This concern is tied closely to efforts to isolate causes, replicate assumptions that are achievable only in idealized randomized control trials, and determine what the cause may be so that we may interact on that one cause. This approach, an effort at causal isolationism, has pushed us ever deeper into anxiety about identifying a particular factor as a cause, because we of course recognize that other factors may be equally meritorious causes.³

This has implications that extend far beyond the methodological confines of any particular discipline and echo efforts in the broader public discussion about how we act to improve health. Our consideration of causes of death, to provide an obvious example, as being centrally behavioral⁴ effectively constrains our spending and commensurate research focus on efforts to modify behavior, neglecting the more upstream social causes of death that are equally responsible for the generation of health and disease.⁵ Loosening the reins on our use of the word “cause,” embodying an appreciation of the multiplicity of causes, stands to create space to introduce the foundational drivers of health in a national health causal conversation, a change that would be welcome indeed.

Second, our reluctance to use the word “cause” has come from the conflation of the act of causal thinking and the application of statistical methods to observed data that collectively have come to be called “causal modeling” approaches. In some respects, this error is not unique to cause and causal thinking. When multilevel modeling emerged as a common useful technique in population health science more than 15 years ago, its widespread adoption also came in some respects at the expense of multilevel thinking.⁶ The adoption of statistical approaches that are new and unfamiliar may, reasonably enough, occasion anxiety, perhaps fueled by the relatively small number of scholars who, at first, are familiar and comfortable with the relevant methods. This anxiety about new methods may be unavoidable. It seems incumbent on the early practitioners to do their part to help us not conflate unease with statistical methods and causal approaches with a solid understanding of the broader notion of causation. Simply put, an appreciation of causal thinking does not depend on causal modeling, and unfamiliarity with the latter need not scare away engagement with the former.

Third, the reluctance to use the word “cause” represents the long tail diffusion of the science and the attendant loss of nuance that is well accepted among leading thinkers in the field as methods and approaches are adopted by the broader mass of scientists and practitioners. Hernàn is saying little new in his editorial; the ideas he brings up have long been recognized and discussed in other articles by leading methodologists, many of whom Hernàn cites. The authors of these articles recognize the challenges inherent in causal inference, the potential and limitations of our methods that can help inform our causal thinking. The authors, additionally, realize that this does not mean that we need to throw the cause baby out with the bathwater of overinference from limited studies. However, the “association is not causation” crutch—perhaps emerging as an antidote to overgenerous interpretations of work that we should not hinge much on—has long spread throughout population health science, has been echoed by the lay press, and, alas all too frequently, has been adopted by reviewers. This oversimplification has gone on to chill the use of “cause” by many throughout the field, leading to the challenges that Hernàn articulates so well. This calls for ongoing, persistent, and relentless communication by leaders in the field about the nuance that should inform our thinking and a refusal to allow simplistic prohibitions to spread without challenge. That Hernàn issues his challenge in the pages of AJPH is laudable; we all probably should have risen to the challenge sooner.

STRIVE FOR CAUSAL THINKING

A recent book defined population health science as “the study of the conditions that shape distributions of health within and across populations, and of the mechanisms through which these conditions manifest as the health of individuals.”^7(p.1) In many ways, this definition is uncontroversial. Surely it is the job of the science of population health to understand the drivers of population health to the end of us intervening and being able to improve the health of populations. The definition should also be uncontroversial in another way. It suggests that population health science is the study of conditions that shape distributions of health and considers how these conditions manifest as health. One imagines that the word “cause” could substitute for both euphemisms in both places in this definition. But the elision of the word “cause” in this definition was a nod to the intellectual baggage it carries and a desire not to have that intellectual encumbrance become the focal point of arguments about whether these conditions are “causes”—an argument that would have been beside the point of what the definition was trying to achieve to begin with.

We consider the move to reintroduce causal language to population health science welcome and long overdue. The task at hand is to ensure that the language of our writing, and hence of our thinking, accurately reflects what we are trying to do. When we are clear that we are studying causes, we open up the opportunity to identify and act on them. Anything less limits the reach and scope of population health science.