Tolerating Uncertainty About the Communication of Risk

Early in my medical career, I was frequently impressed yet discomforted by the power of risk communication. Back then, little of it was done precisely—using numeric risk estimates—but I saw firsthand how mere words expressing a “high” or “low” chance of cure, a “good” or “poor” prognosis, could change people’s minds and the subsequent course of their illnesses and lives. Adding to my discomfort was the routine, nondeliberative way we communicated risks, with little explicit reflection or attention to the basic question: What is the best way to communicate?

Fortunately, things have changed since then. An entire interdisciplinary field of research on risk communication has developed, generating a large body of evidence on the outcomes of different risk communication strategies. Informed and shared decision making have arisen as powerful health care ideals, stimulating the development of new risk communication strategies and training interventions for clinicians.

Despite this great progress, the question remains of how best to communicate risk in health care. Now, however, the problem is a lack of clarity about how to translate existing evidence into practice. To this end, the Making Numbers Meaningful initiative is a hopeful undertaking. It rigorously synthesizes findings from diverse studies of health risk communication (I will use this term given its broad currency, although the authors justifiably prefer “probability communication”), organizes these findings within a useful taxonomy, grades the quality of existing evidence, and details how different communication strategies affect various outcomes. ¹

The result is a comprehensive resource that can help health care professionals understand the potential impact of different risk communication strategies and make better informed choices between them. Its further value, in my view, lies in its overall conclusion: there is no single “best” way to communicate risk information. The authors identify several reasons why, including insufficient evidence and methodological limitations of existing studies. Notably, many of even the strongest studies have been conducted among healthy individuals reacting to hypothetical medical scenarios, potentially limiting their generalizability. Compounding this problem is the review’s analytic approach, which involved breaking down communication strategies into key elements (e.g., data structure, presentation format), subdividing them into different types (e.g., graphical, numerical, and verbal data presentation formats), and assessing their effects when applied to different cognitive tasks. ¹ This reductive, “granular” approach, as the authors put it, clarifies independent but not additive effects of individual message elements and may thus have less practical value for integrative, real-world risk communication efforts. Finally, as the authors emphasize, risk communication strategies have diverse effects that may be desirable or undesirable depending on the goals of communication. This means there is no single best strategy, only better or worse strategies for different individuals and situations.

So where do we go from here? The current review, like the risk estimates we strive to communicate, leaves us with both certainty and uncertainty. It promotes confidence about the effectiveness of various risk communication strategies but also raises new questions about their appropriateness. It identifies promising directions for future research but also pushes us to find other ways of managing and tolerating our uncertainty about risk communication. This latter outcome, I believe, is an especially important contribution of the review, for the reasons identified as well as a more fundamental obstacle to identifying a single best risk communication strategy: the limited applicability of risk information to individual persons.

This is not strictly an empirical problem arising from a lack of evidence or an ethical problem arising from conflicting communication goals. It is also an epistemological problem arising from the multiple meanings of probability and their incommensurability. A primary meaning, the objective interpretation, construes probability as either a property or factual description of the natural world. In this interpretation, probability estimates are formal representations of uncertainty about the future, derived from and expressed in terms of the expected frequency of a given outcome over time or across a population. An alternative meaning of probability, the subjective interpretation, construes probability as a mental state. In this interpretation, probability estimates are expressions of individuals’ personal confidence or degree of belief about the future, revealed by their preferences, judgments, or decisions.

Formal risk communication efforts in health care are predicated on the objective interpretation and the assumption that objective probabilities should correspond to individuals’ subjective probabilities; this correspondence is a principal empirical measure of risk understanding. However, several problems limit its achievable degree and normative significance. Objective probabilities represent mere statistical averages. Their accuracy is restricted by imperfections of all empirical studies as well as unmeasured—and unmeasurable—causal variables. A more fundamental problem, however, is the logical incoherence of objective probabilities when applied to single events experienced by individuals. ² As philosopher Ian Hacking observed, “It does not make sense to speak of the ‘frequency’ of a single event.” ³ No matter what proportion of a population might experience the event, the true risk for a given individual is neither knowable nor expressible in objective terms. Richard von Mises, proponent of the objective interpretation, made this point using a health-related example: “We can say nothing about the probability of death of an individual even if we know his condition of life and health in detail. The phrase ‘probability of death,’ when it refers to a single person, has no meaning at all for us.” ⁴ At the individual, single-event level, probabilities are necessarily subjective expressions of the strength of one’s beliefs, which have no single, verifiably correct level. Bruno de Finetti, proponent of the subjective interpretation, thus famously declared, “Probability does not exist.” ⁵

I briefly raise this well-known problem not to diminish the value of objective probabilities or efforts to communicate them. Objective probabilities are the only means available to translate scientific evidence into practice and inform individuals’ subjective probabilities, which might otherwise be based on noncredible evidence. Yet the epistemological challenges they raise are often lost in translation; objective and subjective probabilities tend to be conflated or regarded as fungible, and discrepancies between them as irrational. The tacit assumption behind the search for a single “best” way of communicating risk is that there is a single right way of understanding risk. My own experiences over the years, both professional and personal, have taught me how naïve this assumption is. I have seen how wrong objective probabilities can be, how inappropriately they are often applied, and how damaging they can be when taken literally—leading to both false alarm and false reassurance, endless information search and profound decisional regret. But objective probabilities satisfy a deep human craving, among both health care recipients and providers, for certainty—the feeling of knowing. In the face of uncertainty about the future, they provide something tangible—concrete, visualizable, precise—on which to stake our decisions and lives. Nevertheless, the idea that an objective probability is my or your “real” probability, that such a thing exists, is a myth. As useful as this myth might be, objective probabilities are imperfect abstractions, their precision a mere mirage.

I believe these problems push us to rethink the nature and goals of risk communication, as the title of the current initiative, Making Numbers Meaningful, implies. Risk communication should ideally be not a passive, 1-way transmission of objective probabilities but an active, broader meaning-making process as Edwards, Elwyn, and Mulley have defined it: “the open two-way exchange of information and opinion about risk, leading to better understanding and better decisions about clinical management” ⁶ (emphasis mine). The goal, in this view, is to enable individuals not to discover their “true,” objective probability—that is a myth—but to construct subjective probabilities by integrating objective probabilities with opinions and other forms of evidence. This broader iterative process, which one could call the “subjectivization of objective probability,” ² might employ multiple strategies directed at different goals (e.g., informing, persuading) to enable individuals to approach uncertainty from different perspectives, to make meaningful numbers. A comprehensive inventory such as the current review could be a valuable tool in this process.

Of course, not all health risk communication efforts afford the opportunity for iterative exchange; brief risk messages or infographics in a public service announcement are 1-way propositions. But I wonder whether even such messages could somehow be redesigned or integrated within broader efforts to foster 2-way dialogue and active co-construction of subjective probabilities by health care professionals and laypersons. Perhaps messages could be developed that communicate not only objective probabilities but also their value and limitations, and the legitimacy and necessity of integrating multiple forms of risk evidence. The overarching goal would be to promote an epistemological perspective that normalizes uncertainty and acknowledges the lack of single right answers, and to reinforce particular moral virtues or character strengths—including humility, flexibility, and courage—that help people tolerate uncertainty. ⁷

Such messages might help patients struggling to understand their own risks but also health care practitioners struggling to understand how best to communicate risk. Explicitly acknowledging that there is no single best risk communication strategy may encourage practitioners and researchers to work together to co-create the best strategies they can, and to accept that uncertainty and discomfort about not only risk but the communication of risk can never be eliminated, only managed and tolerated. The critical need moving forward is to develop integrative communication approaches that can reinforce the habits of mind and character that these challenging endeavors require.