Health Care’s Social Movement Should Not Leave Science Behind

Americans are dying at startling rates, not only from COVID-19 but also its reverberating socioeconomic effects: massive unemployment, food insecurity, social isolation, and stress-related increases in smoking, alcohol use, opioid use, and suicide. To save lives, we need scientific advances, not only in biomedicine—virology, vaccine development, and critical care—but also in interventions that the address social determinants of health. Yet, at a time when a scientific approach is needed more than ever, science has been a key political battleground during the pandemic.

Denial of biomedical science is a much-discussed phenomenon,[1] but health care’s tendency to ignore socio-behavioral science is perhaps even more common and insidious.  In health care, socio-behavioral interventions are often developed via trial and error rather than formative science. For instance, it is common for hospitals and clinics to develop de novo programs to address patients’ unmet social needs and promote healthy behavior. Further, many initiatives are funded, implemented, and scaled without strong evidence, or even in the face of null findings. Light-touch technology-platforms that screen patients for unmet social needs, for example, have received hundreds of millions of dollars in investment capital, but well-designed studies demonstrate low uptake and minimal impact as a standalone solution.[2][3] Multidisciplinary care teams for “super-utilizers” are also common, yet several studies have failed to demonstrate the intended effect of reduced health care utilization.[4][5] Finally, domain-specific interventions like ride-sharing to address transportation barriers[6] are intuitive, but have had null findings. None of this is to say that these approaches cannot be refined, combined, and retested for effectiveness. However, dissemination has currently outpaced evidence, possibly crowding out interventions that are evidence-based.

It is important to examine why science has played a less central role in the development and testing of socio-behavioral interventions, and what needs to be done differently. Although the focus here is on interventions, it is important to recognize that socioeconomic policies—such as minimum wages and paid sick leave—can be even more powerful in addressing social determinants of health, and should also be informed by science.

“You Don’t Need a Randomized Trial to Feed a Hungry Child”

The way we think about socio-behavioral programs is shaped by powerful cognitive biases that promote belief over skepticism.

The first is confirmation bias, a tendency to reinforce what we want to believe. The inhumanity of health-related social needs—such as one in three Americans facing food insecurity—creates an appropriate sense of moral urgency. Well-resourced opportunities to correct injustice are rare. Therefore, we want to believe that a plausible-sounding intervention works; this can tempt us to forgo valid skepticism of morally laudable efforts.

Familiarity bias is also at play. For example, decision makers might yield to experts when considering CRISPR gene therapy but may overestimate their own judgment in “everyday” issues, such as food, housing, or social connection. Unlike biomedical therapies, solutions to these problems seem obvious. But even though it doesn’t take a randomized trial to know you should feed a hungry child, it may take one to know how best to do so.

A third and powerful factor is narrative bias, our tendency to make sense of the world through stories. This bias is fueled by the storytelling often found in media reports of socio-behavioral interventions. We tell and retell success stories of people helped by programs, which creates belief in programs that may or may not be truly effective on the whole.

User (Dis)empowerment

Many biomedical interventions have a broad user base. In contrast, users of socio-behavioral interventions may be disadvantaged or oppressed individuals, who are more likely to be ignored when they raise the alarm about ineffective or harmful interventions. And less empowered users also means fewer resources driving the pipeline of scientific discovery.

“Data-driven” Decisions

Information previously restricted to experts now abounds, allowing decisions to appear data-driven even when divorced from scientific design and interpretation. The scientists who could inform these decisions are increasingly viewed as elitist or irrelevant in an anti-science era when growing numbers of laypeople and health care decision makers alike claim that their personal opinions hold equal weight to those of content experts.[1]

In health care, one example of this trend is an over-reliance on pre-post data. Socio-behavioral programs often target outliers; outcomes are tracked over time and improvements interpreted as program success. Health care leaders can see these trends in their own systems, and thus are likely to believe it. But, without a scientific design that includes appropriate comparison, they are likely observing the greatest mirage in evaluative science: regression to the mean.

The Fraught History of Health Science

The enterprise of health science has, for decades, pointed squarely at biomedicine and pharmacotherapy. When health science has focused on the social aspects of health, it has often been to describe, rather than address, disparities. And health science has been guilty of worse things than indifference, with racism and other ethical breaches from Nuremberg to Tuskegee.

For these reasons, community advocates may think of health science as frustrating at best and exploitative at worst. Skepticism can be worsened when health scientists forego participatory methods that involve members of an intervention’s target population in its design, or ignore lessons from fields like social epidemiology and education about how best to engage participants.

How Science Can Advance Health Care’s Social Movement

The high stakes of socio-behavioral interventions create an imperative for science. Other frame shifts in health care (e.g., digital health, precision genomics) have room to fail and iterate because they are of interest to those with wealth and political power. Conversely, the resources for socio-behavioral interventions will probably dry up unless they can offer real impact. We offer three recommendations for ensuring that science guides this work.

First, socio-behavioral interventions need to be grounded in formative science. When possible, organizations should adapt existing evidence-based interventions instead of reinventing the wheel. If needed, new interventions should be developed only after review of socio-behavioral science literature and contextual inquiry, using methods that explicitly involve the population targeted by an intervention in aspects of its development, testing, and dissemination(for example, community-engaged participatory research). This discipline will lead to better interventions. It will also bridge the chasm between designers of socio-behavioral programs and end-users affected by social injustice.

Second, decision makers should use evidence grading systems for socio-behavioral interventions. This would facilitate scale-up of interventions with the strongest likelihood of success, promote research designs with low risk of bias, and uncover knowledge gaps for further scientific inquiry. It would also expose the risk of harm from socio-behavioral interventions, a topic that has been underexplored. It is understandable to have a bias for action when people are suffering, but it is best to do so with eyes wide open.

Finally, health care organizations should do less pre-post analysis and more randomized studies. Newer methods—pragmatic trials, cross-over designs, stepped wedge designs, and randomized quality improvement—can generate high-quality information for decision makers while respecting important ethical and practical considerations. New York University’s Langone Health system recently completed ten large-scale randomized quality improvement projects in one year, substantially improving their decision making.[7]

The biomedical dimension of health care has not always been grounded in science.  In 1910 perhaps it felt wrong to withhold heroin from patients with consumption, just as it now feels wrong to pause and consider investments in addressing hunger or homelessness. Yet, in retrospect, scientific discipline has driven many improvements in health over the past century. Individuals of any profession or political persuasion can fall victim to the biases noted above; they are commonplace. However, given the enormous consequences of our current decisions and investments, the need to reaffirm science-based policies and programs in health, whether biomedical or socio-behavioral, has never been greater.  As we have already seen during the COVID-19 pandemic, a denial of science-based programs and policymaking will be measured in lives.

References

[1] Nichols TM. The death of expertise: the campaign against established knowledge and why it matters.
[2] Wu AW, Hwang S, Weston CM, et al. Baltimore CONNECT: A Randomized Trial to Build Partnership Between Community Organizations and a Local Health System. Prog Community Health Partnersh. 2018;12(3):297-306.
[3] Lindau ST, Makelarski JA, Abramsohn EM, et al. CommunityRx: A Real-World Controlled Clinical Trial of a Scalable, Low-Intensity Community Resource Referral Intervention. Am J Public Health. Apr 2019;109(4):600-606.
[4] Finkelstein A, Zhou A, Taubman S, Doyle J. Health Care Hotspotting – A Randomized, Controlled Trial. N Engl J Med. Jan 9 2020;382(2):152-162.
[5] Iovan S, Lantz PM, Allan K, Abir M. Interventions to Decrease Use in Prehospital and Emergency Care Settings Among Super-Utilizers in the United States: A Systematic Review. Med Care Res Rev. Apr 26 2019:1077558719845722.
[6] Solomon EM, Wing H, Steiner JF, Gottlieb LM. Impact of Transportation Interventions on Health Care Outcomes: A Systematic Review. Med Care. Jan 23 2020.
[7] Horwitz LI, Kuznetsova M, Jones SA. Creating a Learning Health System through Rapid-Cycle, Randomized Testing. N Engl J Med. Sep 19 2019;381(12):1175-1179.

About the Authors