American Scientist

The words on the computer screen blur into a tapestry of human struggle. Housing evictions. Missed medical appointments because of a lack of transportation. Histories of trauma creating barriers to trust. Oliver J. Bear Don't Walk IV, a postdoctoral researcher in biomedical informatics at the University of Washington, had been manually annotating clinical notes from people living with HIV for use in an AI model. The goal was to train the model to automatically identify social barriers to health, such as homelessness or substance use, directly from patient records. But as they worked, Bear Don’t Walk realized that they weren't just collecting data points; they were witnessing the complexities of human lives being reduced to fit into neat boxes.

"I remember reading through those notes and encountering really difficult material—different hardships, different life events that people were going through," Bear Don't Walk says. "And it made me realize that if I'm a data scientist and I'm viewing these people as data points in a matrix, I'm really doing a disservice to them because I'm not understanding the people behind the data and their stories."

For Bear Don't Walk, as an Indigenous scholar, this experience marked a fundamental shift in their approach to data science as he began to grapple with how we capture human experiences in categories, and, more importantly, what is lost in the process.

Data Beyond Checkboxes

In modern clinics, crucial aspects of health can disappear when information is limited to narrow data fields. Constrictive forms that leave no room to capture vital realities—such as land access, cultural practices, or systemic discrimination—mask the human behind the data, and their health suffers.

These gaps have real consequences. For instance, a transgender or intersex patient might be unable to receive appropriate care because of forms that only recognize binary gender options. Or, an Indigenous person could be labeled "non-compliant" for following traditional healing practices.

Bear Don't Walk gives a personal example: "If my dad, who is Crow, goes into a clinic and says to a doctor, 'I've started practicing my culture more. I'm doing some more sweats, I'm using tobacco in ceremony, and I'm taking peyote during these ceremonies,' what happens next depends on the clinician's understanding. A culturally aware clinician might recognize these as cultural practices, while another might simply mark 'tobacco use' and 'hallucinogenic drug use' in the record." These misrepresentations would then follow his father through the medical system, priming future misconceptions in his care.

It is precisely these kinds of complex, lived realities that current data systems fail to capture. Creating systems that prevent these reductions motivates Bear Don’t Walk and his colleagues to advocate for applying what’s called an intersectional framework to biomedical informatics.

The term intersectionality was coined in 1989 by legal scholar Kimberlé Crenshaw—building on theories from Black, Chicana, Asian, and Indigenous scholars—to describe how social categories like race and gender interact to shape an individual's experience.

For Bear Don’t Walk, adopting this perspective is a matter of scientific responsibility. "My science doesn't happen in a vacuum," they emphasize. "It's impacted by the society we live in and vice versa. So, in my mind, it was my responsibility to better understand the different systems that surround my science."

A Root-Cause Approach to Health

To visualize a more holistic view of health, Bear Don't Walk’s research references an ecological framework called the “Health EquiTREE,” developed by the nonprofit Health Resources in Action with the Massachusetts Department of Health (see figure below). This model guides the essential considerations for developing intersectional research outlined by Bear Don’t Walk and their colleagues in their recent paper in the Journal of Biomedical Informatics.

The ecological metaphor, and the first opportunity for intersectional approaches in research, starts with the soil and roots—the underlying systems of privilege and oppression that shape everything above. An intersectional approach argues that instead of using demographic proxies (such as race), researchers should measure the systems themselves (such as racism). This change could be accomplished by incorporating data on factors such as historical redlining (in which certain neighborhoods were racially segregated) or a local social deprivation index, both of which quantify how systemic forces impact community health.

From the roots sprouts the tree’s trunk, representing the interconnected social determinants of health, such as housing, education, and employment. Current models often make the simplifying assumption that these factors are independent. Intersectionality, however, treats the complex interactions between factors as a primary focus of analysis, not something to be controlled for or explained away.

The trunk splits into various branches, representing how health behaviors and outcomes can vary, even within the same group. For example, analyzing the LGBTQ+ community as a single group can hide the fact that individuals who are both sex and gender minorities often face compounding health disadvantages when compared to those who only belong to one of these identities.

Finally, the branches give way to leaves, representing health outcomes such as a diagnosis of depression or high blood pressure. Modern research and health care tend to focus on this level—cataloging and treating the leaves (diagnoses) themselves. Instead, an intersectional framework encourages researchers to trace the leaves back to their roots. For example, rather than simply documenting higher rates of diabetes among Indigenous adults, researchers can use this framework to illustrate how diabetes in Indigenous populations is heavily driven by food insecurity that resulted from colonization and forced relocation.

This concept is known as an anti-categorical approach, which argues that no checkbox or label can ever capture an individual’s story or health history. In practice, one way of implementing such an approach could be to embrace narrative medicine—by using tools that allow patients and clinicians to record their experiences in their own words, capturing a fuller, more human story.

From Theory to Practice

So, how can a researcher begin to cultivate a holistic, intersectional approach to research? Bear Don't Walk and their coauthors stress that the recommendations they provide are meant to be a framework for critical thinking, not a rigid, step-by-step recipe. As they explain, "There are a lot of ways to do this, and we thought [providing concrete steps] might actually be a disservice, because if you just start using intersectionality without understanding the rich history of activism and work done by academics and activists from Black, Chicana, Indigenous, and many other groups, you're not going to truly be able to incorporate intersectionality into your research."

Instead, the authors offer core principles designed to help scientists examine their own biases and ask deeper questions throughout the research process.

Throughout an intersectional research process, perhaps the most important recommendation highlighted by Bear Don’t Walk and their team is to ensure that research drives positive social change rather than simply documenting existing disparities. This principle calls on researchers to address the soil and roots impacting community health—the cultural narratives, resource distributions, and structural barriers that fundamentally shape well-being.

As Bear Don’t Walk explains, it’s one thing to count the number of yellowed leaves on a branch and note that one side of the tree has more than the other; it's another thing entirely to analyze the contaminated soil that is causing the sickness and work to remediate it. The goal, the researchers say, is to move beyond documenting disparities and focus on why they exist, studying the forces that create inequity in the first place.

Implementing this framework is a tall order, one that requires a fundamental shift in how research is conducted. A key component of Bear Don’t Walk’s approach involves community-based participatory research. "There are lots of people with different expertise that isn't recognized by Western academia," they note. "Creating a community advisory board who have expertise I simply do not have allows me to better understand, for example, Indigenous social drivers of health."

Bear Don't Walk witnessed the importance of community collaboration themselves when developing a project focused on Native American patients in Seattle.

Although familiar with the national urban relocation programs of the 1950s, it was community partners who taught Bear Don’t Walk just how extensively these programs continue to impact Indigenous patients in Seattle today. “I didn’t understand the local context that patients here do because of their experiences,” he says. “But what I can do is talk to folks." For Bear Don't Walk, this experience demonstrated why community partnership isn't just a helpful step—it's the ethical foundation of the work. "Even if I can start to acknowledge the biases and views that I bring to the table,” Bear Don’t Walk says, “the bigger the team I can create who have different expertise, the more likely we will be able to spot issues in how I'm approaching science."

Quantifying Complexity

This work remains complex, but researchers are developing new quantitative methods to meet the challenge. Bear Don't Walk points to what’s called causal decomposition analysis as a promising direction.

This statistical technique helps researchers untangle the different causes of health disparities. It allows them to move beyond simple correlations and ask, for example, “How much of a health gap is because one group has less access to care, and how much is because that same group experiences different health outcomes even when they do get care?” By modeling how groups are exposed to varying risks and how those risks affect them, researchers can more accurately pinpoint the mechanisms of inequity.

Modeling these relationships is an evolving field of study, and it’s what drives Bear Don’t Walk’s current research. "Figuring out how to visualize and model these complex relationships," they say, "is the most exciting part for me."

A New Framework for Health Care

In the pursuit of more equitable health care, Bear Don’t Walk sees a unique opportunity for informaticists. Because they often work directly with health care systems and influence how data is collected, they are perfectly positioned to implement these recommendations in their own work, and to begin changing the perspective on how we view health. By rethinking what counts as medical knowledge and who decides what matters, health care technology could shift from a potential barrier into a powerful tool for health equity.

"Intersectionality is difficult, and I would hope that it doesn't discourage folks from incorporating it into their research, even in small ways," Bear Don't Walk says. The goal is not to achieve a perfect, objective view, but to acknowledge that no view is truly objective, and to encourage curiosity about what factors shape the health of our society.

"There is this idea in science that you can be an objective scientist, but we can't come from nowhere," Bear Don't Walk reflects. "The best thing we can do is acknowledge and reflexively examine how our personal histories impact how we're conducting science. We all have views. We all bring something to the table."