It is now well-known that the Covid-19 pandemic is disproportionately impacting Black, Indigenous, and other disadvantaged communities in the United States. Yet in the mist of the crisis, our understanding of this inequity was delayed and remains limited because many health care institutions, as well as state and federal governments, were slow to capture demographic information on Covid-19 patients. This omission is a striking example of how colorblindness and structural racism are manifested in our approaches to data science in health care and beyond.

Studies show that non-Hispanic American Indian, Alaska Native, and Black individuals were five times more likely — and Latinx individuals four times more likely — to be hospitalized for Covid-19 than non-Hispanic white individuals, after adjustment for age. Black patients are also dying at much higher rates. For example, counties where more than 86% of residents are Black experienced up to 10 times higher death rates from Covid-19. Similar trends of infection, hospitalization, and mortality will inevitably also emerge by ethnicity, language, income, and insurance status, reflecting unfair and systematic differences in access to healthy living and working conditions.

At Brigham Health, a member of the not-for-profit Mass General Brigham health system (formerly Partners HealthCare), we proactively developed a robust data infrastructure to understand the differential impact of Covid-19 on our patients and staff; visualized data through dashboards to inform our hospital operations and infrastructure; and used this data to design high-impact strategies to reduce harm caused by racism and other forms of structural discrimination. We have outlined some key lessons learned throughout this process.

Focus on a Few Key Actionable Measures

Anticipating these trends early in the pandemic, Brigham Health convened clinical and hospital leaders in diversity, equity, and inclusion (DE&I), quality improvement, and data science to develop an equity data strategy to inform and guide our incident command response. To avoid information overload, our team worked with operational leaders to focus on a few key actionable measures that aligned with organizational priorities, such as ensuring equitable access to scarce resources such as ventilators and ICU beds.

We began by visualizing our data using graphs and charts that staff and leadership could access through online software. These “dashboards” included measures such as the rate of those who tested positive for Covid-19 sorted (i.e., stratified) into different subgroups such as race, ethnicity, language, sex, insurance status, geographic location, and health-care-worker status. As the crisis evolved, we added other measures to the dashboards such as inpatient and ICU census, deaths, and discharges.

Examine Data through the Lens of “Intersectionality”

Critically, we incorporated the use of filters, which made it possible to examine the ‘intersectional’ effect of Covid-19 — a framework created by Kimberlé Crenshaw that describes how multiple social and political identities within an individual overlap and interact to create greater oppression for some groups of people due to the combination of identities (e.g., being Black and a woman). In the past, we would examine how Covid-19 differentially impacted Blacks versus whites or men versus women. The use of filters allowed us to go further and understand the differential impact of Covid-19 on Black men as compared to Black women, white men, and white women.

This powerful and underused approach revealed inequities that would have otherwise remained hidden. For example, we found that Hispanic non-English speaking patients were dying at higher rates than Hispanic English-speaking patients. Further risk-adjustment analyses then confirmed the finding and led to quality-improvement efforts to improve patient access to language interpreters.

Similarly, the geographic filter, combined with a visual map of infection-rates by neighborhood, uncovered differences in the rates people tested positive for Covid-19 by neighborhood. Historically segregated and red-lined neighborhoods of color were tested at lower rates but tested positive at higher rates compared to more affluent white neighborhoods. This finding was reinforced by state- and Mass General Brigham system-level data and local observations and knowledge indicating the same pattern. As a result, Brigham Health engaged community partners to establish free Covid-19 testing in hotspot neighborhoods; over 5,800 residents were tested over three months. These hotspot testing centers allowed staff to perform over 7,500 social-determinants-of-health screenings to assess short- and long-term needs, which the hospital helped to address by distributing to residents masks, food, grocery gift cards, and essential goods like diapers and wipes.

Tailor Data to the Audience

For leaders on the emergency preparedness and response task force (i.e., incident command), less (not more) data was intentionally shared to encourage efficient decision-making and focused action. The primary measure highlighted in the incident command dashboard for leadership was the percentage of people who tested positive for Covid-19, stratified by race, ethnicity, and language. This measure was selected by Covid-19 equity leaders to serve as a constant reminder for leadership that certain disadvantaged groups were being disproportionately impacted.

In contrast, the equity-specific dashboard created for a Covid-19 committee focused on equity contained additional measures of interest such as ICU census and mortality. In addition to race, ethnicity, and language, this dashboard also allowed users to filter the data by age, gender, and insurance type for both employees and patients. This allowed the Covid-19 equity committee members to identify and escalate emerging risks to incident command leaders, such as our finding of higher mortality among non-English speaking Hispanic patients compared to English-speaking Hispanic patients.

Don’t Forget Employee-Facing Data

Covid-19 data on employees was collected and also stratified by demographic information. This uncovered a concerning trend that some frontline employee groups — including environmental and food services, materials management, transport, patient care and medical assistants — were tested less often and tested positive at up to 10 times the rate of higher socioeconomic employee groups such as physicians and nurses. By comparing employee- and community-level data and by interviewing those with Covid-19 and their contacts (i.e., contact tracing), we found that the difference in Covid-19-infection rates between employee groups reflected transmission patterns in the community.

Additionally, we recognized that the typical means of communication utilized at our institutions (email, usually in English) was not the best means of reaching this cohort of employees. As a result, hospital leadership initiated health and well-being sessions with small groups of essential frontline employees that connected them to crisis resources, addressed questions and concerns, and facilitated timely testing for symptoms. Experts in diversity, equity, and inclusion, quality and safety, infectious disease, and human resources participated in these sessions, where materials in five languages were handed out. A total of more than 1,000 employees attended them.

Use the Existing Quality/Safety infrastructure to Identify Inequities

Research, quality improvement, and safety data were also systematically collected, tracked, and sorted by demographics. Brigham Health was a major research hub for investigating the potential benefit of Remdesivir for patients admitted with Covid-19. Our incident command team worked with our researchers to ensure that disadvantaged groups were fully represented in the study. Approximately 60% of enrolled research participants identified as patients of color and over 30% as speaking a primary language other than English.

For quality and safety, we turned to our integrated quality-, safety-, and equity-reporting system to identify possible inequities in the quality of our care through individual safety reports. We quickly sorted safety reports for Covid-19 patients, aggregated those with common themes and high acuity, and presented concerns daily at incident command for appropriate action.

For example, frontline clinical providers identified possible barriers to accessing translation services for some Covid-19 inpatients. Some of these providers also filed safety reports, leading to case reviews to identify root causes. Through this process, we discovered that our Covid-19 policy to reduce the number of clinicians entering patient rooms, in order to maintain social distancing, was making it difficult for interpreters to join at the patients’ bedside for clinical rounding on patients. As a result, incident command leadership quickly expanded virtual translation services by purchasing additional iPads to allow interpreters and patients to communicate via online software without having to be face to face.

Challenges and Common Pitfalls

Of course, challenges and limitations surfaced throughout the process. For example, the quality and completeness of disability, sexual-orientation, and gender-identity data was relatively unknown to us at the start of the crisis. Before using this data, we needed to first analyze the data to confirm that it was being recorded accurately for most patients. We found that data on patients’ disabilities in our medical records was surprisingly complete and usable, while sexual-orientation and gender-identity data was poorly documented. Unfortunately, the peak of the crisis had passed by the time we completed our assessment of the data quality. Thus, our ability to understand and address possible inequities for these patients was significantly delayed.

We also found that data was often misinterpreted by individuals who were unfamiliar with data science, health equity, or both. This required extra vigilance in what, how, and with whom data was shared. More complicated data was first processed by a smaller group of individuals with some data experience while more straightforward findings were made readily available to a wider audience. For example, the finding that communities of color were testing positive for Covid-19 at high rates was easily understood and communicated to a large audience; whereas, the problems of risk-adjusting data and managing challenges such as collinearity and over-adjustment, were worked through by a smaller group and the findings were carefully shared to avoid misunderstanding.

The Covid-19 pandemic has been a painful reminder of the urgent need to address inequities in health care and the troubling lack of progress we have made in doing so over the last few decades. For now, the lack of a standardized approach to equity data and the failure of state and federal agencies to collect and report data sorted by demographic factors means that each organization will have to make decisions for itself on what to measure and why, how and when to measure it, whom to share it with, where it is stored and how it is visualized. However, this “go-it-alone” approach is not sustainable. Universal standards, clear benchmarks, and best practices for equity data and dashboards are desperately needed if we hope to make real progress.