Amazon and One Medical: Minding the data gaps

When Amazon announced its intention to acquire One Medical, a primary care organization that leverages its digital health platform and brick-and-mortar clinics to offer on-demand health care services, it declared “health care is high on the list of experiences that need reinvention” as a reason for the acquisition.

Reinventing, reforming, or trying to fix any aspect of the U.S. health care system increasingly relies on analyzing patient data to improve care delivery. It’s possible that the aim of the acquisition is to get One Medical’s data. But key to ensuring equitable reform is having data from a broad range of socioeconomic backgrounds and racial and ethnic groups. Our examination of data about One Medical indicates that Amazon may have a big data gap to overcome.

Data gaps occur when certain groups are not represented in datasets. They hamper the ability to identify health inequities and design approaches that serve everyone.

One the best summaries of this concept comes from the book “Invisible Women: Data Bias in a World Designed for Men,” which describes numerous ways that a world designed for men disadvantages women. Female pianists are nearly 50% more likely to have hand injuries, for example, in part because piano keyboards were designed with the average male hand in mind.

The best way to uncover a data gap is to show that a group is being excluded from data collection efforts. A rigorous analysis for the Amazon-One Medical acquisition would involve comparing the characteristics of patients that One Medical serves to the average people seeking health care, something that isn’t possible without access to One Medical’s data. But a reasonable alternative is to look at the characteristics of the neighborhoods in which One Medical clinics are located. These can act as a proxy for the types of patients served by One Medical clinics and thus indicate which types of patients Amazon will — and will not — have data to direct its health care reinvention efforts.

We did such an analysis using publicly available data. We linked clinic locations from One Medical’s website with their Federal Information Processing Series (FIPS) Codes. These unique geographic area identifiers can be thought of as delineating neighborhoods. FIPS codes can then be linked to various data containing information about the neighborhoods. For example, FIPS codes can be linked with responses from the American Community Survey, which contains information such as average income, racial composition, and education level for neighborhoods, and to the Area Deprivation Index, which is a validated method for ranking neighborhoods by socioeconomic disadvantage.

(Note: Our analysis did not include locations of Iora Health clinics, which One Medical acquired in the third quarter of 2021, because these clinics contributed only about 5% of One Medical’s total patients but a significantly greater percentage of locations. Including Iora locations would have given these locations excess weight relative to the number of patients they served.)

The picture that emerged is that the neighborhoods in which One Medical’s clinics are located do not look much like the neighborhoods in which most Americans live.

On average, One Medical clinics are located in the top 10% of most-fortunate (or least-deprived) communities as measured by the Area Deprivation Index. Some clinics are even in the top 1% of most-fortunate places in the country.

Overall, communities with One Medical clinics are overwhelming comprised of urban, wealthy, non-Hispanic white, college-educated individuals. The table below compares how different One Medical neighborhoods are from the average U.S. neighborhood.

To be sure, there are caveats. It is certainly possible (though not likely) that the people who use One Medical's service travel from areas that are less white, affluent, urban, and educated to do so. However, financial documents published by One Medical confirm that a significant portion of its business model involves taking care of young, educated, employed individuals with health insurance, such as those working in the tech industry. It is reasonable to think that the neighborhood characteristics in which these clinics exist serve as a strong proxy for the types of patients to which the clinics cater, especially given data showing that neighborhood characteristics affect access to care.

Our conclusion is that, though Amazon has positioned itself as a leader in health care innovation, acquiring PillPack and One Medical while previously launching Haven, it will have a significant data gap with which to contend if it tries to reinvent health care at scale. While private industry can advance health care, without paying specific attention to data gaps, their contributions could exacerbate an already inequitable system.

Data gaps matter when it comes to designing health systems. Before the Heckler Report was published in 1985, which marked the first time the Department of Health and Human Services required that the health of underrepresented racial and ethnic groups be comprehensively studied, there was no formal national mandate to study health among racial and ethnic groups. Without explicitly studying how different events, like childbirth, affect different groups, few would have known that the ratio of Black-to-white maternal mortality had doubled from the 1930s through the 2010s.

Fortunately, closing data gaps makes action possible, and in 2018 Black Maternal Health Week was created to help raise awareness about the aforementioned disparities. In 2022, Black Maternal Health Week brought with it tens of millions of dollars in funding explicitly earmarked for addressing racial health disparities in maternal health.

Data gaps may not always be obvious to those leading the charge to improve health care. For example, the median age of tech company founders is 39, which is decades younger than the age groups that spend the most on health care. Without close attention, there could be a data gap as younger entrepreneurs may not know or understand the barriers older adults face when trying to access digital health solutions. One study of older adults living in an independent living facility, for example, found that nearly 1 in 4 could not figure out how to access their telemedicine platform despite nearly all being college-educated.

Minding the data gaps

Amazon and other companies seeking to improve health care can mind the data gaps. The first step is realizing they exist. Even big data will be missing certain swaths of society because people have different barriers to interacting with the economy and different abilities to utilize technology.

The next step is prioritizing health equity and including those from marginalized groups in the creative process. There are many names for this approach — community-based participatory research in the academic world or design thinking in business circles — but at their core is having diverse patient representatives to provide meaningful input in care redesign efforts and listening to those recommendations.

For Amazon to do this well, it should at least ensure that the companies in its health care portfolio serve a broad range of patients and develop an internal infrastructure to hold themselves accountable.

Overcoming data gaps shouldn't be left up to the goodwill of individuals, academics, or companies alone. There is a reason pharmaceutical companies usually bring drugs to U.S. markets first — financial incentives — and it's the same reason Amazon did not acquire a federally qualified health center, which have historically served individuals with no insurance or insurance that have lower reimbursement rates, such as Medicaid. There must be a system in place so anybody seeking to improve health care is actively incentivized to think of those most often left behind.

As companies continue to invest in health care, purposefully including underserved populations in redesign efforts and identifying ways to bolster data systems with more complete information are essential for advancing health equity.

(Editor's note: this article was updated to include an explanation for excluding Iora health clinics from the analysis.)

Alexander Chaitoff is a hospitalist and a research fellow in implementation science at Brigham and Women’s Hospital. Khin-Kyemon Aung is a resident in internal medicine and primary care at Brigham and Women’s Hospital and Atrius Health. Alexander R. Zheutlin is a resident in internal medicine at the University of Utah. Joshua D. Niforatos and Max Jordan Nguemeni Tiako contributed to the research described in this essay.