Strengths and Limitations of All of Us Data Use for Public Health Research

The National Institutes of Health (NIH) All of Us (AoU) Research Program, launched in 2018 with the goal of creating a national health data resource that is more inclusive of groups historically underrepresented in biomedical research, collected data from at least one million participants in the United States.¹ To date, more than 870 000 individuals have been enrolled, representing diverse backgrounds across racial and ethnic groups, gender identity and sexual orientation, disability status, access to care, socioeconomic factors, and geographic location.^1,2 Participants have predominantly been adults aged 18 years or older, but the program added limited enrollment of children from birth through age four years in 2024³ and has included over 1200 children since the program’s expansion.² Recruitment is not limited on the basis of health status¹ and is available both in person at participating health centers (including academic, Department of Veterans Affairs, and community) and online with no health care visit required. Data collected from participants include responses to health questionnaires, standardized information from electronic health records (EHRs), physical measurements, biospecimens including lab tests and genomic data, and readings from digital health technology such as wearable sensors and mobile health devices.¹ Once a participant is enrolled, new health information is captured longitudinally for the duration of their participation. The program has noted plans for participant-level linkage to external sources like death indexes, health care claims, pharmacy data, and geospatially linked environmental data.¹

Although the AoU data source is perhaps best known for its potential to advance research in precision medicine, the breadth of health information collected from a large, diverse, volunteer cohort also positions AoU as a valuable data source for public health research topics, such as population-level health trends, health disparities, and disease prevention. However, several factors should be considered when assessing the suitability of AoU data to address specific public health research questions. We present the relative strengths of AoU data for public health research, suggest potential use cases, highlight areas in which limitations may exist, and outline procedures for new users to access and analyze AoU data.

STRENGTHS AND POTENTIAL FOR PUBLIC HEALTH USE

The AoU Research Program was designed to prioritize the inclusion of persons historically underrepresented in biomedical research. To support this goal, the program uses focused recruitment strategies, including community partner outreach and oversampling of individuals in these groups.¹ In contrast to other large, national public health data sets that sample strictly on the basis of US population proportion estimates, enhanced representation in AoU data may provide more robust information and additional power for analyses of minoritized subgroups. Recent studies have leveraged this aspect of AoU data to examine health disparities and barriers to care in racial and ethnic groups.^4,5

Additionally, the lack of clinical inclusion criteria for participation allows for enrollment of participants with an extensive range of medical conditions along with healthy controls. The “all-comers” nature of the cohort supports a broader range of research topics than targeted research samples. Combining this breadth with AoU’s scale enhances opportunities for research involving rare diseases and conditions, as shown by recent genomics studies using AoU data to identify rare genetic variants.⁶

The large scale of AoU data also provides opportunities to examine population-level health trends. Recently, wearable device data from the AoU cohort have been used to study health behaviors and obesity prevalence across different demographic and geographic groups.⁷ These analyses demonstrate the potential of AoU to support population-specific analyses of behavioral and environmental risk factors and their associated disease burdens.

LIMITATIONS OF ALL OF US DATA USE

The data maintained by the AoU Research Program have documented limitations.

Although AoU emphasizes inclusion in enrollment, the program does not use probability-based sampling methods designed to ensure populationlevel representativeness. Because applying the same analytic method to differently defined cohorts can yield substantially different results,⁸ population-level inferences should not be made without applying appropriate weighting or adjustment strategies to ensure that the selected cohort is statistically representative of the US population.

Additionally, recruitment has lagged for certain minoritized groups, including specific racial and ethnic minorities⁹ and individuals with disabilities.¹⁰ Recruitment imbalances have also been observed across demographic characteristics, with overrepresentation of females and older adults (aged 60–79 years), along with other imbalances based on educational attainment, health insurance coverage status, and health-related behaviors.¹¹

Although many researchers seek to leverage AoU’s EHR data, which consist of over 300 000 participants’ diagnoses, medications, labs, physical measurements, and procedure histories, performing health-related research using EHR data is known to have a range of limitations, including validity and missingness at the person and observation level.¹² Although AoU standardizes EHR information using the Systemized Nomenclature of Medicine (SNOMED) vocabulary, variability within participating health care settings’ EHR documentation practices can hinder consistency. Additionally, longitudinal EHR data may contain left or right censoring in cases when health information is collected outside of participating health systems.

DATA ACCESS AND ANALYSIS FOR NEW USERS

The AoU Research Program supports new user registration via the Researcher Workbench platform for members of over 1300 institutions spanning academia, corporations, and smaller organizations like community treatment centers. An individualized request process exists for other researchers and requires association with a formal organization. Analysis with participant-level data is restricted to this platform; external data can be uploaded onto the platform for comparison. Costs for investigations requiring modest computational runtime are covered by free credits supplied to new users.

Researchers utilizing AoU data must follow required practices within the Researcher Workbench platform that focus on research reproducibility and participant privacy. For each new investigation, researchers must publicly declare their hypotheses and the appropriate scope of data in each AoU workspace that they are a member of. Authors must also inform AoU when an investigation has been accepted for publication, at which time an audit of the workspace may be conducted.

Within a given workspace, the web browser–based Cohort Builder and Dataset Builder include standardized processes for cohort and variable selection. Following this process, scripts are generated to pull the selected participant-level data into a coding environment. To begin exploring the data, users can indicate preference for Python-based Jupyter Notebooks, R language RStudio, and SAS Studio. High-complexity workflows and data linkages are supported through workflow management tools such as Cromwell and Nextflow. The Researcher Workbench enables common analytic procedures, including descriptive analyses, statistical modeling, and data visualization.

CONCLUSION

Although the AoU data source was created with precision medicine discoveries in mind, its scale of data collection and the diversity of the cohort lend itself to many applications in public health research. When used appropriately, data from AoU can yield valuable population-level insights that may not otherwise be available from national data sources of comparable size. The data are widely accessible and can serve as a primary or supporting source for answering a broad range of health-related questions. We encourage public health professionals to leverage these data to uplift health at both the individual and population level.