Introduction

Health databases are comprehensive repositories that store patient health data collected from different healthcare encounters, including health administrative databases, electronic health records (EHRs), electronic medical records (EMRs), clinical administrative data, and health information systems [1]. Notably, EHRs and EMRs are often used interchangeably, however, EMRs typically describe records within a single healthcare organization, while EHRs are shared across multiple settings to provide comprehensive patient care [2]. In general, these databases often capture details regarding hospital admissions, diagnoses, treatments, and demographics that can serve as important tools for health system performance monitoring and research [3–5]. Capturing social determinants of health, such as race, supports equity-focused research, system performance and tailored clinical care and policy [6]. However, race data are often not collected, incomplete, inconsistently reported, or oversimplified [6]. The capacity for health databases to support equity-focused research is therefore limited and may obscure insights into systemic health disparities [6,7].

Despite being historically misunderstood as a biological concept, race is now widely accepted as a social determinant of health, with increasing attention focused on addressing its role in perpetuating health inequities [8,9]. Race often refers to categories of people who share perceived physical characteristics that are socially constructed with meaning. The classification is rooted in discrimination and social oppression to maintain hierarchies of privilege and power [8,10]. Conversely, ethnicity refers to shared cultural identity, social practices, and heritage, such as language, religion and customs [8,10]. Recognizing this difference is critical for designing equitable health interventions and policies. Unfortunately, the current literature uses the terms “race” and “ethnicity” inconsistently, which makes it challenging when designing equitable health interventions and policies. In efforts to clarify the confusion, this review focuses on the definition of "race" while reporting the authors’ original terms and instances of conflation [11,12].

Racialized populations face disproportionate health burdens that are shaped by structural and systemic inequities, including discrimination, food insecurity, and unequal access to healthcare [7]. For instance, in Canada, chronic conditions such as diabetes are 2.3 times more prevalent among South Asian adults, 1.9 times more among Black adults, and 1.8 times more among Arab and West Asian adults compared to White adults [13]. Mental health disparities are also evident, where Southeast Asian and Arab adults are less likely to rate their mental health as excellent or good, and lower life satisfaction is reported among West Asian and Black populations. Socioeconomic inequalities compound these disparities [14]. Food insecurity is 2.8 times more common among Black adults, and core housing need is over twice as high among Arab, West Asian, and Black Canadians [15]. Indigenous groups also experience some of the highest unemployment levels in Canada [16]. Discrimination within healthcare further contributes to inequities, where approximately 50% of racialized Canadians reported unfair treatment when accessing services, including being dismissed by healthcare providers, receiving less thorough assessments, or feeling disrespected during clinical encounters [17,18]. These patterns underscore the urgency of incorporating race data into health research to inform inclusive and effective interventions.

Race is increasingly understood as a fundamental variable in health research, particularly for the detection and reduction of health disparities [19]. Yet, the collection and analysis of race data into health databases remains inconsistent across countries [19]. Some health systems collect race information directly through EHRs, while others rely on surveys, geographic proxies, or algorithmic estimates based on names or locations [19,20]. These inconsistent practices differ in quality and reliability, often leading to misclassification, oversimplification, and non-comparable racial data across settings [20]. The lack of standardized frameworks, such as overgeneralized racial categories, missing data, and the absence of individual-level information, limits the ability to identify meaningful within-group differences, evaluate target interventions, and monitor longitudinal disparities in marginalized populations [21]. In health systems and research, racialized populations remain underrepresented due to variable methods of race data collection and reporting in health databases [20]. As a result, our understanding of how systemic racism and social determinants impact health outcomes for minority populations is hindered [22,23]. Strengthening the consistency and accuracy of race data across health databases is therefore essential to support equity-driven research, policy development, and system-level change.

In summary, most of the literature focused on describing disparities rather than evaluating how race information was identified or classified. To address these gaps, this rapid scoping review study aimed to 1) examine recommendations for race identification in health databases, including best practices, frameworks, guidelines, and protocols for data identification, collection, and accuracy, and 2) determine gaps in literature on recommendations for identifying race in health databases to inform further research.

Methods

Results

Study selection

The database search yielded a total of 7072 articles, with 3041 duplicate studies removed, and 4031 remaining following de-duplication. At the title and abstract level of screening, 3898 articles were excluded, leaving 133 for full-text screening. Of the 133 full-text articles screened, 112 were excluded. A total of 21 studies were included in the data extraction and analysis. Fig 1 represents the PRISMA flow diagram documenting records identified, included, and excluded.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. PRISMA-ScR flow diagram of articles included in the scoping review during the study identification, screening, and inclusion processes.

https://doi.org/10.1371/journal.pone.0339025.g001

Study characteristics

A summary of the study characteristics (author, year, country, objective, study design, health database source, and racial categories analyzed) and recommendations can be found in Table 1. Of the included studies, nearly all were conducted in North America (n = 20, 95%) [20, 26–44] in either the United States (n = 19, 90%) [20,27–44] and Canada (n = 1, 5%) [26]. The remaining study spanned multiple European countries, specifically Belgium, France and the Netherlands (n = 1, 5%) [45]. In terms of publication year, the fewest number of studies were published in 2020 (n = 1, 5%) [29] and 2025 (n = 1, 5%) [26], followed by 2021 (n = 2, 10%) [28,33], 2022 (n = 3, 15%) [37,42,43] and 2024 (n = 3, 14%) [34,38,40], 2019 (n = 4, 19%) [20,31,36,41], and the most published in 2023 (n = 7, 33%) [27,30,32,35,39,44,45].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Summary of Study Characteristics and Extracted Data from Included Articles.

https://doi.org/10.1371/journal.pone.0339025.t001

All included studies employed qualitative or observational methodologies [20,26–45]. The most common study design was validation studies (n = 9, 43%) [20,27,32,34–36,38,41,43], followed by cross-sectional studies (n = 3, 14%) [26,31,42], scoping reviews (n = 2, 10%) [30,45], and systematic reviews (n = 2, 10%) [28,39]. The remaining articles were associated with other approaches (n = 5, 24%) [29,33,37,40,44], such as organizational policy statements or descriptive implementation reports. The included studies assessed a range of health databases, with the most common being EHRs (n = 9, 49%) [27,29,31,35–38,40,42], followed by health administrative databases (n = 5, 24%) [20,32,34,41,43], and EMRs (n = 1, 5%) [26]. Some studies engaged with other databases (n = 3, 14%) [30,33,44] and a few did not examine a specific database (n = 3, 14%) [28,39,45], as they focused on reviewing literature or discussing conceptual frameworks.

Approaches to identifying race and ethnicity varied across the included studies. Seven studies used the five-category classification based on the United States Office of Management and Budget framework, which includes American Indian or Alaska Native, Asian, Black or African American, Native Hawaiian or Other Pacific Islander, and White [20,32,33,35,36,39,42]. Furthermore, six studies adopted an updated seven-category framework that added Hispanic and Multiracial as distinct groups [28,29,34,39,40,44]. Notably, six studies grouped race and ethnicity into a single variable [26,31,37,41,43,45]. One study focused specifically on Indigenous populations [30], and two studies discussed identity-related concepts however did not employ a specific framework to organize race data collection (e.g., immigration status or foreign-born populations) [27,45]. These studies were retained as they provided relevant recommendations for improving race data analysis, despite discussing identity more broadly. This variation highlights ongoing challenges in standardizing race and ethnicity data collection across datasets and research contexts.

Key Recommendations

Five overarching recommendation categories were inductively identified across the included studies: (1) Self-identification and patient-centered practice [20,27,30,33–35,38,40,45], (2) Standardization across healthcare systems [26,34,35,39,40], (3) Data quality and completeness [20,26–31,36,37,39], (4) Algorithmic and predictive methods [28,32,37,39,41–44], and (5) Disaggregated data use and cross sector collaboration [27,29,32,33,37,39,40]. The overlapping and interdisciplinary nature of efforts to improve race data in health databases was evident as numerous studies addressed more than one recommendation category. The key recommendations are summarized in Table 2, which describes each category alongside the corresponding articles and their examples.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Key categories of recommendations described in included studies.

https://doi.org/10.1371/journal.pone.0339025.t002

Discussion

Many important findings were identified in this rapid scoping review. First, self-identification and person-centered practices were described as the gold standard for race data collection, emphasizing patient autonomy and lived experiences [20,27,30,33–35,38,40,45]. Notably, many articles emphasized improvement to race data collection at the point of care [20,26,27,33,34,38,40], such as self-reported patient intake forms or staff-administered surveys, rather than methods that refine or supplement race data after it had been recorded. Second, multiple studies highlighted the importance of transparency in how data are sourced and reported, particularly in distinguishing between self-reported, inferred, or algorithmically imputed data [20,27,39]. Third, studies identified standardization across systems to be essential for having consistent categorization of race and enabling reliable comparisons across databases [26,34,35,39,40]. Finally, strategies to improve data completeness, such as the use of supplementing sources and predictive tools, were proposed to enhance existing datasets when self-reported data were not available [26,28,41–44]. Altogether, these findings highlight opportunities to strengthen race data practices in healthcare and bring up persistent gaps in implementation, standardization, and health research.

Self-identification was the most widely supported and validated method for collecting race data in health systems, but real-world implementation challenges remain an issue [20,26,27,30,33,34,38,40,45]. It is grounded on the premise that individuals are best positioned to define and express their own identities [26]. Several studies highlighted strategies to operationalize self-identification [26,27,30,40], such as embedding self-reported surveys into EMRs [27], with reported response validity ranging from 84% to 100% in primary care settings [45]. Almklov et al. [27] demonstrated that electronic self-reporting tools outperformed both standard EHR documentation and staff-entered race data in terms of completeness and accuracy. Patient-facing tools, such as pre-visit portals or kiosk check-ins, were also proposed for clinical implementation, allowing individuals to directly record and update their demographic information, enhancing data quality and patient control [26,34]. Self-identification would also enhance the validity of race data by reducing common misclassification errors in observational or administrative databases, while also supporting more equitable care by centering patient voices and lived experiences [20,26,27,30,34,38,40]. This shift from extractive to patient-informed data practices may foster trust and align with principles of cultural safety.

Despite these strengths, policy, resource availability, and ethical challenges continue to limit the uptake of widespread implementation. In terms of policy, the absence of national or institutional mandates for standardized race data collection leads to inconsistent adoption [46]. Resource constraints, such as staff training or digital infrastructure, further complicate routine collection within already burdened clinical settings [46]. Ethically, concerns regarding data privacy and potential misuse create hesitancy within institutions and among patients. For instance, higher rates of “decline to answer” were observed when demographic questions lacked clarity or sensitivity, or when patients were unsure how their information would be used [27]. Future directions should explore how to frame race-related self-reporting questions to encourage disclosure, particularly among populations that are hesitant to self-identify. Also, although many studies emphasized the importance of self-identification and proposed techniques for improved data collection, few evaluated their implementation in real-world healthcare settings. Most recommendations remained conceptual, offering limited insight into feasibility, cost, equity impact, and sustainability of this intervention [30,33–35,40,45]. Further research should assess these suggestions based on patient engagement outcomes and determine necessary supports to translate these approaches into routine practice.

Another key finding from our review was the need for transparency in how race data were sourced, recorded and interpreted [20,27,39]. Swilley-Martinez et al. [39] emphasized that researchers and health systems should clearly document if race information was self-reported, observed, or algorithmically assigned, and record who made the classification. This level of documentation would strengthen the credibility of race data and ensure that downstream research and policy decisions were based on credible, well-understood sources [27]. Despite these advantages, few health systems consistently tracked or reported how race data were collected, and such information was often overlooked in research publications and datasets [46]. This introduced uncertainty, limited comparability, and made it difficult to determine whether observed disparities reflected real differences or were artifacts of data collection methods [46]. Therefore, future research should focus on developing clear standards for disclosing race data sources and examining how different data collection techniques may influence the interpretation of race-based analyses in health equity research. Establishing greater transparency will assist in laying the groundwork for standardized race data practices that are credible across healthcare systems.

Our findings highlighted the need for standardized frameworks for race data collection to improve comparability, accuracy, and equity across healthcare settings and databases [32,34,35,39]. Inconsistent definitions and classification schemes were noted as common barriers to accurately identifying and comparing race data, particularly when conducting analyses across institutions or national datasets [32]. Standardization was suggested to improve data interoperability, reduce data cleaning, and enable large-scale equity analyses within health systems. For instance, Huang et al. [32] evaluated Medicare administrative data and determined significant variability in the accuracy of commonly used coding systems depending on the racial group being identified. Thus, if racial groups were organized in the same manner, health databases could consistently produce demographic information that is comparable across sources. This could augment the ability to monitor disparities, allocate resources, and inform equitable policy responses. However, it is important to consider limitations that arise with race data standardization. Implementing harmonized systems requires extensive coordination between stakeholders such as government agencies, health system leaders, and EHR vendors, which may be resource intensive [37]. Future research should investigate strategies that account for institutional readiness, stakeholder alignment, and policy supports needed to adopt standardized data practices in differing health system contexts. Additionally, overly broad and outdated categories could perpetuate exclusion and obscure meaningful differences in health outcomes, such as masking subgroup variation or collapsing multiracial identities [26]. Further studies should explore frameworks for expanding racial categories to reflect evolving identities and capture within-group diversity.

Finally, our review identified the use of supplementary and predictive models as a strategy to improve the completeness of race data where self-reported data were incomplete or missing [26,28,41–44]. This included data linkage techniques, natural language processing, and algorithmic imputations such as the Bayesian Improved Surname Geocoding model. The Bayesian Improved Surname Geocoding model was the most mentioned method and estimates a patient’s race based on their last name and geocoded address [26,28,42]. This was shown to reduce missing data and enhance the reliability of race-based categorization in large datasets [28,42,44]. Some studies explored advanced or locally tailored imputation models that incorporated other factors like age, income, or household structure to improve accuracy [41,43]. To display uncertainty, probabilistic rather than fixed categorical assignments were recommended by multiple sources [28,41–44]. For instance, reporting a patient as 70% Black, 20% Hispanic, and 10% White to provide a more nuanced analysis with room for error [41,42]. Ultimately, this might improve the utility of health datasets that lack self-identified race information and allow an increased sample size of data. However, an important consideration to note was that predictive models relied on assumptions of factors (i.e., name, location, neighbourhood composition) that may not reflect an individual's lived racial identity [41]. This could unintentionally reinforce systematic biases, especially without any data validation or transparency. Ethical concerns such as lack of consent, limited interpretability, and potential misuse in policy or research contexts should also be considered. Future directions include exploring how advancements in artificial intelligence and natural language processing could improve contextual understanding of unstructured clinical notes, enhance data linkage, and automate validation processes to reduce biases and strengthen accuracy in race data identification. Future research should continue to focus on validating these imputation models against self-reported data, identifying methods for local tailoring, and creating ethical standards for probabilistic reporting to ensure these tools are used responsibly. As predictive approaches evolve, ensuring their ethical applications will be essential for advancing the quality of research and equity-informed health policies.

While most studies presented similar recommendations, a few presented conflicting perspectives regarding feasibility and data validity. Specifically, some authors questioned whether standardized categories risk oversimplifying complex identities [20,33,38], while others argued that uniform classification is necessary for comparability across databases [34,37,39]. For instance, certain studies recommended consolidating racial categories to align with United States national standards, such as the Office of Management and Budget standardized codes for comparability [26,34,35,39], while others emphasized the need to expand definitions or disaggregate data to better capture within-group differences and nuanced identities [29,33,37]. Furthermore, while some articles viewed algorithmic approaches as practical solutions for missing data [32,36,42–44], others cautioned that they would perpetuate biases or obscure inequities [26,40,41]. These contradictions demonstrated tension in the field regarding precise data and its comparability, as well as inconsistencies in innovation and ethics. Future research should aim to explore these perspectives by integrating both patient-centered and system level approaches, which manage data accuracy along with lived experience.

Overall, these findings reflected both advancements and ongoing challenges when collecting and reporting race data in health systems. Notably, of the 21 articles identified, most were based in the United States, limiting the global applicability of the recommendations due to differences in health systems, race categorizations, and data governance structures across diverse regions. Therefore, while current research provides a constructive foundation, the small number of applicable studies relative to the broad scope of the review highlights an important area for further research. Strengthening race data practices might enable a more accurate understanding of racial disparities in healthcare use and outcomes, which may guide more effective, equity-informed interventions and policy decisions.

Conclusion

This scoping review identified current recommendations for improving the identification of race within health databases, revealing five key thematic areas: self-identification and patient-centered practices, standardization across healthcare systems, data quality and completeness, algorithmic and predictive methods, and equity-oriented and disaggregated data use. While promising strategies exist, implementation remains inconsistent, and gaps such as a lack of global applicability, limited critical appraisal, and minimal focus on real-world feasibility highlight the need for further research. Strengthening race data identification is essential not only for improving data accuracy but also for supporting equity-driven research, policy, and health system transformation. Continued efforts to refine and operationalize these recommendations are crucial to advancing more inclusive healthcare systems.

Supporting information

Acknowledgments

The authors would like to acknowledge Dr. Laura van Staalduinen for her thoughtful support throughout this project. Thank you to Lauren Cadel and Amy Cho for their assistance with refining the search strategy and to Hannah Zuta for her contributions to screening and data extraction. The authors would also like to thank the Queen’s University librarians, Amanda Ross-White and Angelique Roy, for their expertise and contributions in developing the search strategy.