Abstract:More than 20 years ago, a report on unequal treatment by the Institute of Medicine 1 made several recommendations to address racial and ethnic health disparities; chief among them was ensuring accurate and complete race and ethnicity data to identify disparities and inform interventions. Since then, electronic health records (EHRs) have become one of the main sources of population-level race and ethnicity data used for research, quality improvement efforts, and resource allocation for equitable interventions. However, despite its importance, numerous studies have demonstrated that EHR race and ethnicity data are incomplete and inaccurate, especially for minoritized populations at the highest risk for health disparities. 2 Salhi et al 3 add to this growing body of literature with their retrospective cohort study. Using a statewide clinical registry, they measured the frequency of discordant EHR documentation of race and ethnicity for patients with at least 2 emergency department visits to the same hospital or health system between December 1, 2018, and November 30, 2021. Discordance was defined as a change in race and ethnicity categorization across visits. In their large sample of 403 587 adult and 121 839 pediatric patients, they found that the rate of discordance was 1.7% for adults and 7.0% for children. 3 The process of race and ethnicity data collection was unknown, and authors suggested that the discordance observed could be due to variability in data collection. The authors state that this level of discordance may widen disparities or limit the ability to detect them. This may have downstream effects and highlights the need for best practices in race and ethnicity data collection. The study by Salhi et al 3 is one of only a few studies that has looked specifically at EHR discordance across emergency department visits and stratified for both adult and pediatric data. Most studies on EHR discordance have been conducted among adults and compared EHR data with self-reported surveys; pediatric studies have typically had smaller sample sizes or have been from single clinical sites. Sahil et al 3 found higher rates of discordance for children vs adults, which is not surprising considering there is a lack of guidelines and consistent practices in demographic data collection among pediatric populations. 4 In pediatrics there are unique challenges to address: Should the data be provided by a caregiver or child? What age is appropriate for individuals to self-report? How do we reconcile discordance among caregivers or between caregiver and the child? It may be helpful to have a designated area in the EHR to document who provided the data (eg, self, caregiver, or staff observation). Also, because multiracial children are one of the largest growing racial groups in the US, rendering clear data collection processes that have expansive race and ethnicity categories is very important. The complexity of collecting race and ethnicity data can be used to understand how discordance may occur. Overall, these complexities can be distilled into the 5 Ws (who, what, where, when, and why): who collects and provides the data, what race and ethnicity categories are available to choose from, when are data collected, where are data collected, and making clear to patients why data are collected and what will be done with them. Of note, the Salhi et al 3 study was conducted on data before recent calls to more clearly describe the processes used for race and ethnicity data collection from funding agencies, health systems, and studies published in medical journals. For example, the National Committee on Quality Assurance highlighted the issue of discordance from multiple data sources and developed guiding principles on how to prioritize sources of truth and the need for data management systems to allow documentation of how data are collected to better inform reconciliation processes. 5 In addition, the Centers for Medicare & Medicaid Services, the biggest payer of health care in the US, has made expanding the collection, reporting, and analysis of standardized data, including race and ethnicity, 1 of their 5 priority areas in their Framework for Health Equity 2022-2032. 6 Because minoritized groups are more likely than majority groups to have incomplete and inaccurate data, Centers for Medicare & Medicaid Services and others use various imputation methods to indirectly approximate data on the population level, but there is a tension between using those analytic methods over focusing on the more time-intensive and person-centered approach of improvement efforts to consistently implement the criterion standard of self-reported race and ethnicity in health care spaces. 5 ,6 Our society's understanding of race and ethnicity as fluid, socially constructed identit -Abstract Truncated-

Biases in Race and Ethnicity Introduced by Filtering Electronic Health Records for 'Complete Data'

Implicit bias in Critical Care Data: Factors affecting sampling frequencies and missingness patterns of clinical and biological variables in ICU Patients

Racial Bias in Clinical and Population Health Algorithms: A Critical Review of Current Debates

Disparities in Documentation: Evidence of Race-Based Biases in the Electronic Medical Record

Potential Biases in Machine Learning Algorithms Using Electronic Health Record Data

The Sociodemographic Biases in Machine Learning Algorithms: A Biomedical Informatics Perspective

Availability of information needed to evaluate algorithmic fairness - A systematic review of publicly accessible critical care databases

Conceptualizing bias in EHR data: A case study in performance disparities by demographic subgroups for a pediatric obesity incidence classifier

Race and Ethnicity Data in Electronic Health Records

The Lifecycle of Electronic Health Record Data in HIV-Related Big Data Studies: A Qualitative Study of Instances of and Potential Opportunities to Minimize Bias

Examining Linguistic Differences in Electronic Health Records for Diverse Patients With Diabetes: Natural Language Processing Analysis

Racial and Ethnic Bias in Risk Prediction Models for Colorectal Cancer Recurrence When Race and Ethnicity Are Omitted as Predictors

On Biases in a UK Biobank-based Retinal Image Classification Model

A framework for understanding selection bias in real-world healthcare data

Assessing Electronic Health Records for Describing Racial and Ethnic Health Disparities: A Research Note

Dissecting racial bias in an algorithm used to manage the health of populations

Addressing Information Biases Within Electronic Health Record Data to Improve the Examination of Epidemiologic Associations With Diabetes Prevalence Among Young Adults: Cross-Sectional Study

The bias algorithm: how AI in healthcare exacerbates ethnic and racial disparities – a scoping review

1215-P: Evaluating Alternate Bias Correction Methods in Estimating Diabetes Prevalence with Electronic Health Records

Accuracy of Race, Ethnicity, and Language Preference in an Electronic Health Record