Health Equity Beyond Data: Health Care Worker Perceptions of Race, Ethnicity, and Language Data Collection in Electronic Health Records

Large language models leverage external knowledge to extend clinical insight beyond language boundaries

OBJECTIVES

Large Language Models (LLMs) such as ChatGPT and Med-PaLM have excelled in various medical question-answering tasks. However, these English-centric models encounter challenges in non-English clinical settings, primarily due to limited clinical knowledge in respective languages, a consequence of imbalanced training corpora. We systematically evaluate LLMs in the Chinese medical context and develop a novel in-context learning framework to enhance their performance.

MATERIALS AND METHODS

The latest China National Medical Licensing Examination (CNMLE-2022) served as the benchmark. We collected 53 medical books and 381 149 medical questions to construct the medical knowledge base and question bank. The proposed Knowledge and Few-shot Enhancement In-context Learning (KFE) framework leverages the in-context learning ability of LLMs to integrate diverse external clinical knowledge sources. We evaluated KFE with ChatGPT (GPT-3.5), GPT-4, Baichuan2-7B, Baichuan2-13B, and QWEN-72B in CNMLE-2022 and further investigated the effectiveness of different pathways for incorporating LLMs with medical knowledge from 7 distinct perspectives.

RESULTS

Directly applying ChatGPT failed to qualify for the CNMLE-2022 at a score of 51. Cooperated with the KFE framework, the LLMs with varying sizes yielded consistent and significant improvements. The ChatGPT's performance surged to 70.04 and GPT-4 achieved the highest score of 82.59. This surpasses the qualification threshold (60) and exceeds the average human score of 68.70, affirming the effectiveness and robustness of the framework. It also enabled a smaller Baichuan2-13B to pass the examination, showcasing the great potential in low-resource settings.

DISCUSSION AND CONCLUSION

This study shed light on the optimal practices to enhance the capabilities of LLMs in non-English medical scenarios. By synergizing medical knowledge through in-context learning, LLMs can extend clinical insight beyond language barriers in healthcare, significantly reducing language-related disparities of LLM applications and ensuring global benefit in this field.

Racing the Machine: Data Analytic Technologies and Institutional Inscription of Racialized Health Injustice

Recent scientific and policy initiatives frame clinical settings as sites for intervening upon inequality. Electronic health records and data analytic technologies offer opportunity to record standard data on education, employment, social support, and race-ethnicity, and numerous audiences expect biomedicine to redress social determinants based on newly available data. However, little is known on how health practitioners and institutional actors view data standardization in relation to inequity. This article examines a public safety-net health system's expansion of race, ethnicity, and language data collection, drawing on 10 months of ethnographic fieldwork and 32 qualitative interviews with providers, clinic staff, data scientists, and administrators. Findings suggest that electronic data capture institutes a decontextualized racialization within biomedicine as health practitioners and data workers rely on biological, cultural, and social justifications for collecting racial data. This demonstrates a critical paradox of stratified biomedicalization: The same data-centered interventions expected to redress injustice may ultimately reinscribe it.

Health and Racial Disparities: Importance of Accurate and Reliable Ethnicity, Race, and Language Data

Introduction: Accurate demographic data are essential to identify and monitor differences, trends, and changes in diabetes-related conditions between Hispanics and non-Hispanic Blacks (NHBs). It also provides pertinent information to reduce health and racial disparities among English- and Spanish-speakers. Method: The study's design was a quantitative cross-sectional one. Electronic medical record (EMR) and survey data of the same sample were compared. Descriptive statistics were computed for ethnicity, preferred language, and physiological data. Frequency and percentages were calculated for each continuous and categorical variable. Chi-square was calculated to compare physiological variables by ethnicity and language. Results: During a 5-month period (September 2021-February 2022), 106 individuals from New York City with diabetes took part in this study. Among Hispanics, most from the EMR identified as Other (82.4%), whereas from the survey, most identified as White (57.1%). More Hispanics (19%) and Spanish speakers (18%) had high triglyceride levels compared to NHBs (2%) and English speakers (3%). Conclusion: Ensuring that demographic data are accurate can better inform programs. Because Hispanics and Spanish speakers had the highest triglyceride levels, diabetes programs need to include information on cardiovascular disease and must be available in Spanish, to further reduce risk factors, improve health outcomes, and promote health equity among these populations.

Identifying and improving the "ground truth" of race in disparities research through improved EMR data reporting. A systematic review

BACKGROUND

Studies about racial disparities in healthcare are increasing in quantity; however, they are subject to vast differences in definition, classification, and utilization of race/ethnicity data. Improved standardization of this information can strengthen conclusions drawn from studies using such data. The objective of this study is to examine how data related to race/ethnicity are recorded in research through examining articles on race/ethnicity health disparities and examine problems and solutions in data reporting that may impact overall data quality.

METHODS

In this systematic review, Business Source Complete, Embase.com, IEEE Xplore, PubMed, Scopus and Web of Science Core Collection were searched for relevant articles published from 2000 to 2020. Search terms related to the concepts of electronic medical records, race/ethnicity, and data entry related to race/ethnicity were used. Exclusion criteria included articles not in the English language and those describing pediatric populations. Data were extracted from published articles. This review was organized and reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 statement for systematic reviews.

FINDINGS

In this systematic review, 109 full text articles were reviewed. Weaknesses and possible solutions have been discussed in current literature, with the predominant problem and solution as follows: the electronic medical record (EMR) is vulnerable to inaccuracies and incompleteness in the methods that research staff collect this data; however, improved standardization of the collection and use of race data in patient care may help alleviate these inaccuracies.

INTERPRETATION

Conclusions drawn from large datasets concerning peoples of certain race/ethnic groups should be made cautiously, and a careful review of the methodology of each publication should be considered prior to implementation in patient care.

Assessing the accuracy of electronic health record gender identity and REaL data at an academic medical center

BACKGROUND

Collection of accurate patient race, ethnicity, preferred language (REaL) and gender identity in the electronic health record (EHR) is essential for equitable and inclusive care. Misidentification of these factors limits quality measurement of health outcomes in at-risk populations. Therefore, the aim of our study was to assess the accuracy of REaL and gender identity data at our institution.

METHODS

A survey was administered to 117 random patients, selected from prior day admissions at a large academic medical center in urban central New York. Patients (or guardians) self-reported REaL and gender identity data, selecting from current EHR options. Variables were coded for the presence or absence of a difference from data recorded in the EHR.

RESULTS

Race was misreported in the EHR for 13% of patients and ethnicity for 6%. For most White and Black patients, race was concordant. However, self-identified data for all multiracial patients were discordant with the EHR. Most Non-Hispanic patients had ethnicity correctly documented. Some Hispanic patients were misidentified. There was a significant association between reporting both a race and an ethnicity which differed from the EHR on chi square analysis (P < 0.001). Of those who reported an alternative ethnicity, 71.4% also reported an alternative race. Gender identity was missing for most patients and 11% of the gender-identity entries present in the EHR were discordant with the patient's self-identity. Preferred language was 100% concordant with the EHR.

CONCLUSIONS

At an academic medical center, multiracial and Hispanic patients were more likely to have their demographics misreported in the EHR, and gender identity data were largely missing. Healthcare systems need strategies that support accurate collection of patients' self-reported ReAL and gender identity data to improve the future ability to identify and address healthcare disparities.

Social Risk Screening and Response Equity: Assessment by Race, Ethnicity, and Language in Community Health Centers

INTRODUCTION

Little has previously been reported about the implementation of social risk screening across racial/ethnic/language groups. To address this knowledge gap, the associations between race/ethnicity/language, social risk screening, and patient-reported social risks were examined among adult patients at community health centers.

METHODS

Patient- and encounter-level data from 2016 to 2020 from 651 community health centers in 21 U.S. states were used; data were extracted from a shared Epic electronic health record and analyzed between December 2020 and February 2022. In adjusted logistic regression analyses stratified by language, robust sandwich variance SE estimators were applied with clustering on patient's primary care facility.

RESULTS

Social risk screening occurred at 30% of health centers; 11% of eligible adult patients were screened. Screening and reported needs varied significantly by race/ethnicity/language. Black Hispanic and Black non-Hispanic patients were approximately twice as likely to be screened, and Hispanic White patients were 28% less likely to be screened than non-Hispanic White patients. Hispanic Black patients were 87% less likely to report social risks than non-Hispanic White patients. Among patients who preferred a language other than English or Spanish, Black Hispanic patients were 90% less likely to report social needs than non-Hispanic White patients.

CONCLUSIONS

Social risk screening documentation and patient reports of social risks differed by race/ethnicity/language in community health centers. Although social care initiatives are intended to promote health equity, inequitable screening practices could inadvertently undermine this goal. Future implementation research should explore strategies for equitable screening and related interventions.

Alliance for Innovation on Maternal Health: Consensus Bundle on Cardiac Conditions in Obstetric Care

Cardiac conditions are the leading cause of pregnancy-related deaths and disproportionately affect non-Hispanic Black people. Multidisciplinary maternal mortality review committees have found that most people who died from cardiac conditions during pregnancy or postpartum were not diagnosed with a cardiovascular disease before death and that more than 80% of all pregnancy-related deaths, regardless of cause, were preventable. In addition, other obstetric complications, such as preeclampsia and gestational diabetes, are associated with future cardiovascular disease risk. Those with cardiac risk factors and those with congenital and acquired heart disease require specialized care during pregnancy and postpartum to minimize risk of preventable morbidity and mortality. This bundle provides guidance for health care teams to develop coordinated, multidisciplinary care for pregnant and postpartum people with cardiac conditions and to respond to cardio-obstetric emergencies. This bundle is one of several core patient safety bundles developed by the Alliance for Innovation on Maternal Health that provide condition- or event-specific clinical practices for implementation in appropriate care settings. The Cardiac Conditions in Obstetric Care bundle is organized into five domains: 1) Readiness , 2) Recognition and Prevention , 3) Response , 4) Reporting and Systems Learning , and 5) Respectful Care . This bundle is the first by the Alliance to be developed with the fifth domain of Respectful Care . The Respectful Care domain provides essential best practices to support respectful, equitable, and supportive care to all patients. Further health equity considerations are integrated into elements in each domain.

Geospatial analysis of patients' social determinants of health for health systems science and disparity research

Social context matters for health, healthcare processes/quality and patient outcomes. The social status and circumstances we are born into, grow up in and live under, are called social determinants of health; they drive our health, and how we access and experience care; they are the fundamental causes of disease outcomes. Such circumstances are influenced heavily by our location through neighborhood context, which relates to support networks. Geography can influence proximity to resources and is an important dimension of social determinants of health, which also encompass race/ethnicity, language, health literacy, gender identity, social capital, wealth and income. Beginning with an explanation of social determinants, we explore the use of Geospatial Analysis methods and geocoding, including the importance of collaborating with geography experts, the pitfalls of geocoding, and how geographic analysis can help us to understand patient populations within the context of Social Determinants of Health. We then explain mechanisms and methods of geospatial analysis with two examples: (1) Bayesian hierarchical regression with crossed random effects and (2) discontinuity regression i.e., change point analysis. We leveraged the local University of Utah and Yale cohorts of the Multicenter Perioperative Outcomes Group (MPOG.org ), a perioperative electronic health registry; we enriched the Utah cohort with US-census tract level social determinants of health after geocoding patient addresses and extracting social determinants of health from the National Neighborhood Database (NaNDA). We explain how to investigate the impact of US-census tract level community deprivation indices and racial/ethnic composition on (1) individual clinicians’ administration of risk-adjusted perioperative antiemetic prophylaxis, (2) patients’ decisions to defer cataract surgery at the cusp of Medicare eligibility and finally (3) methods to further characterize patient populations at risk through publicly available datasets in the context of public transit access. Our examples are not rigorous analyses, and our preliminary inferences should not be taken at face value, but rather seen as illustration of geospatial analysis processes and methods. Our worked examples show the potential utility of geospatial analysis, and in particular the power of geocoding patient addresses to extract US-census level social determinants of health from publicly available databases to enrich electronic health registries for healthcare disparity research and targeted health system level countermeasures.

Virtuosic craft or clerical labour: the rise of the electronic health record and challenges to physicians' professional identity (1950-2022)

The electronic health record (EHR) is a focus of contentious debate, having become as essential to contemporary clinical practice as it is polarising. Debates about the EHR raise questions about physicians' professional identity, the nature of clinical work, evolution of the patient/practitioner relationship, and narratives of technological optimism and pessimism. The metaphors by which clinicians stake our identities-are we historians, detectives, educators, technicians, or something else?-animate the history of the early computer-based medical record in the mid-to-late twentieth-century USA. Proponents and detractors were equally interested in what the EHR revealed about clinician identity, and how it might fundamentally reshape it. This paper follows key moments in the history of the early computer-based patient record from the late 1950s to the EHR of the present day. In linking physician identity development, clinical epistemological structures, and the rise of the computer-based medical record in the USA in the mid-to-late twentieth century, we ask why the EHR is such a polarising entity in contemporary medicine, and situate clinician/EHR tensions in a longer history of aspirational physician identity and a kind of technological optimism that soon gave way to pessimism surrounding computer-based clinical work.

Missing Race and Ethnicity Data among COVID-19 Cases in Massachusetts

Infectious disease surveillance frequently lacks complete information on race and ethnicity, making it difficult to identify health inequities. Greater awareness of this issue has occurred due to the COVID-19 pandemic, during which inequities in cases, hospitalizations, and deaths were reported but with evidence of substantial missing demographic details. Although the problem of missing race and ethnicity data in COVID-19 cases has been well documented, neither its spatiotemporal variation nor its particular drivers have been characterized. Using individual-level data on confirmed COVID-19 cases in Massachusetts from March 2020 to February 2021, we show how missing race and ethnicity data: (1) varied over time, appearing to increase sharply during two different periods of rapid case growth; (2) differed substantially between towns, indicating a nonrandom distribution; and (3) was associated significantly with several individual- and town-level characteristics in a mixed-effects regression model, suggesting a combination of personal and infrastructural drivers of missing data that persisted despite state and federal data-collection mandates. We discuss how a variety of factors may contribute to persistent missing data but could potentially be mitigated in future contexts.

Health insurance, healthcare utilization and language use among populations who experience risk for tuberculosis, California 2014–2017

Background

California tuberculosis (TB) prevention goals include testing more than ten million at-risk Californians and treating two million infected with tuberculosis. Adequate health insurance and robust healthcare utilization are crucial to meeting these goals, but information on these factors for populations that experience risk for TB is limited.

Methods

We used data from the 2014–2017 California Health Interview Survey (n = 82,758), a population-based dual-frame telephone survey to calculate survey proportions and 95% confidence intervals (CI) stratified by country of birth, focusing on persons from countries of birth with the highest number of TB cases in California. Survey proportions for recent doctor’s visit, overall health, smoking, and diabetes were age-adjusted.

Results

Among 18–64 year-olds, 27% (CI: 25–30) of persons born in Mexico reported being uninsured in contrast with 3% (CI: 1–5) of persons born in India. Report of recent doctor’s visit was highest among persons born in the Philippines, 84% (CI: 80–89) and lowest among Chinese-born persons, 70% (CI: 63–76). Persons born in Mexico were more likely to report community clinics as their usual source of care than persons born in China, Vietnam, or the Philippines. Poverty was highest among Mexican-born persons, 56% (CI: 54–58) and lowest among Indian-born persons, 9% (CI: 5–13). Of adults with a medical visit in a non-English language, 96% (CI: 96–97) were non-U.S.-born, but only 42% (CI: 40–44) of non-U.S.-born persons had a visit in a non-English language.

Discussion

Many, though not all, of the populations that experience risk for TB had health insurance and used healthcare. We found key differences in usual source of care and language use by country of birth which should be considered when planning outreach to specific providers, clinic systems, insurers and communities for TB prevention and case-finding.