The accuracy of race & ethnicity data in US based healthcare databases: A systematic review

Validation of Multi-State EHR-Based Network for Disease Surveillance (MENDS) Data and Implications for Improving Data Quality and Representativeness

Introduction

Surveillance modernization efforts emphasize the potential use of electronic health record (EHR) data to inform public health surveillance and prevention. However, EHR data streams vary widely in their completeness, accuracy, and representativeness.

Methods

We developed a validation process for the Multi-State EHR-Based Network for Disease Surveillance (MENDS) pilot project to identify and resolve data quality issues that could affect chronic disease prevalence estimates. We examined MENDS validation processes from December 2020 through August 2023 across 5 data-contributing organizations and outlined steps to resolve data quality issues.

Results

We identified gaps in the EHR databases of data contributors and in the processes to extract, map, integrate, and analyze their EHR data. Examples of source-data problems included missing data on race and ethnicity and zip codes. Examples of data processing problems included duplicate or missing patient records, lower-than-expected volumes of data, use of multiple fields for a single data type, and implausible values.

Conclusion

Validation protocols identified critical errors in both EHR source data and in the processes used to transform these data for analysis. Our experience highlights the value and importance of data validation to improve data quality and the accuracy of surveillance estimates that use EHR data. The validation process and lessons learned can be applied broadly to other EHR-based surveillance efforts.

Relationship between Civilian Injuries Caused during Contact with Law Enforcement and Community-Level Sociodemographic Characteristics

Civilian injuries caused during contact with law enforcement personnel erode community trust in policing, impact individual well-being, and exacerbate existing health inequities. We assessed the relationship between ZIP code-level rates of civilian injuries caused during legal interventions and community-level sociodemographic characteristics using Illinois hospital data from 2016 to 2022. We developed multivariable Poisson regression models to examine whether legal intervention injury rates differed by race-ethnicity and community economic disadvantage across three geographic regions of Illinois representing different levels of urbanization. Over the study period, 4976 civilian injuries were treated in Illinois hospitals (rate of 5.6 per 100,000 residents). Compared to non-Hispanic white residents, non-Hispanic Black residents demonstrated 5.5-10.5 times higher injury rates across the three geographic regions, and Hispanic-Latino residents demonstrated higher rates in Chicago and suburban Cook County, but lower rates in the rest of the state. In most regions, models showed that as the percent of minority residents in a ZIP code increased, injury rates among non-Hispanic Black and Hispanic-Latino residents decreased. As community economic disadvantage increased at the ZIP code level, civilian injury rates increased. Communities with the highest injury rates involving non-Hispanic white residents were significantly more economically unequal and disadvantaged. While the injury rates were consistently and substantially higher among non-Hispanic Black residents throughout the state, the findings illustrate that the association between overall civilian injuries caused during contact with law enforcement and community sociodemographic characteristics varied across regions. Data on local law enforcement agency policies and procedures are needed to better identify appropriate interventions.

Addressing bias in preterm birth research: The role of advanced imputation techniques for missing race and ethnicity in perinatal health data

OBJECTIVES

To evaluate the effectiveness of Bayesian Improved Surname Geocoding (BISG) and Bayesian Improved First Name Surname Geocoding (BIFSG) in estimating race and ethnicity, and how they influence odds ratios for preterm birth.

METHODS

We analyzed hospital birth admission electronic health records (EHR) data (N = 9985). We created two simulation sets with 40 % of race and ethnicity data missing randomly or more likely for non-Hispanic black birthing people who had preterm birth. We calculated C-statistics to evaluate how accurately BISG and BIFSG estimate race and ethnicity. We examined the association between race and ethnicity and preterm birth using logistic regression and reported odds ratios (OR).

RESULTS

BISG and BIFSG showed high accuracy for most racial and ethnic categories (C-statistics = 0.94-0.97, 95 % confidence intervals [CI] = 0.92-0.97). When race and ethnicity were not missing at random, BISG (OR = 1.25, CI = 0.97-1.62) and BIFSG (OR = 1.38, CI = 1.08-1.76) resulted in positive estimates mirroring the true association (OR = 1.68, CI = 1.34-2.09) for Non-Hispanic Black birthing people, while traditional methods showed contrasting estimates (Complete case OR = 0.62, CI = 0.41-0.94; multiple imputation OR = 0.63, CI = 0.40-0.98).

CONCLUSIONS

BISG and BIFSG accurately estimate missing race and ethnicity in perinatal EHR data, decreasing bias in preterm birth research, and are recommended over traditional methods to reduce potential bias.

Development and Validation of an 18-Gene Urine Test for High-Grade Prostate Cancer

Importance

Benefits of prostate cancer (PCa) screening with prostate-specific antigen (PSA) alone are largely offset by excess negative biopsies and overdetection of indolent cancers resulting from the poor specificity of PSA for high-grade PCa (ie, grade group [GG] 2 or greater).

Objective

To develop a multiplex urinary panel for high-grade PCa and validate its external performance relative to current guideline-endorsed biomarkers.

Design, Setting, and Participants

RNA sequencing analysis of 58 724 genes identified 54 markers of PCa, including 17 markers uniquely overexpressed by high-grade cancers. Gene expression and clinical factors were modeled in a new urinary test for high-grade PCa (MyProstateScore 2.0 [MPS2]). Optimal models were developed in parallel without prostate volume (MPS2) and with prostate volume (MPS2+). The locked models underwent blinded external validation in a prospective National Cancer Institute trial cohort. Data were collected from January 2008 to December 2020, and data were analyzed from November 2022 to November 2023.

Exposure

Protocolized blood and urine collection and transrectal ultrasound-guided systematic prostate biopsy.

Main Outcomes and Measures

Multiple biomarker tests were assessed in the validation cohort, including serum PSA alone, the Prostate Cancer Prevention Trial risk calculator, and the Prostate Health Index (PHI) as well as derived multiplex 2-gene and 3-gene models, the original 2-gene MPS test, and the 18-gene MPS2 models. Under a testing approach with 95% sensitivity for PCa of GG 2 or greater, measures of diagnostic accuracy and clinical consequences of testing were calculated. Cancers of GG 3 or greater were assessed secondarily.

Results

Of 761 men included in the development cohort, the median (IQR) age was 63 (58-68) years, and the median (IQR) PSA level was 5.6 (4.6-7.2) ng/mL; of 743 men included in the validation cohort, the median (IQR) age was 62 (57-68) years, and the median (IQR) PSA level was 5.6 (4.1-8.0) ng/mL. In the validation cohort, 151 (20.3%) had high-grade PCa on biopsy. Area under the receiver operating characteristic curve values were 0.60 using PSA alone, 0.66 using the risk calculator, 0.77 using PHI, 0.76 using the derived multiplex 2-gene model, 0.72 using the derived multiplex 3-gene model, and 0.74 using the original MPS model compared with 0.81 using the MPS2 model and 0.82 using the MPS2+ model. At 95% sensitivity, the MPS2 model would have reduced unnecessary biopsies performed in the initial biopsy population (range for other tests, 15% to 30%; range for MPS2, 35% to 42%) and repeat biopsy population (range for other tests, 9% to 21%; range for MPS2, 46% to 51%). Across pertinent subgroups, the MPS2 models had negative predictive values of 95% to 99% for cancers of GG 2 or greater and of 99% for cancers of GG 3 or greater.

Conclusions and Relevance

In this study, a new 18-gene PCa test had higher diagnostic accuracy for high-grade PCa relative to existing biomarker tests. Clinically, use of this test would have meaningfully reduced unnecessary biopsies performed while maintaining highly sensitive detection of high-grade cancers. These data support use of this new PCa biomarker test in patients with elevated PSA levels to reduce the potential harms of PCa screening while preserving its long-term benefits.

Time Differences From Abnormal Cervical Cancer Screening to Colposcopy Between Insurance Statuses

OBJECTIVE

Screening and diagnostic follow-up to prevent cervical cancer are influenced by socioeconomic and systemic factors. This study sought to characterize intervals from abnormal cervical cancer screening to colposcopy between practices differing by insurance status at a large, urban academic center.

MATERIALS AND METHODS

This retrospective cohort study included patients aged 21-65 who presented for colposcopy between January 1, 2021, and January 1, 2022, at the resident and faculty gynecology practices of a single large urban academic medical center. Patient characteristics were compared using t tests or Wilcoxon rank sum tests for continuous measures and χ2 or Fisher exact tests for categorical measures. Intervals from abnormal cervical cancer screening to colposcopy were compared using the Wilcoxon rank sum test and linear regression analysis with multivariable models adjusted for age, cervical cytology result, human papillomavirus result, and HIV status.

RESULTS

Resident practice patients were publicly insured and more likely to be Black or Hispanic (p < .0001); rates of high-risk human papillomavirus and smoking were similar. Resident practice patients had longer intervals from abnormal cervical cancer screening to colposcopy compared with faculty practice patients (median 79.5 vs 34 d, p < .0001). On adjusted analysis, resident practice patients faced a 95% longer interval (p < .0001).

CONCLUSIONS

Publicly insured patients of a resident-based practice faced significantly longer intervals from abnormal cervical cancer screening to colposcopy than faculty practice patients at a single urban academic center. Effort to address these differences may be an area of focus in improving health disparities.

A Novel Approach for Conducting a Catchment Area Analysis of Breast Cancer by Age and Stage for a Community Cancer Center

BACKGROUND

The U.S. Preventive Services Task Force recently issued an updated draft recommendation statement to initiate breast cancer screening at age 40, reflecting well-documented disparities in breast cancer-related mortality that disproportionately impact younger Black women. This study applied a novel approach to identify hotspots of breast cancer diagnosed before age 50 and/or at an advanced stage to improve breast cancer detection within these communities.

METHODS

Cancer registry data for 3,497 women with invasive breast cancer diagnosed or treated between 2012 and 2020 at the Helen F. Graham Cancer Center and Research Institute (HFGCCRI) and who resided in the HFGCCRI catchment area, defined as New Castle County, Delaware, were geocoded and analyzed with spatial intensity. Standardized incidence ratios stratified by age and race were calculated for each hotspot.

RESULTS

Four hotspots were identified, two for breast cancer diagnosed before age 50, one for advanced breast cancer, and one for advanced breast cancer diagnosed before age 50. Younger Black women were overrepresented in these hotspots relative to the full-catchment area.

CONCLUSIONS

The novel use of spatial methods to analyze a community cancer center catchment area identified geographic areas with higher rates of breast cancer with poor prognostic factors and evidence that these areas made an outsized contribution to racial disparities in breast cancer.

IMPACT

Identifying and prioritizing hotspot breast cancer communities for community outreach and engagement activities designed to improve breast cancer detection have the potential to reduce the overall burden of breast cancer and narrow racial disparities in breast cancer.

Racial and Ethnic Disparities in Perioperative Health Care Among Patients Undergoing Cardiac Surgery: JACC State-of-the-Art Review

There has been little progress in reducing health care disparities since the 2003 landmark Institute of Medicine's report Unequal Treatment. Despite the higher burden of cardiovascular disease in underrepresented racial and ethnic groups, they have less access to cardiologists and cardiothoracic surgeons, and have higher rates of morbidity and mortality with cardiac surgical interventions. This review summarizes existing literature and highlights disparities in cardiovascular perioperative health care. We propose actionable solutions utilizing multidisciplinary perspectives from cardiology, cardiac surgery, cardiothoracic anesthesiology, critical care, medical ethics, and health disparity experts. Applying a health equity lens to multipronged interventions is necessary to eliminate the disparities in perioperative health care among patients undergoing cardiac surgery.

Racial and Ethnic Disparities in Delivery In-Hospital Mortality or Maternal End-Organ Injury: A Multistate Analysis, 2007-2020

Background: In the United States, Black maternal mortality is 2-4 × higher than that of White maternal mortality, with differences also present in severe maternal morbidity and other measures. However, limited research has comprehensively studied multilevel social determinants of health, and their confounding and effect modification on obstetrical outcomes. Materials and Methods: We performed a retrospective multistate analysis of adult inpatient delivery hospitalizations (Florida, Kentucky, Maryland, New Jersey, New York, North Carolina, and Washington) between 2007 and 2020. Multilevel multivariable models were used to test the confounder-adjusted association for race/ethnicity and the binary outcomes (1) in-hospital mortality or maternal end-organ injury and (2) in-hospital mortality only. Stratified analyses were performed to test effect modification. Results: The confounder-adjusted odds ratio showed that Black (1.33, 95% confidence interval [CI]: 1.30-1.36) and Hispanic (1.14, 95% CI: 1.11-1.18) as compared with White patients were more likely to die in-hospital or experience maternal end-organ injury. For Black and Hispanic patients, stratified analysis showed that findings remained significant in almost all homogeneous strata. After statistical adjustment, Black as compared with White patients were more likely to die in-hospital (1.49, 95% CI: 1.21-1.82). Conclusions: Black and Hispanic patients had higher adjusted odds of in-patient mortality and end-organ damage after birth than White patients. Race and ethnicity serve as strong predictors of health care inequality, and differences in outcomes may reflect broader structural racism and individual implicit bias. Proposed solutions require immense and multifaceted active efforts to restructure how obstetrical care is provided on the societal, hospital, and patient level.

Limited English Proficiency Is Associated With Diabetic Retinopathy in Patients Presenting for Cataract Surgery

Purpose

To investigate the relationship between limited English proficiency (LEP) and diabetic retinopathy (DR) in patients presenting for cataract surgery.

Methods

This is a retrospective observational study of patients who underwent cataract surgery between January 2014 and February 2020. Patients who self-identified as needing or preferring an interpreter were defined as having LEP. Differences in demographics, characteristics, and outcomes including history of type 2 diabetes (T2DM), DR, preoperative best corrected visual acuity (BCVA), macular edema, and anti-vascular endothelial growth factor injections were analyzed. Statistical comparisons were assessed using logistic regression with generalized estimating equations.

Results

We included 13,590 eyes. Of these, 868 (6.4%) were from LEP patients. Patients with LEP were more likely to be Hispanic (P < 0.001), female sex (P = 0.008), or older age (P = 0.003) and have worse mean BCVA at presentation (P < 0.001). Patients with LEP had a significantly higher rate of T2DM (P < 0.001), macular edema (P = 0.033), and DR (18.1% vs. 5.8%, P < 0.001). Findings remained significant when controlling for age, sex, race/ethnicity, and type of health insurance. Patients with LEP and DR were more likely to have had later stages of DR (P = 0.023).

Conclusions

Patients with LEP presenting for cataract surgery had a higher rate of DR and associated complications compared to patients with English proficiency. Further studies are needed to understand how language disparities influence health and what measures could be taken to improve healthcare in this vulnerable population.

Translational Relevance

Our study highlights healthcare disparities within ophthalmology and emphasizes the importance of advocating for improved healthcare delivery for patients with LEP.

Retaining Race in Chronic Kidney Disease Diagnosis and Treatment

The best overall measure of kidney function is glomerular filtration rate (GFR) as commonly estimated from serum creatinine concentrations (eGFRcr) using formulas that correct for the higher average creatinine concentrations in Blacks. After two decades of use, these formulas have come under scrutiny for estimating GFR differently in Blacks and non-Blacks. Discussions of whether to include race (Black vs. non-Black) in the calculation of eGFRcr fail to acknowledge that the original race-based eGFRcr provided the same CKD treatment recommendations for Blacks and non-Blacks based on directly (exogenously) measured GFR. Nevertheless, the National Kidney Foundation and the American Society of Nephrology Task Force on Reassessing the Inclusion of Race in Diagnosing Kidney Disease removed race in CKD treatment guidelines and pushed for the immediate adoption of a race-free eGFRcr formula by physicians and clinical laboratories. This formula is projected to negate CKD in 5.51 million White and other non-Black adults and reclassify CKD to less severe stages in another 4.59 million non-Blacks, in order to expand treatment eligibility to 434,000 Blacks not previously diagnosed and to 584,000 Blacks previously diagnosed with less severe CKD. This review examines: 1) the validity of the arguments for removing the original race correction, and 2) the performance of the proposed replacement formula. Excluding race in the derivation of eGFRcr changed the statistical bias from +3.7 to -3.6 ml/min/1.73m2 in Blacks and from +0.5 to +3.9 in non-Blacks, i.e., promoting CKD diagnosis in Blacks at the cost of restricting diagnosis in non-Blacks. By doing so, the revised eGFRcr greatly exaggerates the purported racial disparity in CKD burden. Claims that the revised formulas identify heretofore undiagnosed CKD in Blacks are not supported when studies that used kidney failure replacement therapy and mortality are interpreted as proxies for baseline CKD. Alternatively, a race-stratified eGFRcr (i.e., separate equations for Blacks and non-Blacks) would provide the least biased eGFRcr for both Blacks and non-Blacks and the best medical treatment for all patients.

A Customized Human Mitochondrial DNA Database (hMITO DB v1.0) for Rapid Sequence Analysis, Haplotyping and Geo-Mapping

The field of mitochondrial genomics has advanced rapidly and has revolutionized disciplines such as molecular anthropology, population genetics, and medical genetics/oncogenetics. However, mtDNA next-generation sequencing (NGS) analysis for matrilineal haplotyping and phylogeographic inference remains hindered by the lack of a consolidated mitogenome database and an efficient bioinformatics pipeline. To address this, we developed a customized human mitogenome database (hMITO DB) embedded in a CLC Genomics workflow for read mapping, variant analysis, haplotyping, and geo-mapping. The database was constructed from 4286 mitogenomes. The macro-haplogroup (A to Z) distribution and representative phylogenetic tree were found to be consistent with published literature. The hMITO DB automated workflow was tested using mtDNA-NGS sequences derived from Pap smears and cervical cancer cell lines. The auto-generated read mapping, variants track, and table of haplotypes and geo-origins were completed in 15 min for 47 samples. The mtDNA workflow proved to be a rapid, efficient, and accurate means of sequence analysis for translational mitogenomics.