Detecting racial/ethnic health disparities using deep learning from frontal chest radiography
Open accessAbstract
Purpose The aim of this study was to assess racial/ethnic and socioeconomic disparities in the difference between atherosclerotic vascular disease prevalence measured by a multitask convolutional neural network (CNN) deep learning model using frontal chest radiographs (CXRs) and the prevalence reflected by administrative hierarchical condition category codes in two cohorts of patients with coronavirus disease 2019 (COVID-19).Methods A CNN model, previously published, was trained to predict atherosclerotic disease from ambulatory frontal CXRs. The model was then validated on two cohorts of patients with COVID-19: 814 ambulatory patients from a suburban location (presenting from March 14, 2020, to October 24, 2020, the internal ambulatory cohort) and 485 hospitalized patients from an inner-city location (hospitalized from March 14, 2020, to August 12, 2020, the external hospitalized cohort). The CNN model predictions were validated against electronic health record administrative codes in both cohorts and assessed using the area under the receiver operating characteristic curve (AUC). The CXRs from the ambulatory cohort were also reviewed by two board-certified radiologists and compared with the CNN-predicted values for the same cohort to produce a receiver operating characteristic curve and the AUC. The atherosclerosis diagnosis discrepancy, Δ vasc, referring to the difference between the predicted value and presence or absence of the vascular disease HCC categorical code, was calculated. Linear regression was performed to determine the association of Δ vasc with the covariates of age, sex, race/ethnicity, language preference, and social deprivation index. Logistic regression was used to look for an association between the presence of any hierarchical condition category codes with Δ vasc and other covariates.Results The CNN prediction for vascular disease from frontal CXRs in the ambulatory cohort had an AUC of 0.85 (95% confidence interval, 0.82-0.89) and in the hospitalized cohort had an AUC of 0.69 (95% confidence interval, 0.64-0.75) against the electronic health record data. In the ambulatory cohort, the consensus radiologists' reading had an AUC of 0.89 (95% confidence interval, 0.86-0.92) relative to the CNN. Multivariate linear regression of Δ vasc in the ambulatory cohort demonstrated significant negative associations with non-English-language preference (β = -0.083, P < .05) and Black or Hispanic race/ethnicity (β = -0.048, P < .05) and positive associations with age (β = 0.005, P < .001) and sex (β = 0.044, P < .05). For the hospitalized cohort, age was also significant (β = 0.003, P < .01), as was social deprivation index (β = 0.002, P < .05). The Δ vasc variable (odds ratio [OR], 0.34), Black or Hispanic race/ethnicity (OR, 1.58), non-English-language preference (OR, 1.74), and site (OR, 0.22) were independent predictors of having one or more hierarchical condition category codes (P < .01 for all) in the combined patient cohort.Conclusions A CNN model was predictive of aortic atherosclerosis in two cohorts (one ambulatory and one hospitalized) with COVID-19. The discrepancy between the CNN model and the administrative code, Δ vasc, was associated with language preference in the ambulatory cohort; in the hospitalized cohort, this discrepancy was associated with social deprivation index. The absence of administrative code(s) was associated with Δ vasc in the combined cohorts, suggesting that Δ vasc is an independent predictor of health disparities. This may suggest that biomarkers extracted from routine imaging studies and compared with electronic health record data could play a role in enhancing value-based health care for traditionally underserved or disadvantaged patients for whom barriers to care exist.
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
AI may exacerbate societal healthcare disparities.
“There is a significant concern that AI may exacerbate societal healthcare disparities [24].”
From Implementing Artificial Intelligence in Critical Care Medicine: a consensus of 22. (2025) · cites this paper as [24]
Recent studies have shown that AI models can learn features that are 'invisible' to a human expert.
“Recent studies have shown that AI models can learn features that are “invisible” to a human expert.”
From AI pitfalls and what not to do: mitigating bias in AI (2023) · cites this paper as [27]
Chest X-rays can be used to predict ICD-10 codes related to cardiac disease, including conduction abnormalities, with accuracies above 0.76.
“For example, CXRs can be used to predict ICD-10 codes related to cardiac disease (including conduction abnormalities) with accuracies of above 0.76. 27”
From AI pitfalls and what not to do: mitigating bias in AI (2023) · cites this paper as [27]
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Cites
DeLong, E. R., DeLong, D. M., & Clarke-Pearson, D. L. (1988) · Biometrics · doi:10.2307/2531595 · Cited by 22,649 · cited 2× · see citing works
“…classi fi er ’ s predictions for vascular disease compared with the EHR-based administrative data (HCC-108) showed areas under the curve of 0.851 (95% CI, 0.816-0.887) in the ambulatory cohort and 0.694 (95% CI, 0.641-0.748) in the hospitalized cohort, with P .001, using DeLong ’ s method [16].”
MethodInfluentialRichardson, S., Hirsch, J. S., Narasimhan, M., Crawford, J. M., McGinn, T., Davidson, K. W., and the Northwell COVID-19 Research Consortium, Barnaby, D. P., Becker, L. B., Chelico, J. D., et al. (2020) · JAMA · doi:10.1001/jama.2020.6775 · Cited by 9,254 · cited 1× · see citing works
“Comorbidities are strongly associated with increased severity of coronavirus disease 2019 (COVID-19), disproportionately affecting racial and ethnic minorities [1].”
BackgroundFactors associated with COVID-19-related death using OpenSAFELY
Williamson, E. J., Walker, A. J., Bhaskaran, K., Bacon, S., Bates, C., Morton, C. E., Curtis, H. J., Mehrkar, A., Evans, D., Inglesby, P., et al. (2020) · Nature · doi:10.1038/s41586-020-2521-4 · Cited by 6,674 · see citing works
Yang, J., Zheng, Y., Gou, X., Pu, K., Chen, Z., Guo, Q., Ji, R., Wang, H., Wang, Y., & Zhou, Y. (2020) · International Journal of Infectious Diseases · doi:10.1016/j.ijid.2020.03.017 · Cited by 4,206 · cited 1× · see citing works
“Racial/ethnic minority populations tend to have a greater degree of underlying comorbidities, which can increase the risk for mortality from COVID-19 [2-4].”
BackgroundCOVID-19 and Racial/Ethnic Disparities
Hooper, M. W., Nápoles, A. M., & Pérez‐Stable, E. J. (2020) · JAMA · doi:10.1001/jama.2020.8598 · Cited by 2,047 · cited 2× · see citing works
“Racial/ethnic minority populations tend to have a greater degree of underlying comorbidities, which can increase the risk for mortality from COVID-19 [2-4].”
BackgroundWard, J. B., Gartner, D. R., Keyes, K. M., Fliss, M. D., McClure, E. S., & Robinson, W. R. (2018) · Annals of Epidemiology · doi:10.1016/j.annepidem.2018.09.007 · Cited by 233 · cited 1× · see citing works
“Detecting and measuring health disparities remains dif fi cult and complex [6] and requires new technologies and techniques in a multidisciplinary approach.”
BackgroundGallino, A., Aboyans, V., Diehm, C., Cosentino, F., Stricker, H., Falk, E., Schouten, O., Lekakis, J., Amann-Vesti, B. R., Siclari, F., et al. (2014) · European Heart Journal · doi:10.1093/eurheartj/ehu071 · Cited by 184 · cited 1× · see citing works
“Although aortic atherosclerosis is exceedingly common with advanced age [19], the discrepancy in its reporting may indicate more important health disparities.”
BackgroundKing, J. T., Yoon, J. S., Rentsch, C. T., Tate, J. P., Park, L. S., Kidwai‐Khan, F., Skanderson, M., Hauser, R. G., Jacobson, D., Erdos, J., et al. (2020) · PLoS ONE · doi:10.1371/journal.pone.0241825 · Cited by 89 · cited 1× · see citing works
“These administrative data also predict mortality in patients with COVID-19 [9].”
BackgroundStraw, I. (2020) · Artificial Intelligence in Medicine · doi:10.1016/j.artmed.2020.101965 · Cited by 86 · cited 1× · see citing works
“Deep learning (DL) has been extensively documented to propagate health care disparities and biases, mostly through the use of biased training data, limiting its generalizability [11].”
BackgroundLanguage and Health Equity during COVID-19: Lessons and Opportunities
Ortega, P., Martínez, G. A., & Diamond, L. C. (2020) · Journal of Health Care for the Poor and Underserved · doi:10.1353/hpu.2020.0114 · Cited by 77 · cited 1× · see citing works
“In addition, there are known linguistic barriers, such as speaking only a non-English language, that have been magni fi ed during the COVID-19 pandemic [18].”
BackgroundBrady, A. P., Bello, J. A., Derchi, L. E., Fuchsjäger, M., Goergen, S., Krestin, G. P., Lee, E. J. Y., Levin, D. C., Pressacco, J., Rao, V. M., et al. (2020) · Radiology · doi:10.1148/radiol.2020209027 · Cited by 73 · see citing works
Joseph, N. P., Reid, N., Som, A., Li, M., Hyle, E. P., Dugdale, C. M., Lang, M., Betancourt, J. R., Deng, F. M., Mendoza, D. P., et al. (2020) · Radiology · doi:10.1148/radiol.2020202602 · Cited by 67 · see citing works
A Review on Methods of Risk Adjustment and their Use in Integrated Healthcare Systems
Juhnke, C., Bethge, S., & Mühlbacher, A. C. (2016) · International Journal of Integrated Care · doi:10.5334/ijic.2500 · Cited by 59 · cited 1× · see citing works
“categories (HCCs), that have predictive power for future hospital costs [8].”
BackgroundRacial disparities in provider-patient communication of incidental medical findings
Schut, R. A. (2021) · Social Science & Medicine · doi:10.1016/j.socscimed.2021.113901 · Cited by 31 · cited 2× · see citing works
“Social Deprivation Index To control for geographic health inequities, we imputed the publicly available social deprivation index (SDI) by referencing the associated ZIP code tabulation areas [14].”
BackgroundMethodMosley, D. G., Peterson, E., & Martin, D. C. (2009) · Journal of the American Geriatrics Society · doi:10.1111/j.1532-5415.2009.02558.x · Cited by 27 · cited 1× · see citing works
“HCC codes are also predictive of repeat admission [22].”
BackgroundKERR, I. H. (1974) · British Journal of Anaesthesia · doi:10.1093/bja/46.8.558 · Cited by 27 · cited 1× · see citing works
“Imaging biomarkers extracted from CXRs offer unique opportunities to identify undocumented, underdiagnosed, or undiagnosed illnesses in high-risk patients because they are relatively common and inexpensive tests, frequently performed either around or at admission [20,21].”
BackgroundThe Risks and Rewards of Value-Based Reimbursement
Henkel, R. J., & Maryland, P. A. (2015) · Frontiers of Health Services Management · doi:10.1097/01974520-201510000-00002 · Cited by 26 · cited 1× · see citing works
“As we transition to a more value-based model for health care delivery and reimbursement, it will be increasingly important to extract the maximum possible administrative codes from available data, because cost-effective care and institutional pro fi tability are tightly linked in VBH [17].”
MethodKerr, I. H., Tm, O., & Dictionary, E. (1980) · PubMed · Cited by 23 · see citing works
Pyrros, A. T., Flanders, A. E., Rodríguez-Fernández, J. M., Chen, A., Cole, P., Wenzke, D. R., Hart, E. M., Harford, S., Horowitz, J. M., Nikolaidis, P., et al. (2021) · Academic Radiology · doi:10.1016/j.acra.2021.05.002 · Cited by 8 · cited 3× · see citing works
“Using a CNN-based DL classi fi er [10] on CXRs of patients with COVID-19, we can predict the presence of vascular disease.”
MethodInfluential
Flagged first, then by citations
Cited by
Sorted by citations
AI pitfalls and what not to do: mitigating bias in AI
Gichoya, J. W., Thomas, K., Celi, L. A., Safdar, N., Banerjee, I., Banja, J. D., Seyyed-Kalantari, L., Trivedi, H., & Purkayastha, S. (2023) · The British Journal of Radiology · doi:10.1259/bjr.20230023 · Cited by 277 · see citing works
“27 More concerning about these models is that information on social determinants of health like area deprivation index are learned and encoded on the CXRs. CXRs can also be used to predict demographic information—including insurance, self-reported race, and age.”
Implementing Artificial Intelligence in Critical Care Medicine: a consensus of 22.
Cecconi, M., Greco, M., Shickel, B., Angus, D. C., Bailey, H., Bignami, E., Calandra, T., Celi, L. A., Einav, S., Elbers, P., et al. (2025) · Critical Care · doi:10.1186/s13054-025-05532-2 · Cited by 80 · see citing works
“When considering AI’s potential influence on physicians'choices and behaviour, the possibility of including or reinforcing biases should be examined rigorously to avoid perpetuating existing health inequities and unfair data-driven associations [24].”
The Algorithmic Divide: A Systematic Review on AI-Driven Racial Disparities in Healthcare
Haider, S. A., Borna, S., Gomez-Cabello, C. A., Pressman, S. M., Haider, C. R., & Forte, A. J. (2024) · Journal of Racial and Ethnic Health Disparities · doi:10.1007/s40615-024-02237-0 · Cited by 66 · see citing works
“For instance, using zip codes as proxies for socio-economic status can reinforce geographic and racial disparities [71, 72].”
Opportunistic detection of type 2 diabetes using deep learning from frontal chest radiographs
Pyrros, A. T., Borstelmann, S. M., Mantravadi, R., Zaiman, Z., Thomas, K., Price, B., Greenstein, E., Siddiqui, N., Willis, M., Shulhan, I., et al. (2023) · Nature Communications · doi:10.1038/s41467-023-39631-x · Cited by 60 · see citing works
An intentional approach to managing bias in general purpose embedding models
Weng, W., Sellergen, A., Kiraly, A. P., D’Amour, A., Park, J., Pilgrim, R., Pfohl, S., Lau, C. T., Natarajan, V., Azizi, S., et al. (2024) · The Lancet Digital Health · doi:10.1016/s2589-7500(23)00227-3 · Cited by 17 · see citing works
Fiske, A., Blacker, S., Geneviève, L. D., Willem, T., Fritzsche, M.-C., Buyx, A., Celi, L. A., & McLennan, S. (2025) · The Lancet Digital Health · doi:10.1016/j.landig.2025.01.003 · Cited by 20 · see citing works
Pyrros, A. T., Rodríguez-Fernández, J. M., Borstelmann, S. M., Flanders, A. E., Wenzke, D. R., Hart, E. M., Horowitz, J. M., Nikolaidis, P., Willis, M., Chen, A., et al. (2022) · PLOS Digital Health · doi:10.1371/journal.pdig.0000057 · Cited by 15 · see citing works
“Frontal chest radiographs (CXRs) have been used to directly predict or quantify patient comorbidities that contribute to outcomes [13].”
BackgroundGandomkar, Z., Khong, P. L., Punch, A., & Lewis, S. J. (2022) · Journal of Imaging Informatics in Medicine · doi:10.1007/s10278-022-00631-w · Cited by 10 · see citing works
“In medical imaging, studies on brain MRI [17, 18, 24] and studies involving optical coherence tomography images, histopathological images [19], and chest X-rays [25] have shown the effectiveness of this approach to identify regions of importance.”
BackgroundClunie, D., Taylor, A. J., Bisson, T., Gutman, D. A., Xiao, Y., Schwarz, C. G., Greve, D. N., Gichoya, J. W., Shih, G., Kline, A. S., et al. (2024) · Journal of Imaging Informatics in Medicine · doi:10.1007/s10278-024-01183-x · Cited by 10 · see citing works
Machine Learning in Cardiovascular Imaging: A Scoping Review of Published Literature
Rouzrokh, P., Khosravi, B., Vahdati, S., Moassefi, M., Faghani, S., Mahmoudi, E., Chalian, H., & Erickson, B. J. (2022) · Current Radiology Reports · doi:10.1007/s40134-022-00407-8 · Cited by 10 · see citing works
“, which applied ML to CXRs to analyze racial/ethnic and socioeconomic differences in the prevalence of atherosclerotic vascular disease [14].”
Dual smart sensor data-based deep learning network for premature infant hypoglycemia detection
Shafiq, M., Kavitha, J., Rinku, D. R., Kumar, N. K. S., Poon, K., Jaffar, A. Y., & Saravanan, V. (2025) · Scientific Reports · doi:10.1038/s41598-025-03864-1 · Cited by 2 · see citing works
Responsible Use of AI: Ethics, Biases, and Fairness
Blades, K. (2025) · Springer eBooks · doi:10.1007/978-3-031-83526-1_10 · Cited by 1 · see citing works
Huang, S. S., Chaisson, L. H., Galanter, W., Jalali, A., Menchaca, M., Parde, N., Rodríguez-Fernández, J. M., Trotter, A., & Kochendorfer, K. M. (2023) · Journal of Clinical and Translational Science · doi:10.1017/cts.2023.26
Yazbeck, H., Assaf, J. F., Wheary, L., Upaphong, P., Elsharawi, R., Jackson, J., Wittberg, D. M., Nash, S. D., Admassu, F., Tadesse, Z., et al. (2026) · Ophthalmology Science · doi:10.1016/j.xops.2026.101184
Implementing responsible artificial intelligence-driven precision medicine in critical care
Ennouri, E., Boussarsar, M., Saida, I. B., Meddeb, K., & Toumi, R. (2026) · Elsevier eBooks · doi:10.1016/b978-0-443-36554-6.00021-0
- Authors
- Pyrros, A., Rodríguez-Fernández, J. M., Borstelmann, S. M., Gichoya, J. W., Horowitz, J. M., Fornelli, B., Siddiqui, N., Velichko, Y., Koyejo, O., & Galanter, W.
- Affiliations
- Adventist Hinsdale Hospital (US)
- University of Illinois Chicago (US)
- University of Central Florida (US)
- Emory University (US)
- Northwestern University (US)
- Corresponding author
- Pyrros, Ayis
- ORCID
- Venue
- Journal of the American College of Radiology
- Publisher
- Elsevier BV
- Published
- 2022-01
- Volume
- 19
- Issue
- 1
- Pages
- 184-191
- ISSN
- 1546-1440 (ISSN-L)
- PMC
- PMC8820271
- OpenAlex
- W4205855895
- Cited by
- 15
- References
- 19
- Citations by year
- 2026: 2 · 2025: 4 · 2024: 3 · 2023: 3 · 2022: 3
- FWCI
- 1.85
- Type
- journal article
- Language
- en
- Update notices
- erratum (2022-03-01)
- Open access
- green
- www.ncbi.nlm.nih.gov
- Locations
- Journal of the American College of Radiology (publishedVersion)
- PubMed (publishedVersion)
- PubMed Central (acceptedVersion)
- Funders
- University of Illinois at Chicago Center for Clinical and Translational Science
- National Institutes of Health
- National Institute of Biomedical Imaging and Bioengineering (75N92020C00008, 75N92020C00021)
- Topics
- Health Sciences › Medicine › Radiology, Nuclear Medicine and Imaging › COVID-19 diagnosis using AI
- Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
- Physical Sciences › Computer Science › Artificial Intelligence › Machine Learning in Healthcare
- Keywords
- racial and ethnic health disparities, socioeconomic disparities, atherosclerosis, convolutional neural networks, COVID-19 pandemic, language preference, social deprivation index, radiologist consensus, hospitalized patients, outpatients, cardiovascular risk stratification
- MeSH terms
- Deep Learning, COVID-19, SARS-CoV-2, Social Deprivation, Ethnicity, Carcinoma, Hepatocellular, Humans, Liver Neoplasms, Radiography, Retrospective Studies
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
