Representation and misdiagnosis of dark skin in a large-scale visual diagnostic challenge

Abstract

To the Editor: Images of dark skin are routinely underrepresented in medical education and research.1Alvarado S.M. Feng H. Representation of dark skin images of common dermatologic conditions in educational resources: a cross-sectional analysis.J Am Acad Dermatol. 2021; 84: 1427-1431Abstract Full Text Full Text PDF PubMed Scopus (19) Google Scholar,2Lester J.C. Jia J.L. Zhang L. Okoye G.A. Linos E. Absence of images of skin of colour in publications of COVID-19 skin manifestations.Br J Dermatol. 2020; 183: 593-595Crossref PubMed Scopus (39) Google Scholar This lack of representation may influence diagnostic ability among clinicians and health outcomes for patients of color. In turn, this may disproportionately affect certain race/ethnicity groups; for example, nearly half of the dermatologists report inadequate training on skin conditions in Black patients.3Buster K.J. Stevens E.I. Elmets C.A. Dermatologic health disparities.Dermatol Clin. 2012; 30 (viii): 53-59Abstract Full Text Full Text PDF PubMed Scopus (96) Google Scholar In addition, a recent single-center study of medical students showed decreased diagnostic accuracy for images of darker skin, raising the possibility that this may be a widespread phenomenon.4Fenton A. Elliott E. Shahbandi A. et al.Medical students' ability to diagnose common dermatologic conditions in skin of color.J Am Acad Dermatol. 2020; 83: 957-958Abstract Full Text Full Text PDF PubMed Scopus (24) Google Scholar Using over 40 million responses from the weekly multiple-choice New England Journal of Medicine Image Challenge, we aimed to investigate the differences in representation and image-based diagnostic accuracy by skin type.5Image challenge. New England Journal of Medicine.https://www.nejm.org/image-challengeDate accessed: July 21, 2020Google Scholar Our study was deemed exempt by the institutional review board of the University of Texas at Austin. Between October 13, 2005 and July 9, 2020, 424 of 767 images (55.3%) had an identifiable skin type. Of the 424 images, 307 (72.4%) contained cutaneous or subcutaneous pathology such as rashes and lesions. The remaining 117 (27.6%) images contained pathology in which the skin type was less likely to influence the diagnosis, including oral (n = 53), ocular (n = 27), acral (n = 16), hair (n = 5), genital mucosa (n = 4), and other sites; these images comprised the reference group. Each image was labeled with the Fitzpatrick skin type by a board-certified dermatologist and classified into “light” (I-II), “intermediate” (III-IV), or “dark” (V-VI). The differences in represented proportions and diagnostic accuracy were evaluated using 2-sample t test, and trends were evaluated using the Mann-Kendall test. All questions recorded between 35,000 and 220,000 responses. There were 205 images of light skin (47.6%), 175 of intermediate skin (41.7%), and 44 of dark skin (10.7%). These differences are significant (P < .001) and have not trended significantly either upward or downward since 2005 (Fig 1). On average, 44.3% of responses correctly diagnosed cutaneous/subcutaneous pathology in dark skin compared with 50.5% in intermediate skin (P = .020) and 50.4% in light skin (P = .023) (Fig 2). By contrast, neither of these differences in diagnostic accuracy across the skin type were significant in the reference group (P = .80 and P = .52, data not shown). We did not detect significant trends in proportion correct since 2005 for any group.Fig 2Diagnostic accuracy among skin types in the Image Challenge. The differences in proportion correct between the Fitzpatrick skin types in images of cutaneous or subcutaneous pathology from the New England Journal of Medicine Image Challenge. Statistically significant differences are denoted by an asterisk (∗).View Large Image Figure ViewerDownload Hi-res image Download (PPT) To our knowledge, this is the largest empirical evaluation of diagnostic accuracy for conditions in dark skin and the first to study their trends over time. Limitations include the semi-subjectivity of the Fitzpatrick skin type, potential skin type misclassification due to lighting or tanned skin, image review by a single dermatologist, sample bias toward more notable cases, lack of stratification by respondent specialty or training level, and lack of data on disease class or body region. Our data indicate that New England Journal of Medicine Image Challenge respondents—who likely include clinicians and clinician-trainees involved in patient care—may be less prepared to recognize the diagnostic signs on dark skin and that neither representation nor diagnostic differences have changed significantly since 2005. Although it is likely that most respondents are non-dermatologists, these results point to a potential deficiency in general medical education that needs to be addressed, and the dermatologists may be poised to help. None disclosed.