The evaluation of physical differences, or dysmorphology, is foundational to the practice of medical genetics (Innes and Lynch 2021; Solomon et al. 2023). Although molecular sequencing technologies now anchor many diagnoses, careful phenotyping remains essential for variant interpretation, clinical correlation, syndrome recognition, and guiding evaluation when testing is limited (Solomon et al. 2023). Modern clinical genetics developed largely in Europe and North America, and many of the field's teaching resources were assembled in those settings. As a result, the visual and descriptive language of facial morphology has often treated European ancestry as the implicit reference point, increasing the risk that common ancestry-associated variation is misclassified as “dysmorphic”. Today, the practice of medical genetics spans the entire globe, with both new and classically described diagnoses being made in patients of all ethnic backgrounds and geographical origins. The language we use to describe facial morphology, however, continues to reflect foundations of whiteness, with European facial features treated as “normal” or “default”. Examples include using terms such as depressed nasal bridge as though there is a single baseline (where, in practice, nasal bridge height must be interpreted in the context of family resemblance and ancestry) or making a special note of synophrys, despite it being a common trait in some ancestries and individual families. Importantly, there is a growing appreciation for Eurocentric bias in reference and educational materials in other fields of medicine, such as dermatology (Louie and Wilkes 2018). Studies have shown that dermatology resources often predominantly feature conditions as they appear on lighter skin (Louie and Wilkes 2018). This lack of diversity can lead to educational deficiencies and a potential decrease in diagnostic accuracy for patients from underrepresented groups (Louie and Wilkes 2018). Others have made the case for increased diversity in morphology teaching, and the potential obstacles to achieving this in medical genetics (Koretzky et al. 2016), but available approaches have not provided a simple, reproducible way to quantify gaps in representation within teaching resources to the populations served. Here, we employ a novel approach to move beyond a conceptual critique, performing a structured audit of the eighth and latest edition of Smith's Recognizable Patterns of Human Malformation, a foundational dysmorphology textbook for the study of medical genetics (Innes and Lynch 2021), comparing the representation of apparent ancestry in its textbook images to three relative comparator populations (World, United States, Toronto). Twelve auditors, all postgraduate physician trainees in medical genetics at the University of Toronto, were provided with copies of Smith's Recognizable Patterns of Human Malformation, 8th edition (Jones et al. 2021). The contents of the textbook were divided amongst the auditors who, in groups of two or three, assessed the photos in their assigned section of the text in duplicate or triplicate. Reviewers categorized the race of each patient depicted as one of the following: White, Black, Asian (including East Asian and Southeast Asian), or other (including Latine, North American Indigenous, South Asian, West Asian, or those of mixed or ambiguous ancestry who could not be confidently assigned to another category). Multiple images of the same patient were not counted separately where this was possible to determine. Categorizations were then discussed by other members of the same team, with conflicts resolved through re-review and group discussion. Categorizations were then rechecked by DBP or ES for accuracy and to ensure consistency across sections. To act as population comparators to the individuals represented in the text, we chose three populations: The world population, which is served by the practice of medical genetics; the US population, which is served by the primary contributors of Smith's;5 and the population of the city of Toronto, reflecting the clinical context that prompted this audit. The world population was derived from UN statistics using continental origin as a surrogate for ancestry (United Nations Department of Economic and Social Affairs Population Division 2022). The population of the United States was drawn from US Census 2020–2023 using race as identified (U.S. Census Bureau 2024a). Modifications to these statistics to fit our four-category model were made by including Latine White people and 29% of the Asian demographic which identifies as South Asian in the Other Racialized group (U.S. Census Bureau 2024b). The population of Toronto was drawn from the Canadian Census 2021, using Visible Minority Status as a surrogate for ancestry (Statistics Canada 2023). Given the work outlined above in dermatology (Louie and Wilkes 2018), we were particularly interested in two sections related to genetic disorders where skin differences are distinctive and can be essential to making the correct diagnosis (Chapter 1, Sections Q: Hamartoses and R: Ectodermal Dysplasias). This subgroup is therefore considered specifically and presented alongside the overall text and the comparator populations. In total, 664 individuals in Smith's were categorized by apparent race/ancestry group, of which 42 were in the dermatology sections. As illustrated in Figure 1, Smith's over-represents individuals of European ancestry, with 505 apparently unique individuals being categorized as white (76.1%). Furthermore, there is not only a significant underrepresentation of diverse skin tones in the important dermatology sections Q and R (14.3% non-white), but less skin tone variation in these sections compared to the overall text (23.9% non-white). These findings, in an important repository of morphological reference material for medical genetics, clearly demonstrate the Eurocentricity of dysmorphology practice and must be addressed. This is not an indictment of this essential text, but a call to action to the medical genetics community at-large to actively work to close this resource gap. Our findings further support the work of Koretzky et al. (2016), which highlighted the need for more representative morphology in medical genetics. Importantly, valuable efforts are already underway to address the issue, including the electronic atlas being assembled by Muenke et al. (2016) under the auspices of the NIH. Moreover, ongoing discussions about the future of dysmorphology within medical genetics signal the field's commitment to having this skillset evolve to meet the needs of patients (Solomon et al. 2023). The drivers of underrepresentation in published dysmorphology image archives are likely multifactorial. Structural inequities in access to specialty genetics services and differential referral pathways may shape which patients are evaluated in tertiary settings and therefore who is most likely to be photographed and represented in legacy resources (Gold et al. 2026). Evolving norms around consent and publication of identifiable clinical photographs may have influenced what images have been collected and disseminated (Cunniff et al. 2000), particularly in an increasingly digital environment where there may be farther reaching consequences (Roguljić et al. 2022). Finally, and notably, a broader history of discrimination and medical mistrust may affect the willingness of certain patients to share images for educational dissemination (Halwani 2004). Together, these factors underscore why representation cannot be assumed in well-established references and why updated, intentionally inclusive resources are needed. Pragmatic strategies can partially mitigate these limitations during training and clinical care even before major reference textbooks are updated. First, morphology should be taught and documented as context-dependent: many facial descriptors are meaningful only relative to a familial baseline. When feasible, incorporating comparison with parents and siblings as a routine assessment standard can reduce misclassification of ancestry-associated variants as “dysmorphic”, and thus improve the specificity of phenotypic data used to guide diagnosis. Second, mixed ancestry should be explicitly acknowledged in teaching resources, case discussions, and requisitions; rigid racial bins are an oversimplification, and many individuals will not map cleanly into a single category. Emphasizing family context, proportionality of features, and gestalt, rather than a single “default” face, is therefore essential. This analysis has limitations; continental origin and race are imperfect surrogates for ancestry, which is itself an imperfect marker of diversity. Although we attempted to reduce inter-observer variability in the review, it is possible that some patients, particularly those in the “other racialized” category, were miscategorized. However, prior studies such as those which established the Chicago Faces Database (Ma et al. 2015) support that members of the general population have proficiency in this superficial categorization of apparent ancestry, and where there was uncertainty in the classification of a photo the study group erred on the side of placing individuals into the “other racialized” category. As a result, the present analysis likely represents a more liberal quantification of the text's diversity. Finally, while our audit quantifies representation in one (albeit important) medical genetics textbook, we acknowledge that it does not directly measure downstream clinical outcomes such as diagnostic delays, misdiagnosis, or unnecessary testing; nor does it imply that other tools and references could not be complementary to the interest of an accurate diagnosis across multiple patient groups. Other approaches, such as computer-assisted facial phenotyping, have increasingly been leveraged to support syndrome recognition across diverse populations. For example, in cohorts of individuals with Cornelia de Lange syndrome and Noonan syndrome spanning multiple geographic and ancestry groups, reported diagnostic sensitivities are high (Dowsett et al. 2019; Kruszka et al. 2017), suggesting that at least some facial differences do generalize reliably across ancestry. Although this facial analysis data is useful context and will continue to be a potential adjunct to our practice, there remains an important opportunity to study how trainees educated with predominantly Eurocentric reference images perform in real-world settings, and whether training environments influence outcomes such as diagnostic confidence, anchoring bias, or testing behavior, including over-investigation of ancestry-associated variation. It is imperative that reference materials genetics education reflect the diversity of the population we serve, not only to ensure accurate diagnoses but also to promote equitable care for historically marginalized groups. We hope that ongoing efforts to address such gaps in our discipline will be integrated into key resources like Smith's in the future. Improving representation in morphology will also challenge the Eurocentric biases that have shaped the field and may even lead to the description of new classical morphological patterns that are less visible against the background of European ancestry. Ultimately, it is incumbent on authors and editors to continue to prioritize publishing images that showcase the breadth of morphology across ancestry groups, as well as all of those practicing medical genetics to further these important conversations. D' Arcy B. Prendergast: study design, first and second pass audit participant, data analysis, manuscript drafting and approval of final version. Emma Sullivan: second pass audit lead, data collation and analysis, manuscript revisions and approval of final version. Michael P. Mackley: first pass audit participant, manuscript revisions and approval of final version. Hanna Faghfoury: study design and initial conception, first pass audit lead and second pass audit participant, research oversight, manuscript revisions, and approval of final version. We thank the following colleagues who participated in the first pass audit of Smith's presented here: Mona Alhamed, Ashish Deshwar, Joseph Brenton Dubé, Isabel Friedman, Rebecca Mantha, Adaeze Odigwe, Rachel Youjin Oh, Sarah Redmond, Zhuo Shao, and Yiming Wang. Dr. Faghfoury thanks Dr. Neda Maghbouleh for generous conversations on race and representation, as well as directing us to resources relevant to this work. We also acknowledge David W. Smith and the authors of Smith's Recognizable Patterns of Human Malformation for its lasting educational influence in the study of dysmorphology. The authors have nothing to report. The data that support the findings of this study are available from the corresponding author upon reasonable request.
Prendergast et al. (Wed,) studied this question.