Does the diagnosis a patient receives vary when all that differs is their sex or race? Systemic lupus erythematosus (SLE) is a protean, chronic autoimmune disease that affects multiple organ systems, with significant morbidity and decreased life expectancy (1). Symptoms vary but include fatigue, fever, rashes, joint pain, chest pain, and proteinuria. The reported average time from first symptom to diagnosis has been as high as 6 years (2). Women of childbearing age are disproportionately affected (female:male ratio of approximately 9:1), according to the literature (3). Although female patients generally have more frequent exacerbations of disease, male patients may have more significant accrual of multisystem damage and disease severity (4). Disparities by race/ethnicity are also well-described, with higher prevalence and more severe disease (including the lupus nephritis phenotype) among Black, Hispanic, and Asian individuals compared with White individuals (5–7). In a recent US study, Izmirly et al. (7) reported that per 100,000 population, the prevalence of SLE was 230.9 in Black females, 84.7 in White females, 26.7 in Black males, and 8.9 in White males. While we understand that the clinical presentation of SLE differs by sex and race, we do not understand how much the path to diagnosis differs more upstream of the observed natural history of the disease. Classification criteria exist for research; however, the field lacks definitive diagnostic criteria for clinical decision-making in real-world practice. This is further complicated by the heterogeneity of SLE. Thus, the physician, who some would describe as a natural Bayesian, ultimately plays a key role in identifying SLE and initiating appropriate treatment. Understanding how physicians’ cognitive processes are involved in diagnosing SLE may inform how clinical decision-making practices unfold in the future. Further, uncovering and addressing potential cognitive biases may reduce delayed and missed opportunities for SLE diagnosis during clinical encounters and prevent unnecessary damage. Using an experimental approach, we evaluated the extent to which a patient’s sex, and secondarily race, factors into clinical decision-making. We hypothesized that when presented with the same clinical scenario with sex reversed, physicians would be less likely to diagnose males with SLE than females. Between January and May 2019, US rheumatologists listed in the American Medical Association’s Physician Masterfile were e-mailed via MMS Lists (Medical Marketing Service, Inc., Schaumburg, Illinois), which broadcast messages to 4,909 rheumatologists. Participants were offered an incentive (a $20 gift card for a large online retailer which increased to $40 at the third mailing). Two board-certified rheumatologists (Y.C., C.H.F.) developed 5 clinical scenarios in a format similar to that of a board certification examination. Because our emphasis was on SLE, we included 2 SLE cases (mild lupus and lupus nephritis) and 3 “control” cases (gout, ankylosing spondylitis, and sarcoidosis). Gout and ankylosing spondylitis are reportedly more common in males (8–10). Each case scenario provided information on the patient’s sex and race, which varied randomly, and outlined relevant symptoms, medical history, physical examination findings, and laboratory results—for example, “A 32-year-old (INSERT RACE) (INSERT SEX) is evaluated in the urgent care clinic for a 1-week history of fever, joint pain, and fatigue.” Eleven rheumatologists and 10 primary-care providers blinded to the survey’s purpose pilot-tested the instrument and received a $20 gift card incentive. We consulted physicians in rheumatology, nephrology, and renal pathology to reduce ambiguity of cases. Participating physicians provided information about their training, expertise, and demographic characteristics, including race/ethnicity, sex, and birth year. Each participant received a random race/sex variant (Black female, White female, Black male, White male) of each of the 5 cases (Figure 1). We determined whether the participants correctly diagnosed the cases as our primary outcome. Each case had an official diagnosis based on established classification criteria, which was also supported by clinician consensus during development and pilot work. Review time (amount of time needed to complete the question for each vignette; automatically recorded by the survey software) was a secondary measure serving as a proxy for cognitive bias, where a longer review time, even with the correct diagnosis, might indicate greater uncertainty when considering a particular patient (and their sex and race) for that diagnosis (11). Case vignette pool and randomization mechanism used in a survey of how patient sex and race influence clinical decision-making among rheumatologists, United States, January–May 2019. A) Pool of clinical cases by sex and race; B) an example of 6 sets of surveys. Each participant received the same 5 clinical cases, randomized by race and sex. The scenarios were presented in the same order, with the exception of systemic lupus erythematosus (SLE) and lupus nephritis, as shown in the set of examples. BF, Black female; BM, Black male; P, participant; WF, White female; WM, White male. We summarized data on respondent characteristics using frequencies and proportions. Each case’s correct diagnosis was compared with the participant’s response. We calculated the numbers of correct and incorrect assessments by variant for each case and the mean amount of review time needed for each variant for each case, further stratifying the results by correct response. Of the 4,909 rheumatologists who were sent an e-mail, 296 participated. A majority identified as White (60.5%) and male (53.0%). Most participants had over 10 years of clinical experience (52.0%), worked in a private group practice (40.5%) or an academic setting (39.2%), and were located in urban centers (52.7%) (see Web Table 1, available at https://doi.org/10.1093/aje/kwab199). When considering the intersection of race and sex, we found that Black females were most likely to be correctly diagnosed with SLE (89.2%), followed by White females (85.7%), Black males (84.1%), and White males (77.6%). This discrepancy was less pronounced for lupus nephritis, but still pronounced when compared with the 3 control cases. The pattern was similar for our secondary outcome of review time. Among the respondents who ultimately answered correctly, review time was longer for White male SLE cases (36.4 seconds) than for Black female cases (16.0 seconds) (Table 1). Survey Results From a Randomized Experiment Carried Out Among 296 Rheumatologists on How Patient Sex and Race Influence Clinical Decision-Making, United States, January–May 2019 Abbreviation: SLE, systemic lupus erythematosus. a Time needed to complete vignette; automatically recorded by survey software. Our goal in this pilot work was to assess whether a patient’s sex and/or race influences what diagnosis they receive. For the lupus nephritis case, there was no difference in the proportion of physicians responding correctly for White females (92.9%), Black females (92.1%), and Black males (91.3%). However, respondents were modestly less likely to provide the correct diagnosis for White males (86.4%). For milder SLE, we noted a “dose response” in the proportion correctly diagnosed commensurate with the demographic distribution of SLE prevalence in the evidence base, with White males having the lowest correct response (77.6%), followed by Black males (84.1%), White females (85.7%), and Black females (89.2%). This begs the question, Is sex or race implicitly a diagnostic criterion for SLE? Researchers have used alternative approaches when examining sex bias in other diagnoses. In a small survey-based study of clinical decision-making for fibromyalgia, rheumatologists were more likely to require a physical finding to support the diagnosis in males (12). Combining these findings with the present study, with the observation that the minority (i.e., the patient group less likely to have the disease on the basis of the evidence) often has more severe disease and poorer outcomes for numerous conditions (13), raises some questions: Is this simply a manifestation of how clinicians tend to diagnose consistent with what the evidence says is most common? Or, as has been characterized in psychology and psychiatry, where symptoms are often subjective (14), is this specific to SLE in our study because it is a complex disease without diagnostic criteria? The classification criteria for SLE, intended to be used for research and not diagnostic purposes, include neither sex nor race. The evidence base states that females comprise approximately 90% of the SLE patient population; therefore (and appropriately), one’s prior probability of SLE in clinical decision-making may be lower when presented with a male case. Differences in response time may reflect extra caution by the respondent. These results may not be surprising because males comprise the minority of lupus patients and these findings simply perpetuate that truth, as cognitive biases related to both anchoring and representativeness might lead us to expect (15). However, rheumatologists did better for cases generally considered more male-predominant (e.g., gout; approximately 1:3.5 female:male ratio) regardless of case variant, suggesting that these cognitive biases alone may not explain our findings (8, 9). While we do not think this explains the reported sex ratio in SLE, misdiagnosis or potentially delayed diagnosis of SLE in males may perpetuate the established sex ratio and disparities in severity and outcomes. We theorize that White males, for example, are faced with 2 possible futures. First, patients with mild cases could continue to obtain care for the wrong diagnosis and simply never be diagnosed; or, second, the patient could develop end-organ manifestations of SLE (e.g., lupus nephritis) and contribute to the narrative that males present more severely or with more organ involvement than females. This is worrisome, because it could mean that male patients with SLE may inadvertently be diagnosed later, with irreversible damage, increasing the costs and likelihood of refractory disease and reducing the quality of these patients’ lives. Although our response rate was low, the participating rheumatologists were generally similar to the rheumatology workforce according to the ACR 2015 Work Force Study Report, except that our group was more racially and ethnically diverse (16). We cannot exclude the possibility of selection bias and limited generalizability on the basis of participation, and we were unable to examine heterogeneity by respondent-vignette race/sex concordance in our small pilot study. In future work, we plan to examine race, intersectionality, and how diagnosis varies by respondent characteristics. We are less concerned about internal validity threats given the randomization of our exposure (race/sex variants). Additionally, a random subset of surveys included only 4 vignettes, where ankylosing spondylitis cases were randomly excluded from the survey. Physicians adhere to evidence-based guidelines where the “evidence” in the literature suggest that females comprise most patients with SLE. While we are not suggesting that the evidence is flawed, herein lies the paradox of personalized medicine: The evidence base is most informative when the patient looks like the “average” patient in the literature. We argue that assigning characteristics of the population to the individual may lead to diagnostic errors. This is especially salient in SLE, an infrequent, heterogeneous disease. We acknowledge that the lack of an “average” patient in SLE may contribute to the need to find a heuristic that aids clinical decision-making. This work was supported by the Discovery Innovation Award from Stanford University School of Medicine through the Biomedical Innovation Initiative. Our data availability statement is available online (https://tinyurl.com/dxbiasrheum) through the Open Science Framework. We thank Janine Bodea Sung for her technical assistance with our survey instrument and Renee M. Miller for her help with incentive distribution and project management. Conflict of interest: none declared.
No takes yet. Share an insight, caveat, or question.
Simard et al. (2021) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: