A 2026 study<\/a> conducted by researchers at Ann & Robert H. Lurie Children\u2019s Hospital of Chicago provides a clear example. In its analysis of over 74,000 pediatric emergency department visits, researchers found that undertriage\u2014assigning a lower severity score than the level of care a child actually needed\u2014was more likely for children who were Black or Hispanic or those who preferred Spanish over English.<\/p>\n Jennifer Hoffmann, MD\u2014the study\u2019s lead author\u2014explains that these disparities have multiple drivers. \u201cThe disparities we observed may be driven by a combination of implicit bias, communication barriers, and structural factors within the healthcare system,\u201d she says. \u201cFor example, clinicians may unconsciously perceive symptoms differently based on race or ethnicity, and underuse of professional interpreters can lead to miscommunication for Spanish-speaking families. These factors, along with broader systemic inequities, can influence how urgency is assessed during triage.\u201d<\/p>\n Empirical research has repeatedly shown disparities in how patients are assessed and treated. Race, gender, and socioeconomic status<\/a> influence diagnostic outcomes, access to care, and perceived risk levels. These patterns are rooted in long-standing structural inequalities that shape the entire mental health ecosystem. Any attempt to address bias must contend with these systemic realities rather than treat bias as an isolated issue.<\/p>\n \u200e<\/a> Triage platforms\u2014digital clinical decision support tools designed to standardize and support clinical decision-making\u2014appear promising for mitigating bias. They include artificial intelligence (AI)\u2013driven risk prediction models, structured digital intake platforms, and symptom assessment tools that guide patients<\/a> and clinicians through consistent workflows. By applying the same criteria across cases, these systems aim to limit the influence of individual judgment. They can prioritize patients based on risk, flag urgent cases, and suggest care pathways, particularly in settings where specialist expertise is limited.<\/p>\n Sam Zand, MD\u2014a psychiatrist and CEO of Anywhere Clinic<\/a>\u2014notes the complexity of mental health triage. \u201cWhen evaluating mental health patients in the emergency department, it can often be difficult to assess the level of risk related to care of these patients,\u201d he explains. \u201cMisclassification of risk will lead to either an under-triaged patient and\/or over-triaged patient\u2014both of which have dire consequences regarding patient safety and staff resources.\u201d<\/p>\n Kenneth Perry, MD\u2014an emergency physician in South Carolina\u2014sees potential for AI to improve these challenging decisions. \u201cAI tools would be very helpful in determining which patients need inpatient medical care and which are safe enough for outpatient follow up,\u201d he says. \u201cIf there was a way to keep the bias of the questioner from the interaction, more correct, efficient dispositions could be possible.\u201d<\/p>\n The Challenge of Algorithmic Bias<\/p>\n Despite their potential, however, the effectiveness of these tools is fundamentally constrained by data quality. These systems are trained on historical datasets that often reflect the very disparities<\/a> found in the Ann & Robert H. Lurie Children\u2019s Hospital of Chicago study.<\/p>\n Mai Uchida, MD\u2014a pediatric psychiatrist and AI investigator at Harvard University and the NTT Research Brain Science Center, and director of pediatric depression at Massachusetts General Hospital\u2014explains how bias becomes embedded in these systems. \u201cIf historical data reflects disparities\u2014such as underdiagnosis of internalizing disorders in boys, overdiagnosis of behavioral disorders in Black children, or differential access to care across socioeconomic groups\u2014then AI may not only reproduce these patterns but reinforce them with a veneer of objectivity,\u201d she says.<\/p>\n Flawed assumptions can get amplified. According to Dr Saravanan Thirumalai Thangarajan, visiting scientist at Harvard T.H. Chan School of Public Health and international fellow at Ariadne Labs, Brigham and Women\u2019s Hospital, \u201cAI triage tools in pediatric mental health do not neutralize bias. They can industrialize it. Many triage models use prior healthcare utilization as a proxy for need, but utilization reflects access, not illness burden. A model trained on unequal access patterns does not simply detect risk. It learns who the system already knows how to see.\u201d<\/p>\n \u200e<\/a> Yet, these tools have practical value that shouldn\u2019t be dismissed. When designed and implemented carefully, they can reduce certain types of bias and strengthen triage systems.<\/p>\n![](\"https:\/\/asset.jmir.pub\/assets\/337f8c65-5a95-11f1-9eca-c167e27f4f75.png\"\/)
<\/a>The Role of Clinical Decision Support<\/p>\n![](\"https:\/\/asset.jmir.pub\/assets\/338f91f1-5a95-11f1-9eca-c167e27f4f75.png\"\/)
<\/a>The Value of Standardization and Transparency<\/p>\n