Emergency departments in medical centers across the United States face challenges in patient triage, which is essential to prioritize patients based on the urgency of their conditions. Traditional triage methods often rely on experienced clinical staff such as registered nurses, using structured systems like the Manchester Triage System (MTS) or the Emergency Severity Index (ESI). These methods depend on human judgment but can be affected by workload stress and limited resources.
AI-based decision-support tools use machine learning algorithms or large language models trained on extensive clinical data to analyze patient symptoms, history, and electronic health records quickly. Their goal is to replicate or improve triage accuracy with less human intervention.
Two AI-based triage systems have gained recent attention:
A study at Kepler University Hospital in Austria compared SMASS with the traditional Manchester Triage System. It looked at 1,021 adult patients with non-traumatic complaints and found notable differences:
The agreement between SMASS and MTS was low, with a Cohen’s kappa of 0.167, which suggests significant mismatches in patient categorization. This raises concerns about both overtriage and undertriage, indicating SMASS needs further validation before regular clinical use.
In contrast, U.S. studies on LLM-based tools such as ChatGPT and Microsoft Copilot showed different results. In a study involving 468 adult patients in a busy urban emergency department:
A multicenter study involving 6,657 patients confirmed these findings, with ChatGPT-4o showing near-perfect agreement with emergency physician triage decisions (Cohen’s kappa of 0.833) and superior classification scores compared to triage nurses.
These studies suggest advanced AI models could assist clinical staff, especially in identifying severely ill patients in need of immediate care. However, implementing AI needs caution, including attention to factors like emergency department capacity to avoid overcrowding or delays.
AI has also influenced emergency radiology workflows. One study analyzed AI-driven triage in 600 CT and MRI emergency cases, finding:
These improvements are relevant to U.S. emergency departments handling large imaging volumes. Faster triage can enable quicker interventions and reduce delays that affect patient outcomes.
Emergency departments in the United States often encounter overcrowding, staffing shortages, and varied patient complexities. When evaluating AI triage tools, healthcare leaders should consider the following:
AI-powered automation is becoming a key element in improving front-office operations in emergency departments. For example, Simbo AI provides phone automation and AI-driven answering services that help health facilities handle patient calls, appointment scheduling, and initial assessments more efficiently.
Automated phone systems reduce the workload of administrative staff, streamline patient intake, and improve communication during busy periods. This aids patient decisions on whether to seek urgent care or wait for medical evaluation.
Besides phone automation, AI supports several areas:
These AI workflow tools can help hospitals meet regulatory standards for quality and efficiency, which is important for compliance with organizations such as The Joint Commission and the Centers for Medicare & Medicaid Services (CMS).
Healthcare decision-makers in the United States must balance AI benefits with operational challenges. Recommended steps include conducting pilot studies to evaluate AI triage locally, integrating AI gradually alongside traditional methods, and establishing protocols for human oversight.
Companies like Simbo AI offer technologies for front-office automation that ease administrative tasks, giving clinical staff more time for patient care. Combining these communication solutions with AI triage tools may improve emergency department responsiveness.
Administrators should also consider:
Reviewing the performance of AI triage systems from SMASS to LLM models like ChatGPT, alongside advances in AI radiology triage and workflow automation, can help U.S. healthcare facilities decide on technology adoption. While AI presents useful tools for emergency medicine, implementing them requires careful evaluation to maintain patient safety and operational effectiveness.
The study aims to evaluate the performance of SMASS, an AI-based decision-support tool, in comparison to the traditional Manchester Triage System (MTS) for rapid patient assessment in emergency medicine.
Patients aged 18 years or older presenting with non-traumatic complaints to the emergency department were included in the retrospective analysis.
A total of 1,021 patients were triaged with both the Manchester Triage System and the Swiss Medical Assessment System during the study period.
The mean age of the patients included in the study was 60 years, with a standard deviation of 21 years.
19% of patients categorized as ‘orange’ by MTS were classified as non-urgent by SMASS.
The agreement between SMASS and MTS classifications was low, with a Cohen’s kappa of 0.167, indicating significant discrepancies in triage categorization.
23% of patients initially classified as non-urgent by SMASS ultimately required hospitalization after evaluation and treatment.
The study revealed considerable rates of both overtriage and undertriage, indicating that AI-based triage tools may require further validation before routine clinical use.
The conclusion emphasizes that while SMASS is an innovative AI-based tool, its current discrepancies with established methods like MTS necessitate further validation before integration into emergency care.
The findings suggest that while AI has potential in triage processes, careful assessment and refinement of such systems are essential to ensure patient safety and effective emergency care delivery.
Simbo is using AI agents to automate all phone related workflows. Connect now to know more.
Schedule AI Agents Demo Now© Since 2019, Simbo AI.
