Banner
  • Revenue Cycle Management
  • COVID-19
  • Reimbursement
  • Diabetes Awareness Month
  • Risk Management
  • Patient Retention
  • Staffing
  • Medical Economics® 100th Anniversary
  • Coding and documentation
  • Business of Endocrinology
  • Telehealth
  • Physicians Financial News
  • Cybersecurity
  • Cardiovascular Clinical Consult
  • Locum Tenens, brought to you by LocumLife®
  • Weight Management
  • Business of Women's Health
  • Practice Efficiency
  • Finance and Wealth
  • EHRs
  • Remote Patient Monitoring
  • Sponsored Webinars
  • Medical Technology
  • Billing and collections
  • Acute Pain Management
  • Exclusive Content
  • Value-based Care
  • Business of Pediatrics
  • Concierge Medicine 2.0 by Castle Connolly Private Health Partners
  • Practice Growth
  • Concierge Medicine
  • Business of Cardiology
  • Implementing the Topcon Ocular Telehealth Platform
  • Malpractice
  • Influenza
  • Sexual Health
  • Chronic Conditions
  • Technology
  • Legal and Policy
  • Money
  • Opinion
  • Vaccines
  • Practice Management
  • Patient Relations
  • Careers

Study: AI does well with emergency triage

News
Article

AI was correct more than physicians, but it’s not ready to take over

ED triage using AI: ©Sudok1 - stock.adobe.com

ED triage using AI: ©Sudok1 - stock.adobe.com

Emergency departments worldwide face immense pressure, often dealing with overcrowding and limited resources. However, a study led by UC San Francisco suggests that artificial intelligence (AI) could potentially revolutionize the way patients are prioritized for treatment.

Published May 7, 2024, in JAMA Network Open, the study evaluated the efficacy of an AI model in determining the urgency of patient care based on their symptoms. Researchers utilized anonymized data from 251,000 adult emergency department visits, comparing the AI's analysis with patients' scores on the Emergency Severity Index (ESI), a tool used by ED nurses for triage.

The study employed the ChatGPT-4 large language model (LLM) through UCSF's secure generative AI platform, ensuring strict privacy measures. In a sample of 10,000 matched pairs (20,000 patients), where one patient had a serious condition and the other had a less urgent one, the AI accurately identified the more critical patient 89% of the time.

Moreover, in a subset of 500 pairs assessed by both the AI and a physician, the AI proved to be correct 88% of the time, compared to the physician's 86%.

Lead author Christopher Williams, MB, BChir, emphasized the potential impact of AI on prioritizing patient care, envisioning scenarios where AI could assist clinicians facing multiple urgent requests, thereby optimizing resource allocation and potentially saving lives.

However, despite its promising performance, Williams cautioned against immediate implementation of AI in emergency departments. He stressed the need for further validation and clinical trials, highlighting concerns such as bias in AI models. Previous research has indicated that AI models may perpetuate racial and gender biases present in the training data.

Williams emphasized the importance of thoroughly understanding and mitigating biases before deploying AI in clinical settings. He noted ongoing efforts to address these challenges and optimize the responsible use of AI in emergency care.

While the study marks a significant step forward in harnessing AI for emergency triage, Williams underscored the necessity of ensuring that AI not only works but also benefits all patients equitably.

Related Videos
Kyle Zebley headshot
Kyle Zebley headshot
Kyle Zebley headshot
Michael J. Barry, MD
Hadi Chaudhry, President and CEO, CareCloud
Claire Ernst, JD, gives expert advice
Arien Malec