TL/DR –
US healthcare often uses a triage process to assign queuing priority, and recently AI has been increasingly used for this task. Professors Itai Gurvich and Jan Van Mieghem from the Kellogg School, in collaboration with Simrita Singh, compared two AI triage methods: one that uses a preliminary diagnosis to assign priority and another that directly assigns priority based on raw data from the patient’s x-ray. The researchers found that the direct model was more effective at reducing wait times and costs, and was generally superior unless the AI was able to provide a 100% accurate diagnosis.
U.S. Healthcare System Turns to AI for Patient Triage
The U.S. healthcare system typically uses a triage process where clinicians assign patients a preliminary diagnosis. The aim is to treat the most urgent cases first and also reduce overall wait times and costs. Recently, hospitals have been increasingly adopting AI for patient triage to expedite this process.
A Study Comparing AI Triage Methods
Itai Gurvich and Jan Van Mieghem, both professors of operations at the Kellogg School, and Simrita Singh, an assistant professor at Santa Clara University conducted a study on AI for patient triage. They looked at two AI methods: one that diagnoses a patient’s “type” from the x-ray then assigns priority, and another that decides patient priority based directly on the raw x-ray features.
Direct-to-Queue Method Proves More Effective
The researchers found that the “direct model,” or the second method, was more effective at reducing patient wait times. This approach bypassed an initial diagnosis when triaging patients and significantly reduced the total waiting cost. “Our objective is not necessarily accuracy [of diagnosis] but rather reducing the cost of making the wrong patient wait too long,” Van Mieghem says.
Head-to-Head Comparison of AI Triage Methods
The researchers compared the direct-to-queue method with the typical diagnosis-first approach using a mathematical model. They discovered the two methods performed equally well only if the AI accurately diagnosed a patient’s disease type 100% of the time, a highly unlikely scenario. In all other situations, the direct-to-queue method proved more effective at minimizing wait times.
Reducing Waiting Costs with AI Triage
Testing their theory on a dataset of 112,120 anonymized patient chest x-rays, the researchers used an AI image-classification system called MobileNet to analyze and sort the x-rays into urgency-based queues. The direct-to-queue method outperformed the diagnosis-first model, reducing total waiting costs by over 30%.
Implications for Everyday Queues
The findings could be applied to other scenarios involving queues, such as customer service at help centers or bank call centers. For AI engineers, the study emphasizes the importance of focusing on an AI model’s core objective. In the case of chest x-rays, it was to reduce the total cost of waiting, not necessarily to achieve an accurate preliminary diagnosis.
—
Read More Health & Wellness News ; US News