IDANN Triage utilizes interpretable deep neural networks with attention to predict critical patient outcomes and resources, aiding emergency departments with patient prioritization.
Prioritizing patients is a challenge for emergency departments all over the world, as departments need to balance ever-changing patient needs, wait times, and hospital resources. One significant issue is the inability to quickly distinguish significantly ill patients; under-triaging these patients can delay time-sensitive treatment.
Using deep neural networks with attention, our model provides interpretable joint predictions of critical patient outcomes and resource usage, based on the information that nurses already collect during triage. Our REST API is built to interface with electronic health records, allowing nurses to receive a recommended Emergency Severity Index (ESI) rating and view the top factors contributing to the recommendation.
On completely unseen test data, IDANN Triage was able to increase the true positive rate for critical outcomes by 13 percentage points over nurse-determined ESI assignments, without a tradeoff increase in the false positive rate. IDANN captured 37.4% more critical outcomes.