AI could assist clinicians in making complex and quick decisions, especially in intensive care units, according to a study by researchers from Carnegie Mellon University's Human-Computer Interaction Institute (HCII), the University of Pittsburgh, and UPMC.
Clinical Decision Support
The team introduced an AI Clinician model, an interactive clinical decision support (CDS) interface named the AI Clinician Explorer, that can recommend treatments for sepsis.
Sepsis is a potentially life-threatening condition that occurs when the body's response to an infection damages its own tissues and organs.
It can develop when the chemicals the immune system releases into the bloodstream to fight an infection cause inflammation throughout the entire body instead.
During their ICU stays, more than 18,000 patients who met the standard diagnostic criteria for sepsis were the basis for the training of the model.
Clinical experts can now filter and search through the dataset, visualize the trajectory of the patient's illnesses, and compare the predictions made by the model to the actual treatment decisions that were made at the bedside.
Ignore, Rely, Consider, Negotiate
In a think-aloud study, 24 ICU physicians experienced in treating sepsis used a simplified AI Clinician Explorer interface to make treatment decisions for four simulated patient cases.
The team categorized the physicians into four groups based on their behavior: ignore, rely, consider, and negotiate.
The "ignore" group made their decisions without considering the AI's recommendation, while the "rely" group accepted the AI's input consistently.
The "consider" group evaluated the AI's recommendation before accepting or rejecting it. The majority of participants belonged to the "negotiate" group, which accepted some aspects of the AI's recommendation but not all.
The team discovered that most clinicians incorporated the AI Clinician model into some of their decisions. Venkatesh Sivaraman, a Ph.D. student in the HCII and part of the research team, believes that clinicians are excited about AI's potential to help them but may not be familiar with how these AI tools work.
Read Also : AI May Just Prevent the Next Pandemic-but How?
AI in the ICU soon?
The goal of the research is not to replace or copy the decisions made by clinicians. Instead, it aims to uncover patterns in patient outcomes that may have been overlooked in the past using AI.
The team claims the system could provide guidance to clinicians in a new direction or support their current approach.
However, some clinicians raised concerns about the AI's limited access to holistic patient data, such as general appearance. They were also skeptical when the AI suggested different approaches than what they had been taught.
"When the CDS deviates from what clinicians would normally do or consider to be best practice, there was not a good sense of why," Sivaraman said in a press release statement.
"So right now, we're focusing on determining how to provide that data and validate these recommendations, which is a challenging problem that will require machine learning and AI."
The team's findings were published in arXiv.
