An artificial intelligence tool to help clinicians make decisions regarding hospital patients at risk of sepsis has an interesting feature: it accounts for uncertainty and suggests what demographic data, vital signs, and lab test results are required to improve its prediction performance (1). The SepsisLab system was created based on feedback from doctors and nurses who treat patients in emergency departments and intensive care units (ICUs), where sepsis, the body’s overwhelming response to an infection, is most common. They expressed unhappiness with an existing AI-assisted technology that creates a patient risk prediction score based solely on electronic health records and without clinical input.

Advertisement

SepsisLab: Revolutionizing Sepsis Prediction with Real-Time Data Analysis and Enhanced Accuracy

Scientists at The Ohio State University created SepsisLab to predict a patient’s sepsis risk within four hours; however, while the clock ticks, the system identifies missing patient information, quantifies its importance, and provides clinicians with a visual representation of how specific information will affect the final risk prediction. Experiments using a combination of publicly accessible and proprietary patient data revealed that including 8% of the suggested data increased the system’s sepsis prediction accuracy by 11%.

“The existing model represents a more traditional human-AI competition paradigm, generating numerous annoying false alarms in ICUs and emergency rooms without listening to clinicians,” said Ping Zhang, senior study author and associate professor of computer science and engineering and biomedical informatics at Ohio State.

Advertisement

Integrating AI with Human Expertise: Building a Physician-Controlled Sepsis Prediction System

“The idea is to incorporate AI into every intermediate step of decision-making by implementing the ‘AI-in-the-human-loop’ concept.” We are not only designing a technology; we are also bringing physicians on board. “This is a true collaboration between computer scientists and clinicians to create a human-centered system that puts the physician in control.”

The findings were published on August 24 in KDD ’24: Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, and they will be presented orally on Wednesday (August 28) at SIGKDD 2024 in Barcelona, Spain.

Advertisement

What is Sepsis?

Sepsis is a life-threatening medical emergency that can quickly lead to organ failure, but it’s difficult to identify because the symptoms of fever, low blood pressure, increased heart rate, and breathing issues can mimic a variety of other disorders. This study expands on Zhang and colleagues’ prior machine learning model, which identified the best timing to administer antibiotics to patients with suspected sepsis.

SepsisLab’s Active Sensing System

SepsisLab is designed to provide a fast risk prediction, but it generates a fresh estimate every hour after new patient data is entered into the system.

“When a patient first comes in, there are many missing values, especially for lab tests,” said Changchang Yin, lead author and a Ph.D. student in Zhang’s Artificial Intelligence in Medicine group.

In most AI models, missing data points are compensated for with a single given value – a process known as imputation – “but the imputation model could suffer from uncertainty that can be propagated to the downstream prediction model,” Yin added.

“If the imputation model fails to accurately impute the missing value, which is critical, the variable should be observed. Our active sensing system seeks to identify such missing values and informs doctors of any additional variables that may be required to improve the prediction model’s accuracy.”

How AI and Clinician Actions Improve Sepsis Prediction Accuracy

It is also critical to provide clinicians with practical recommendations in order to gradually remove uncertainty from the system. These include lab tests ranked according to their usefulness in the diagnosis process, as well as estimations of how a patient’s sepsis risk would vary if particular therapeutic therapies were implemented.

Experiments demonstrated that incorporating 8% of the additional data from lab tests, vital signs, and other high-value variables lowered the model’s propagated uncertainty by 70%, amounting to an 11% improvement in sepsis risk accuracy.

“The algorithm can select the most important variables, and the physician’s action reduces uncertainty,” said Zhang, a core faculty member at Ohio State’s Translational Data Analytics Institute. “This fundamental mathematics work is the most important technical innovation – the backbone of the research.”

Zhang believes that human-centered AI will be part of the future of medicine, but only if it interacts with doctors in a way that builds trust in the system.

“This is not about building an AI system that can conquer the world,” he told me. “The center of medicine is hypothesis testing and making minute-by-minute decisions that are not simply ‘yes’ or ‘no.'” We picture a person at the core of the engagement, with AI assisting that person in feeling superhuman.

References:

1. A human-centered AI tool to improve sepsis management
   (https://www.eurekalert.org/news-releases/1055728)

Source-Medindia