New AI Tool Predicts Sepsis Risk Within 4 Hours, Identifies Important Patient Data

Ping Zhang

Credit: The Ohio State University, College of Engineering

Investigators proposed an active sensing algorithm, SepsisLab, to assist clinicians in decision-making for hospital patients at risk for sepsis by recommending observing demographic data, vital signs, and lab test results.¹

Doctors and nurses who treat patients in the emergency departments and ICUs with sepsis have reported dissatisfaction with the existing AI-assisted tool used to generate a patient’s sepsis risk. The current AI tool is based only on electronic health records and no inputted data from clinicians.

“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 senior study author Ping Zhang, associate professor of computer science and engineering and biomedical informatics at Ohio State, in a press release.²

Ohio State University investigators sought to investigate and reduce the uncertainty caused by missing values in AI models predicting the sepsis risk.¹ In this study, investigators designed SepsisLab to predict a patient’s sepsis within 4 hours. The algorithm is supposed to identify missing patient information, and from there determine how important the missing information is and provide a visual image to clinicians of how that specific information will impact the final sepsis risk prediction.

SepsisLab builds on an earlier machine learning model developed by Zhang and colleagues that estimated the optimal time to give antibiotics to patients with suspected sepsis.² However, with Sepsis lab, this AI tool is designed to predict the sepsis risk quickly, producing a new prediction every hour after new patient data has been added to the system.

Research has shown adding 8% of publicly available and proprietary patient data, such as new data from lab tests, vital signs, and other high-value variables, into the algorithm improves the tool’s sepsis prediction accuracy by 11%. Adding this type of data reduces uncertainty in the model by 70%.

“The idea is we need to involve AI in every intermediate step of decision-making by adopting the ‘AI-in-the-human-loop’ concept,” Zhang said. “We’re not just developing a tool – we also recruited physicians into the project. This is a real collaboration between computer scientists and clinicians to develop a human-centered system that puts the physician in the driver’s seat.”

Sepsis is a life-threatening medical emergency, contributing to half of all hospital deaths. The body’s inability to fight an infection can rapidly lead to organ failure. Despite sepsis’ prevalence, it is difficult to diagnose. People with sepsis may experience symptoms of fever, low blood pressure, increasing heart rate, and breathing conditions—symptoms that can appear like other conditions.

Typically, AI models account for missing data by putting in a single assigned value—an imputation. However, this leaves room for uncertainty.

“If the imputation model cannot accurately impute the missing value and it’s a very important value, the variable should be observed,” said first author Changchang Yin, a computer science and engineering PhD student in Zhang’s Artificial Intelligence in Medicine lab, in the press release. “Our active sensing algorithm aims to find such missing values and tell clinicians what additional variables they might need to observe – variables that can make the prediction model more accurate.”

Ultimately, the AI will take all the inputted data on lab tests, vital signs, and other high-value variables and rank them in order based on the value they add to the diagnostic process. The system will also estimate how a patient’s sepsis risk will change after specific clinical treatments.

“The algorithm can select the most important variables, and the physician’s action reduces the uncertainty,” Zhang said. “This fundamental mathematics work is the most important technical innovation – the backbone of the research.”

References

1. ^{Yin, C, Chen, P, Yao, B. SepsisLab: Early Sepsis Prediction with Uncertainty Quantification and Active Sensing. 2024. ASM. https://doi.org/10.1145/3637528.3671586}
2. ^{A Human-Centered AI Tool To Improve Sepsis Management. EurekAlert! August 27, 2024. https://www.eurekalert.org/news-releases/1055728. Accessed August 28, 2024.}