Researchers claim a new machine learning model that predicts the best time to treat sepsis could open the door for support systems that assist doctors in personalising treatment choices while treating patients at the bedside.
Researchers from The Ohio State University describe the new model, which makes use of artificial intelligence to tackle the challenging problem of when to provide antibiotics to patients with a suspected case of sepsis, in a report that was published in Nature Machine Intelligence.
Time is of the essence since organ failure can occur very quickly as a result of sepsis, the body’s excessive reaction to an infection. Yet its signs—fever, low blood pressure, accelerated heart rate, and breathing issues—can resemble those of numerous other illnesses. As a first line of defense, federal guidelines recommend immediate treatment with broad-spectrum antibiotics. This course of action is generally necessary before a lab can provide cultures confirming a bacterial infection.
These uncertainty and time constraints were taken into consideration when creating the model.
Researchers evaluated the model’s performance by comparing the outcomes of patients whose actual treatment followed the model’s suggested timeline for treatment with those of patients whose actual treatment deviated from what the model would have suggested based on their vital signs, lab results, and risk-related demographic data. Patient survival 30 and 60 days following sepsis treatment served as the outcome’s proxy.
“We demonstrated that our mortality rate is lower when artificial intelligence and actual treatment are in sync. According to senior author Ping Zhang, Ph.D., assistant professor of computer science and engineering, as well as biomedical informatics at Ohio State, “the mortality rate can be as high as 25% if they don’t agree.”
On a dataset taken from the MIMIC-III database, which is a freely accessible resource, the model was trained and verified. A new external dataset from AmsterdamUMCdb and several MIMIC-III subsets were used to test the model.
Changes in patient vital signs and lab test results over time, which serve as indicators of illness severity and type of infection, as well as a novel method developed to compare outcomes between patients who received and did not receive antibiotics at a particular time, were among the key measures from almost 14,000 sepsis patients.
“We want the modeling to predict whether it’s beneficial to use antibiotics at a given time—yes or no. But we’ll never know what happens if we don’t give the antibiotic. So we applied a clinical trial concept to this model: For every patient who had taken the drug, we included a matched, clinically similar patient who didn’t take antibiotics at that time,” said Zhang, who leads the Artificial Intelligence in Medicine Lab and is also a core faculty member in Ohio State’s Translational Data Analytics Institute.
“That way, we can predict the counterfactual outcome, and train the counterfactual treatment model to find whether treatment for sepsis works or not.”
According to Katherine Buck, MD, assistant professor of emergency medicine in the College of Medicine and director of the Geriatric Emergency Department at Ohio State Wexner Medical Center, sepsis accounts for more than one-third of in-hospital deaths and is most frequently seen in intensive care units and emergency departments, “where we’re often making decisions without the gold standard—results from a culture.” Not all patients who fulfil sepsis criteria go on to show evidence of a bacterial infection, according to experts.
Antibiotics carry hazards; they may be harmful to the kidneys, produce an allergic reaction, or result in C. difficile infection, which causes severe diarrhoea and colon inflammation.
“What this paper starts to get at is, can we use information available to the clinicians, sometimes at the forefront and sometimes not, to say, Things are changing in a way that suggests the patient will benefit from antibiotics,” Buck said. “A decision-support tool could tell clinicians if it matches what we’re already thinking or prompt us to ask ourselves what we’re missing. Hopefully, with time, all the electronic health record data we have will reveal signals—and from there it’s a matter of figuring out how to use them and how to get that to clinicians.”
These insights—as well as the availability of data from electronic health records—were crucial for supplying the model with the appropriate data and configuring it to take into account various factors that are associated with changing medical circumstances, according to Zhang.
“We modeled the patient record like it’s language,” he said. “And for machine learning, we always train the model batch by batch—you need the model to analyze the pattern of data, set parameters, and based on these parameters, add another training dataset to make improvements. And then the machine always finds better parameters to fit the model.”
The Sequential Organ Failure Assessment (SOFA) score, which is used to routinely evaluate how an ICU patient’s organ systems are functioning based on findings from six lab tests, is a critical indicator used to guide how the model arrives at a suggestion. The SOFA scores vary when the model updates the suggested treatment timeline based on changes to personalised patient data, as seen in example case studies the researchers did to highlight what an interface designed for the clinical situation may look like.
“Our paper is the first to use AI to pursue an antibiotic recommendation for sepsis, using real-world data to help clinical decision making,” Zhang said. “Any research like this needs clinical validation—this is phase one for retrospective data analysis, and phase two will involve human-AI collaboration for better patient care.”
