Researchers from the Massachusetts Institute of Technology in Cambridge and Massachusetts General Hospital in Boston created a predictive modeling system to help inform clinical decisions for sepsis patients in the emergency room.
To create the model, researchers compiled medical records for nearly 186,000 patients treated at Massachusetts General's emergency room between 2014-16. Some patients received vasopressors within the first 48 hours of their hospital visit. Researchers reviewed every medical record for patients with likely septic shock to identify the exact time vasopressors were given.
Researchers used 70 percent of the records to train the machine learning model, which identified more than two dozen clinical factors, such as blood pressure, total fluid volume administered and respiratory rate, present in the cases. The model works by analyzing these clinical features during set time intervals to look for patterns that indicate whether a patient needs vasopressors.
Researchers tested the model using the remaining 30 percent of medical records and found it could correctly predict whether patients would need vasopressors in the next two hours 80 to 90 percent of the time.
"It's important to have good discriminating ability between who needs vasopressors and who doesn't [in the ER]," lead author Varesh Prasad, a PhD student in the Harvard-MIT Program in Health Sciences and Technology at Cambridge, said in a press release. "We can predict within a couple of hours if a patient needs vasopressors. If, in that time, patients got three liters of IV fluid, that might be excessive. If we knew in advance those liters weren't going to help anyway, they could have started on vasopressors earlier."
The machine learning model is the first system specifically designed to inform sepsis treatment in the emergency room. Researchers said they hope to create additional tools to predict sepsis risk in real-time in the ER. "The idea is to integrate all these tools into one pipeline that will help manage care from when [patients] first come into the ER," Mr. Prasad said.
The researchers presented their findings at the American Medical Informatics Association's Annual Symposium in San Francisco this week.