Sepsis is one of the most common reasons for readmission to the hospital and one of the most common causes of death in the ICU. The researchers suggest that most of the ICU patients are admitted through the emergency room.
Treatment typically begins with antibiotics and IV fluids at a couple liters at a time, according to the researchers. Sepsis shock can happen if a patient’s body doesn’t respond well to treatment, which results in blood pressure dropping dangerously low with organ failure. Once that happens, the patient goes to ICU where clinicians can reduce and stop fluids to start vasopressor medications to raise and maintain blood pressure.
However, giving a patient fluids for too long could cause more organ damage. The researchers say that vasopressor intervention could be helpful and has previously been linked to improved mortality in septic shock. But administering vasopressors too early can cause heart arrhythmias and cell damage, leaving clinicians with an unclear answer on when to administer treatment.
MIT researchers have developed a model to alleviate that problem. The model learns from health data on emergency-care sepsis patients and can predict if a patient will need vasopressors within the next few hours.
In a study, the researchers compiled a dataset for ER sepsis patients. When they tested the algorithm, the model was able to predict the need for a vasopressor more than 80% of the time.
The researchers suggest that early prediction could prevent an unnecessary ICU stay for patients that don’t need vasopressors while allowing clinicians to prepare the ICU early for patients who do need vasopressors.
“It’s important to have good discriminating ability between who needs vasopressors and who doesn’t [in the ER],” Varesh Prasad, a researcher on the study, 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 system could be used in a bedside monitor to track patients and send alerts to clinicians in the ER about when to start vasopressors and reduce fluids.
“This model would be a vigilance or surveillance system working in the background,” Thomas Heldt, the study’s co-author, said. “There are many cases of sepsis that [clinicians] clearly understand, or don’t need any support with. The patients might be so sick at initial presentation that the physicians know exactly what to do. But there’s also a ‘gray zone,’ where these kinds of tools become very important.”
Other models have been built to predict who is at risk of developing sepsis or when to administer vasopressors in the ICU. The MIT-developed model is the first one to be trained on data from the ER.
“[The ICU] is a later stage for most sepsis patients. The ER is the first point of patient contact, where you can make important decisions that can make a difference in outcome,” Heldt said.
One of the challenges of building the machine-learning system has been the lack of an ER database. The MIT researchers worked with Massachusetts General Hospital over several years to gather medical records of about 186,000 patients who were treated in the ER from 2014 to 2016. Some of the patients had vasopressors in the first 48 hours of their hospital visit. Two of the researchers reviewed all of the data from patients who might have had septic shock manually to figure out the exact time vasopressors were administered.
About 70% of the records were used for training the model and the remaining 30% were used for testing it. The model extracted up to 28 of 58 possible features from patients who needed or didn’t need vasopressors. These features included blood pressure, elapsed time from initial ER admission, total fluid volume administered, respiratory rate, mental status, oxygen saturation and changes in cardiac stroke volume.
When the researchers tested the model, the model could analyze the features in a new patient at set time intervals and could look for patterns that could indicate that a patient would need vasopressors or not. using that information, the system made a prediction at each time interval and was accurate 80-90% of the time.
“The model basically takes a set of current vital signs, and a little bit of what the trajectory looks like, and determines that this current observation suggests this patient might need vasopressors, or this set of variables suggests this patient would not need them,” Prasad said.
The researchers hope to expand their work to make more tools that can predict in real-time if patients in the ER would initially be at risk for sepsis or septic shock.
“The idea is to integrate all these tools into one pipeline that will help manage care from when they first come into the ER,” said Prasad.
The researchers also say that the system could help clinicians in emergency room departments in major hospitals focus on patients who are most at-risk of developing sepsis.
“The problem with sepsis is the presentation of the patient often belies the seriousness of the underlying disease process,” Heldt said. “If someone comes in with weakness and doesn’t feel right, a little bit of fluids may often do the trick. But, in some cases, they have underlying sepsis and can deteriorate very quickly. We want to be able to tell which patients have become better and which are on a critical path if left untreated.”
The research was funded in part by a National Defense Science and Engineering Graduate Fellowship, the MIT-MGH Strategic Partnership and by CRICO Risk Management Foundation and Nihon Kohden Corporation.