More than 1.5 million Americans develop sepsis every year with about 250,000 people dying from the disease, according to the Centers for Disease Control and Prevention. Sepsis is the body’s overwhelming response to an infection.
The disease is hard to detect as many of the symptoms, such as fever and difficulty breathing, mimic other conditions. However, if not treated quickly it can lead to tissue damage, organ failure and even death. Research indicates that for every hour a patient goes untreated for sepsis, their chance for mortality increases 8 percent.
OSF HealthCare enlisted the Advanced Analytics team, a part of OSF Innovation, to help identify individuals with the disease sooner. The goal was to predict the probability of a patient having sepsis soon after they arrive to the emergency department.
Pre- and Post-Lab sepsis models
Understanding the difficulty of diagnosing sepsis, the Advanced Analytics team developed and deployed a model in late 2017 that enhances a clinician’s decision on whether to treat someone for the disease. Nurses and physicians were previously alerted that a patient might have sepsis via the electronic medical record (EMR) based on systemic inflammatory response syndrome (SIRS) criteria that takes into account a person’s vitals and lab results. However, the alert was often slow and inaccurate, leading to false-positives.
Advanced Analytics developed an in-house, early warning model that is designed to alert clinicians before labs are drawn. This Pre-Lab model pulls a patient’s vitals and historical data to estimate the likelihood of a patient having sepsis.
“We want our model to make a prediction while a patient is being triaged in the ED—or within 15 minutes of arriving to the hospital,” said Jason Weinberg, a data scientist on the Advanced Analytics team. “As soon as a nurse or advanced practice provider takes an individual’s vitals and hits save in the EMR, all patient data gets collected, runs through the model and then a probability score pops up, helping clinicians decide whether to treat an individual for sepsis.”
If there’s just a 15% chance a patient has sepsis, clinicians are expected to begin the treatment process that must take place within three hours of a diagnosis. Those found to have a 15% chance of sepsis are up to eight times the standard population’s risk of getting the disease.
“Following implementation of the Pre-Lab sepsis model, we are catching more people with sepsis in a timelier fashion,” said Chris Franciskovich, director of Advanced Analytics. “We have also been able to reduce the number of false positive alerts that have come up with previous models.”
It’s estimated the Pre-Lab model is saving OSF HealthCare $508,000 in workload reduction as clinicians are no longer having to respond to as many false-positive alerts. If the Pre-Lab model doesn’t identify a patient at risk, there’s a chance the Post-Lab model could do so. That model uses lab results to determine the probability of a patient having sepsis.
While the Pre- and Post-Lab models were built to help clinicians track patients with sepsis entering the ED, the Advanced Analytics team is also working to determine whether those same models can be applied in the inpatient setting.
The group is also working on a couple of studies around sepsis treatment to find any gaps in care. The first one compares the outcomes of septic patients who are transferred versus those who are treated at the same hospital they arrived in. The second takes a look at whether medication treatment is delayed for patients waiting for an inpatient bed.
Advanced Analytics isn’t the only team working to improve sepsis outcomes. A multidisciplinary group, made up of leaders and clinicians across the Ministry, came together to design and standardize how clinicians respond to sepsis alerts, bringing clinical teams to the bedside sooner to care for patients.