Sepsis accounts for half of hospital deaths, according to research published in the Journal of the American Medical Association. It’s the leading cause of readmissions, and more than $20 billion is spent on it annually.
According to the Centers for Disease Control and Prevention, sepsis has become a medical emergency. Every year there are at least 1.7 million cases of sepsis implicated in as many as 270,000 deaths, the CDC says.
Sepsis starts with any kind of infection or any time germs enter the bloodstream. The human body tries to fight the infection but fails. The organs shut down one by one. But what’s unique about sepsis, compared to cancer, for instance, is that there is a known cure – usually antibiotics and fluids are sufficient. This means, that for the medical community, the most critical issue with regard to sepsis is detection, not treatment.
More than two-thirds of the providers recently polled by KLAS, say surveillance technology employed for sepsis detection has led to a significant improvement in patient safety outcomes, with some reporting a 50 percent drop in mortality. Other benefits are shorter lengths of stay, fewer readmissions, and lower treatment costs.
An essential best practice has been the preparation of an order set for each detection of sepsis, so that patients, once identified, can start treatment immediately. This is critical. Studies reveal that until antimicrobials are administered, the chance of survival drops 7.6 percent each hour.
Sepsis is a rapidly progressing disease. As it advances, it brings higher costs and worse outcomes. A major challenge in the early identification of sepsis is that the symptoms associated with sepsis are the same as many other conditions and diseases. This makes it particularly difficult to distinguish septic patients from other high-acuity patients. For the purposes of clinical identification, most sepsis detection solutions must determine whether the patient meets the criteria of systemic inflammatory response syndrome (SIRS).
Clinical elements of SIRS include: temp >38 degrees C (100.4 degrees F) or <36 degrees C (97.8 degrees F); heart rate >90; respiration rate >20 or PaCO2 <32 mm Hg; WBC >12,000/mm3, <4,000/mm3, or >10 percent bands.
Other detection algorithms in use include q SOFA (quick Sequential sepsis-related Organ Failure Assessment), Modified Early Warning Score (MEWS), Cerner’s St. John Sepsis Agent, and the Rothman Index (RI). The latter is often viewed as superior, because it includes all the clinical elements used in both SIRS and qSOFA, additional vital signs and labs, and a full range of body-system nursing assessments known to be leading indicators of deterioration.
Paramount to a successful sepsis detection solution is performance with regard to sensitivity and specificity. Sensitivity is the ability of a test to recognize true positives, while specificity measures the number of true negatives correctly identified. In a screening test for a potentially life-threatening disease such as severe sepsis, high sensitivity would be valued over high specificity.
The majority of healthcare delivery organizations look to leverage sepsis detection capabilities native to their EMR. Depending on the vendor, sepsis detection manifests itself within the EMR through specific modules, frameworks or tools, notification and alerting, surveillance dashboards, and in some cases, custom solutions through extensibility offered by the vendor. In some cases, users report the need to adopt in-house tools for their workflows. In addition, EMR users report alert fatigue with the system, potentially due to sensitivity issues.
In interviews we regularly conduct with Chief Information Officers at health delivery organizations, we find the following comment—from Brian Patty, vice president and CMIO of Clinical Information Systems at Rush University Medical Center—representative of sentiment.
"The issue with sepsis is it’s a sensitivity and specificity challenge. You can set up an alert that monitors blood pressure, heart rate, temperature and things like that, but the sensitivity can be very poor, so alerts are firing way too often. As much as we tried to refine alerts on our own, we were still only having a sensitivity and specificity with our sepsis alerts in our EMR in the upper 60s. So, as you can imagine, a little over 30% of the time, when there is sepsis, the alerts are not picking it up. We need to reach at least the low 90s, both sensitivity and specificity, so alert fatigue is reduced and you’re picking up more sepsis cases earlier.
"For decades, the medical community has struggled to improve sepsis outcomes, in part because the condition’s initial symptoms mimic several common illnesses, but also because today’s EHRs cannot support the enterprise-level surveillance required to impact sepsis rates. We must overcome the challenges inherent in early identification so we can achieve timelier and more accurate diagnoses for faster and more effective treatments. This will, in turn, reduce the severity of sepsis and the number of associated deaths."
According to a study published in the Public Library of Science, (“Creating an Automated Trigger for Sepsis Clinical Decision Support at Emergency Department Triage Using Machine Learning,”) free text is significantly more accurate than structured data such as vital signs and demographic information in identifying infection.
Another study, “Effect of a Machine Learning-Based Severe Sepsis Prediction Algorithm on Patient Survival and Hospital Length of Stay,” published in the British Medical Journal, demonstrated average length of stay decreased from 13.0 days in the control to 10.3 days in the experimental group, with in-hospital mortality decreasing 12.4 percentage points when using a machine learning algorithm. This study was the first randomized, controlled trial of a sepsis surveillance system to demonstrate statistically significant differences in length of stay and in-hospital mortality.
Accurate triggering of clinical decision support will become increasingly important as clinical decision support is integrated into EMRs. Since decision support has the potential to interrupt the clinical workflow, every attempt should be made to ensure all eligible patients receive decision support (sensitivity), and that non-eligible patients are not mistakenly targeted (specificity) thus leading to alert fatigue.