The Case for EMR Data in Identifying Early Signs of Patient Deterioration
Because of the high morbidity, mortality, and costs of treatment, the early identification of sepsis has been a major focus of numerous initiatives.1 Foremost among these is the Surviving Sepsis Campaign (SSC), a collaboration of the Society of Critical Care Medicine and the European Society of Intensive Care Medicine in the development of evidence-based guidelines, or “sepsis bundles.” The SSC’s most recent “Hour-One Bundle”, adopted by the Centers for Medicare & Medicaid Services (CMS) as a quality core measure, includes immediate steps to be initiated on the first documentation in the chart of the signs of sepsis, emphasizing the need to treat sepsis as a medical emergency.2
Yet, because the symptoms of sepsis can be attributed to other conditions and comorbidities—and there is no one single diagnostic tool or screen for sepsis—the role of protocols among hospital clinicians and nurses has been controversial.3 In fact, according to recent data on the CMS Hospital Compare quality reporting website, the national average compliance rate for the Severe Sepsis and Septic Shock Early Management Bundle is only slightly over 50 percent.4
In addition, in spite of efforts by the SCC, and other hospital-based sepsis initiatives, there is a significant amount of variability in diagnosing sepsis among those clinicians most closely involved in evaluating and caring for patients at risk.1
So Much Data, So Little Time
Not surprisingly, given the magnitude and deadliness of sepsis, as well as medical advances that have increased the numbers of patients living longer, within the past decade a number of predictive sepsis alert tools, using electronic medical record (EMR) data, have seen widespread growth not just in intensive care units (ICUs) and emergency departments (EDs), but throughout the hospital.5 Yet, since sepsis in its early stages can be so difficult to identify and differentiate from other conditions, clinicians must also be able to confirm their suspicions by sorting through disparate amounts of patient data.5,6
Hence the question arises: What is considered too much noise in the data, or too little data in identifying and treating patient deterioration and sepsis? While numerous studies have shown that predictive models may reduce mortality from severe sepsis, many of the models are based on data from patients who meet the clinical criteria for severe sepsis, such as systemic inflammatory response syndrome or organ dysfunction; the point at which the sepsis has already become life-threatening.5,7 Machine learning systems that can sort through complex and heterogeneous EMR data, alerting clinicians at the onset of symptoms when time is of the essence and the overwhelming deterioration of sepsis can be prevented, is far more challenging and nuanced.7
To address the need for more accurate and timely identification of sepsis signs, some systems have embedded sepsis predictive models in hospital workflows designed to alert nurses or physicians of the need to assess the patient and implement the appropriate steps to improve early detection.5,7 However, recent analyses of embedded models report that in many instances, where the EMR data are inconsistently applied, alert-driven work-flows may not actually flag important changes in a patient’s condition or propel immediate action.5 In fact, in real world settings too many alert systems may result in alert fatigue due to low specificity, leading to false alarms and a waste of resources.5
How Natural Language Processing Can Inform the Conversation
Given the tremendous heterogeneity in EMR data, coupled with high-paced work flow of ICU and ED settings, the need to quickly differentiate the signs of early sepsis from all the EMR “noise” remains a challenge. More recently, the introduction of artificial intelligence natural language processing (NLP) brings a new tool that may aid in a far more meaningful interaction between the clinician and the EMR data. Ideally, the introduction of cognitive agents, or “smart chatbots” could provide the ability to draw from both the clinician’s various communications and the EMR data itself. That combination may then provide information the clinician needs to respond to the patient without delay.
While the cognitive agents would not replace predictive algorithms, they could help monitor the EMR data and notify the clinicians when the data they need to review become available. In this way, the cognitive agent could help provide the data the clinician needs to make the right decision in the moments that matter most. While there is no perfect answer, this could be the start of a more meaningful conversation around patient deterioration and sepsis.
- Diagnosing Sepsis Is Subjective and Highly Variable: A Survey of Intensivists Using Case Vignettes, Critical Care, 2016. https://dash.harvard.edu/bitstream/handle/1/26859971/4822273.pdf?sequence=1
- The Surviving Sepsis Campaign Bundle: 2018 Update, Copyright © 2018 by the Society of Critical Care Medicine and the European Society of Intensive Medicine. http://www.survivingsepsis.org/SiteCollectionDocuments/Surviving-Sepsis-Campaign-Hour-1-Bundle-2018.pdf
- The New “Hour-One” Sepsis Bundle: Key Takeaways and Controversies, Nursing Center, June, 2018. https://www.nursingcenter.com/ncblog/june-2018/hour-one-sepsis-bundle
- Using Remote Surveillance, Interventions to Treat Sepsis Patients Early, Modern Healthcare, 2018. https://www.modernhealthcare.com/article/20181215/TRANSFORMATION02/181219978/using-remote-surveillance-interventions-to-treat-sepsis-patients-early
- To Catch A Killer: Electronic Sepsis Alert Tools Reaching A Fever Pitch? BMJ Quality and Safety, September 2019. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6702042/
- Sepsis As 2 Problems: Identifying Sepsis at Admission and Predicting Onset in The Hospital Using an Electronic Medical Record–Based Acuity Score, Journal of Critical Care, April, 2017. https://www.sciencedirect.com/science/article/pii/S0883944116302775?via%3Dihub
- Validation of Test Performance and Clinical Time Zero for An Electronic Health Record Embedded Severe Sepsis Alert, Applied Clinical Informatics, 2017.https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4941860/
Access related articles, videos, webinars at PRECISELY, by GE Healthcare and Roche.