While speaking on a panel titled “AI & Sepsis: How Can It Inform Bedside Care?” at the 2021 Sepsis Alliance Summit, Cerner Vice President Tanuj K. Gupta, MD, MBA, referred to estimates that some 60 percent of potentially clinically relevant information in healthcare appears in unstructured data, while others assess that rate to be up to 80%. That estimate conveys much of what we need to know about why the use of natural language processing (NLP) in clinical surveillance tools is the most crucial development for the early identification of and management of patients with sepsis.
Sepsis remains the problem
The human and financial costs of sepsis are well known but cannot be overstated. Sepsis is the biggest in-hospital killer in the United States. According to the Centers for Disease Control and Prevention, at least 1.7 million adults in the U.S. alone develop sepsis each year, and nearly 270,000 die as a result. The cost of sepsis care at inpatient and skilled nursing facilities in the US is over $60 billion annually.
A focus on sepsis and standardizing care has garnered improvements over the last couple of decades, but progress has slowed and, in some cases, stalled entirely. The reasons are almost all tied to the difficulty of identifying patients early and managing the condition. Symptoms for other diseases can mimic sepsis, so multiple data points monitored over time are necessary to diagnose conclusively. Early diagnosis and timely execution of standard sepsis protocols have been shown to improve patient outcomes, but overstretched clinical staff struggle to meet those challenges. Cognitive overload, alert fatigue, outdated workflows, and inadequate training in the treatment of sepsis all contribute.
Clinicians hoped EHRs would ease the burdens. Most EHR vendors now offer sepsis detection solutions, but poor alert accuracy has contributed to increasing levels of alert fatigue for bedside care teams. Many specialized sepsis detection solutions aren’t much better and often the common denominator is that they do not use natural language processing (NLP), a form of artificial intelligence, to inform patient context so they are limited to monitoring structured data. Most sepsis experts believe capturing the information contained in clinical notes will improve the accuracy of sepsis detection algorithms, making them considerably more accurate – highly specific without sacrificing sensitivity.
NLP changes the dynamic for sepsis detection
NLP uses advanced algorithms to extract relevant clinical facts from the clinician’s note, then standardizes that data so that it can be used along with structured data in disease detection surveillance algorithms. Automated clinical surveillance tools monitor and evaluate patients as clinicians do every day – but do so continually for every appropriate hospital patient, accounting for complex interactions, as well as trends over time. This means these tools achieve coverage that far exceeds human capacity, particularly bedside clinicians who are taxed with patient volumes that are driving clinician burnout to unprecedented levels.
Wolters Kluwer’s Sepsis Monitor is one such tool, and it has demonstrated that the incorporation of NLP is a key factor in its ability to deliver early and accurate patient identification alerts. Highly accurate alerts delivered to the right person are key to earn the trust of clinical staff. Sepsis Monitor has shown it can deliver accurate alerts, and it uses NLP to suppress alerts when the right care is already in progress. Through NLP, Sepsis Monitor can recognize when a clinician documents sepsis in a clinical note and instead of generating a sepsis alert the system can track the patient’s compliance with treatment protocols in the background.
In addition, because it was developed by expert clinicians, Sepsis Monitor appropriately identifies co-morbidities, conditions, and even medications documented in a note that can mimic SIRS and will often be misidentified as sepsis in less sophisticated systems. Finally, Sepsis Monitor details the provenance of its alerts including referencing which note contains information used in the sepsis identification algorithm, thus enabling clinicians to quickly evaluate Sepsis Monitor’s alert accuracy rather than guessing at how a black box algorithm generated its recommendations.
Early case studies have demonstrated that Sepsis Monitor is highly specific without any loss in sensitivity and dramatically reduces false alerts. Moreover, a study published in 2020 in the Journal of Patient Safety reviewed six studies that provided evidence of patient monitoring systems reducing sepsis-related mortality; in one study, the risk of death was nearly 50 percent lower.
Addressing burnout: boosting the productivity, morale, and effectiveness of frontline clinicians
Identification, stratification, and proactive management of high-risk patients are critical components of inpatient care. Patients with sepsis top the list of concerns, but absent reliable sepsis surveillance tools, stretched out bedside teams struggle to do these high-priority tasks.
Sepsis surveillance tools with poor alert quality are a distraction to the care team and create frustration. Continually responding to false alerts takes attention from critical patient care activities, reduces trust in monitoring systems, and ultimately may be a part of the reason that more than 30 percent of nurses leave the profession due to burnout.
Yet if nurses have a sepsis surveillance tool they trust, they can work more confidently and efficiently and deliver better patient care. Creating that scenario depends on a surveillance tool that minimizes false alerts and incorporates clinical notes to tell the larger story of the patient. The tool becomes even more helpful if it monitors CMS sepsis bundle compliance and is configured to send the highest status alerts – and only the highest status alerts – to nurses’ hand-held devices within each nurse’s workflow. That is how sepsis care improves while easing the cognitive and emotional burden on nursing staff.
It’s time for the next leap forward in sepsis care
Healthcare workers are demanding more sophistication in decision support tools, and technologies like NLP are making significant advances possible. NLP-driven sepsis surveillance is already demonstrating its ability to drive more precise patient identification, reduce alert fatigue, improve patient outcomes, and improve overall efficiency. Implementing high-quality surveillance tools supported by the appropriate training and change management support can support hospitals and health systems in their goal to improve sepsis care.
Learn how Sentri7 Sepsis Monitor can improve your sepsis care with our white paper, AI-driven Clinical Surveillance Accurately Identifies Patient Risk