Sepsis is the body’s life-threatening response to infection that affects 1.7 million people and takes 350,000 adult lives in the U.S. every year.
Sepsis is a medical emergency that requires rapid diagnosis and treatment. For every hour sepsis goes untreated, mortality increases by about 8%. Nurses play an integral role in the fight against sepsis. According to the U.S. Bureau of Labor Statistics registered nurses held over 3 million jobs in 2021. Sixty percent were in a hospital setting which accounted for approximately 30% of total hospital employment. Nurses in these settings care for patients 24x7 and are often the first professionals to recognize adverse changes in a patient’s condition.
Sepsis surveillance at the point of care
Sepsis surveillance technology allows for early and accurate detection of sepsis, which is critical in reducing mortality and overall healthcare-associated costs. The goal is to identify patients at risk before they decompensate by continuously aggregating and analyzing data documented within the electronic health record (EHR). This allows for timely and accurate alerting at the bedside. Workflow integration, sophisticated algorithms to identify patients with sepsis, and the use of AI technology like natural language processing (NLP), should be considered when reviewing sepsis surveillance technologies.
The role of clinical NLP in sepsis management
Natural language processing (NLP), a type of artificial intelligence (AI), is an important tool gaining attention and increased usage in the fight against sepsis. Along with algorithms, rules, and machine learning (ML), NLP is a form of AI that can be used to augment a clinician’s critical thinking skills and thus is referred to as, “Augmented Intelligence,” rather than “Artificial Intelligence.” Clinical NLP assists in the tagging and identification of critical clinical indicators within the EHR that could otherwise be overlooked in the sea of information available in the record. Presenting the right information to the right team members at the right time is critical in managing patients who may be decompensating or experiencing signs and symptoms of sepsis.
However, this is not an easy task in healthcare. Let’s explore some of the reasons that clinical NLP in healthcare differs from other ecosystems, and why it is so important to tailor NLP to your specific use case.
Clinical documentation and records
Clinical notes come in various formats and include structured and unstructured data types. Documentation is multidisciplinary and can be entered using methods such as point-and-click templates, free-text type, speech recognition, and dictation. Notes can be entered directly into an EHR, with outside documents often getting scanned into the system. Since most clinical notes contain unstructured or semi-structured data, it is imperative to implement a solution that is clinically trained to instantly extract clinically relevant information from these unstructured notes, and in real-time present vital information such as, patient conditions, medications, labs, vital signs, and other value-add clinical concepts to the clinician.
Three C’s for clinical NLP: content, codes and context
Content
Clinical NLP goes a step beyond traditional NLP because it is carefully tuned to be able to understand the rich clinical context of medical terminology and documentation that is critical to quality patient care. Think of the many unique concepts, terms, abbreviations, acronyms, and phrases clinicians use to document patient assessments, problems, diagnoses, interventions, goals, and outcomes – patient charts are full of them. Examples commonly seen in the EHR include “PNA” (pneumonia), “DIC” (disseminated intravascular coagulation), and septic lung emboli. Traditional, non-medical NLP solutions are unable to identify these unique nuances, and therefore miss the mark in not only identifying patients decompensating or experiencing indicators of sepsis, but also showing early signs of other disease states that need to be treated urgently. That said, it is so important to leverage a clinical NLP solution that can properly identify not only clinical concepts, but also have the intelligence to recognize and understand the unique ways a clinician can document using synonyms, acronyms, and abbreviations that are commonly found throughout the patient record.
Codes
Along with concept identification, your clinical NLP solution must have the ability to codify data to standards to enable semantic interoperability to empower and inform continuity of care, analytics, and other downstream initiatives. Examples of standardized terminologies used to normalize data unique to healthcare are, SNOMED-CT for problems, ICD-10-CM for diagnoses, RxNorm for medications, LOINC for observations, and CPT, HCPCS, and ICD-CM-PCS for procedures.
Context
Context — one of the most important features in a clinical NLP solution — includes being able to identify for the clinician at the point-of-care where in the record the critical data is pulled from. The surrounding notes often provide relevant information that help assist with medical decision making. In order to extrapolate key patient details from the documentation, it is important to extract from essential sections within the note, reducing the risk of noise, thus providing more accurate identification of a patient’s condition.
Along with sectioning, organizations must ensure that a solution not only identify valuable clinical data, but also present these data based on their context. For example, is the note entry referring to the patient, or someone other than the patient? If in the Family History section, the note reads: “Mother-Breast cancer, Father-CHF”, the presented data should not attribute breast cancer or CHF to the patient. Negation is another extremely important contextual element to consider. Can the solution distinguish if the concept or indicator is affirmative or negated? For example, does the documentation state, “The patient complains of shortness of breath” or does it state, “The patient denies shortness of breath”? Temporality (present, historical) and certainty (certain, uncertain) are additional context related attributes to consider when looking for a solution.
Tuning for success – reduce the noise
Once a system has been implemented, it is essential that organizations work closely with the solution provider to continuously improve the system. In the case of sepsis surveillance, information needed to be brought forward to the attention of the clinician caring for the patient must identify the clinical indicators for that condition. Organizations can identify which indicators are involved in false or non-value added alerts. Reducing noise will not only help ensure NLP is doing its job in augmenting a clinician’s critical thinking skills, but will also help to reduce alert fatigue and clinician burnout.
Tuning is a multidisciplinary action
As organizations review evidence-based guidelines, regulations, and other guiding references that best address their use cases, it’s essential to involve the users of the system, as well as clinical terminologists and informaticists who can help tweak the system to meet the needs of those using the solution. This will help ensure that the clinical NLP solution’s output is standardized and codified to match the specified use case, such as SNOMED-CT for problems in the case of sepsis management. Further, a multidisciplinary approach ensures that terminology libraries remain up-to date, facilitating accurate responses.
Ensuring that an organization gets the most out of a clinical NLP solution requires a methodical approach to selection and implementation. Above all, this process must be clinician-friendly, with nursing and other clinical professionals involved at all stages of the selection process to ensure that the technology will be efficient and accurate while complementing existing clinical workflows. Taking these initial steps will increase the likelihood of adoption and use, enabling the clinical NLP solution to generate meaningful ROI while improving clinical outcomes.
Having the ability to identify provider documentation related to sepsis is important for healthcare facilities like Blessing Health System. Providers rely on clinical note documentation, which incorporates key elements of care and patient conditions. A collaborative group of clinicians and informaticists, including those with a nursing background, review documentation to ensure key concepts that might indicate sepsis are identified prior to implementation. This collaborative team approach ensures NLP is accurate and timely.
The impact of sepsis surveillance on nursing
The implementation of sepsis surveillance tools have a profound impact on bedside nursing. With the global pandemic, nurses all over the world have faced a myriad of obstacles when caring for patients at the bedside. Staffing challenges, increased patient ratios, new protocols, and changes to the way education is received have all had a dramatic impact. While there is no easy fix to some of these obstacles that many still face today, there are technological solutions that can help the bedside nurse at the point of care.
Solutions, such as Sentri7® Sepsis Monitor allow care teams to continually monitor patients for sepsis, allowing nurses to focus on providing bedside care to their patients. Nurses can rest easy knowing that while they are caring for another patient, the system is still monitoring other patients and will send alerts to the bedside nurse, charge nurse, and even the physician when necessary. These sepsis surveillance solutions are not meant to replace clinical judgment but augments nursing's clinical expertise. This empowers nurses to remain focused on patients, and allows more time for bedside care.
Learn how clinical surveillance solutions like Sentri7 Sepsis Monitor and clinical NLP solutions like Health Language can support bedside ICU nursing care teams.
