Picture this common critical care setting scenario: Your critically ill sepsis patient needs immediate rapid and effective intervention. Your early diagnosis and prompt initiation of treatment, especially with antimicrobials and fluid resuscitation, are linked to better outcomes. With a possible 40% mortality rate, this is a life or death situation, and you are not about to let your patient be the one in three to die within the first 48 hours of admission to the intensive care unit (ICU). There is nothing new about this situation; you know that sepsis is responsible for more than a third of all hospital admissions and approximately 50% of all ICU admissions. Scores such as SOFA were designed to specifically assess the severity of organ dysfunction in septic patients but early warning scores and predictive models that assess general disease severity on admission allow for early prediction of potential adverse outcomes, ensuring your patients are treated safely, effectively, and in time.
Early warning systems
These systems were developed as far back as the 1980s in response to the recognition that a significant proportion of patients who suffered adverse events, such as cardiac arrest or an unplanned ICU admission, had abnormal physiological parameters for many hours before the event. It therefore made sense that linking regular physiological observations to thresholds or triggers for seeking assistance, coupled with the development of rapid response teams (RRTs), could prevent deterioration. Indeed, these “track-and-trigger” systems, where an increasing score determines an escalated response, are able to identify patients at risk of deterioration in a busy clinical environment. Such early warning systems provide a simple, uniform method for categorizing a patient’s condition, and guidance to indicate when a deteriorating patient may require additional attention. They are also used to determine in-hospital mortality and ICU transfer.
Scoring systems used in critically ill patients can be broadly divided into those that are specific for an organ or disease and those that are generic for all ICU patients. A proliferation of early warning systems has been developed, including NEWS, MEWS, SAPS, MPM, and APACHE. They use scores for single parameter criteria, multiple parameters, or aggregate weighted parameters with no single superior scoring system. Physiological parameters (e.g., heart rate, systolic blood pressure) are measured and a numerical weighting applied according to the degree of deviation from a “normal range”. The numbers are then totaled to create a score that predicts clinical outcome. Some scores use data obtained during the first 24 hours after admission to the ICU, while others use data that is continuously collected to track patients throughout their illness pathway to trigger at predefined thresholds of aggregate scores. 1,2,3,4,5
The limitations of early warning systems
As with any other clinical tool, early warning systems may be subject to failure and their accuracy is dependent on the quality of the data input. For example, if physiologic data is incorrectly measured, this can lead to false alarms and increased workload. These scoring systems were developed using static population data but the ICU population is dynamic, resulting in inaccurate predictions. Most early warning systems were developed by expert opinion. NEWS was originally only validated in a single center UK hospital cohort, while APACHE was developed in the U.S., and SAPS used data from ICUs across the world and is commonly used in Europe. 5 While a high sensitivity trigger is likely to lead to earlier identification of patients, with the potential to provide early treatment and hence improve outcome, this may create trigger fatigue, overwork and distraction. Furthermore, the blind application of a score of deterioration to all patients may not be appropriate. Indeed, there may be a danger of over treatment.
General illness severity scores are widely used in the ICU to assess resource use, predict outcome, and characterize disease severity and degree of organ dysfunction. The different scoring systems should be seen as complementing each other, rather than competing with one another as they have different purposes and measure different parameters. For example, outcome prediction models cannot be used to assess the severity of individual organ dysfunctions or to monitor patient progress over time. Although organ dysfunction scores correlate with outcomes, this is not what they were developed for. It is envisoned that, increasingly, all patients will be initially evaluated using a general outcome prediction model computed on admission or within the first 24 hours, then by repeated organ failure (for example, SOFA) and nursing workload (for example, TISS-28) scores during their ICU stay.5
Early warning systems perform best on similar patient populations or data sets that were used to develop the score and therefore may not cope well with other patient populations or chronically abnormal physiology. As ICU populations change and new diagnostic, therapeutic and prognostic techniques become available, all the scoring systems will need to be updated.
An optimal usage for an early warning system is to provide a high sensitivity trigger to alert a rapid response team (RRT) who have specific skills in identification and treatment of sepsis, and who subsequently use more refined tools (i.e., SOFA) to guide management decisions. Ultimately, experienced clinical judgment must always be pre-eminent. Nevertheless, early warning systems are important in the care of acutely ill patients and are used across the world.
The added value of predictive models and how they benefit from the huge amount of data in EMRs
Large investments by the U.S. government and the Meaningful Use Act have accelerated the adoption of Electronic Medical Records (EMR). These information systems contain real-time granular data made over the course of a patient’s stay in the hospital, including detailed records of symptoms, test measurements, data from monitoring devices, clinicians’ observations and billing data. Most current ICU mortality modeling methods use a small fraction of the data available on a patient, primarily the single most abnormal value of laboratory test results and vital signs. Developed before EMRs were widely adopted, these models relied on manual data abstraction and thus had a compelling rationale to limit the data collected. For example, a manual APACHE medical record review by a trained nurse takes an average of 30 minutes per patient. The increasing adoption of EMRs means this process can be automated and include more values and variables.
This data is now being used in machine learning algorithms to better and more accurately predict outcomes and specific adverse events. Early predictions allow for interventions that can possibly prevent these events, improve outcomes and determine more cost-effective allocations of resources. A recent retrospective multi-center study by Marafino and colleagues of 101,196 patients with a first-time admission to the ICU found that capturing more of the available data and applying machine learning and natural language processing from doctors’ notes can improve and automate the prediction of outcomes in ICU patients.
The CLEW platform
The power of harnessing vast amounts of clinical and patient data to deliver highly accurate predictive clinical analytics is the basis of CLEW’s platform. CLEW’s analytics engine identifies relationships between real-time physiological data and latent medical conditions, recognizing – in real-time – any changes in the patient’s condition that may indicate the possibility of life-threatening situations. The platform utilizes innovative prediction models derived from Big Data analysis and advanced high-dimensional analytics, to provide hospital management and medical personnel with the preemptive information they require to better manage their resources.
Deployable directly from bedside to a centralized command and control application, the CLEW system combines an easy to use interface with prediction technology, workflow guidance, at-a-glance views and no required user input to improve clinical outcomes, lower costs, and enhance the healthcare provider patient and family experience. CLEW’s unique technology helps focus attention to where it is needed most, staying ahead of impending problems.
- Royal College of Physicians. National Early Warning Score (NEWS) 2: Standardising the assessment of acute-illness severity in the NHS. Updated report of a working party. London: RCP, 2017. Available from: https://www.rcplondon.ac.uk/projects/outputs/national-early-warning-score-news-2 Accessed October 14, 2020.
- Blankush JM, Freeman R, McIlvaine J, Tran T, Nassani S, Leitman IM. Implementation of a novel postoperative monitoring system using automated Modified Early Warning Scores (MEWS) incorporating end-tidal capnography. J Clin Monit Comput. 2017;31(5):1081-1092. doi: 10.1007/s10877-016-9943-4. Epub 2016 Oct 20. PMID: 27766526.
- Churpek MM, Yuen TC, Park SY, Gibbons R, Edelson DP. Using electronic health record data to develop and validate a prediction model for adverse outcomes in the wards. Crit Care Med. 2014;42(4):841-848. doi:10.1097/CCM.0000000000000038
- Marafino BJ, Park M, Davies JM, et al. Validation of Prediction Models for Critical Care Outcomes Using Natural Language Processing of Electronic Health Record Data. JAMA Netw Open. 2018;1(8):e185097. Published 2018 Dec 7. doi:10.1001/jamanetworkopen.2018.5097
- Vincent JL, Moreno R. Clinical review: scoring systems in the critically ill. Crit Care. 2010;14(2):207 http://ccforum.com/content/14/2/207