Early Warning Scores
Early Warning Scores are a subset of clinical decision support (CDS) algorithms focused on the prediction of clinical decompensation in patients.
Early Warning Score scores are typically (although not always) implemented as CDS in Electronic Medical Record systems (EHRs). These systems use charted elements to predict clinical deterioration in patients. The elements often used include nursing assessments, vital signs, orders, laboratory results, and problem lists. These algorithms typically return a score that has a clinical workflow associated with in to indicate a level of intervention ranging from re-assessing the patient to consultation with an additional physician to transferring a patient to a higher level of care. The algorithms themselves typically are associated with a specific level of sensitivity and specificity for a specific clinical outcome at a specific score. For example a score of 3 in a sample system might indicate an 80% sensitivity and 60% specificity that a patient will be intubated (an escalation in care) in the next 24 hours. Often times the implementation process and local clinical environment dictate modifications to the interventions as compared with the initial study.
Goals of the Algorithms
The ultimate goal of clinical care is generally improvement of the patient. In inpatient care environments, a key part of caring for these patients is to observe these patients and perform interventions to improve the patient and prevent/treat deteriorations of the patient. Early warning scoring systems are designed to assist in the prevention and early intervention on patient decompensation. It is well established that many medical conditions associated with decompensation are improved by early intervention. This improves outcomes for the patient. This can be applied in a number of ways such as predicting decompensation of patients on medical/surgical units (general inpatient units) to prevent transfers to Intensive Care Units. Other example applications can be raising situational awareness in ICUs for patients with concern for death.
Note that goals for all patients are not necessarily clinical improvement such as end of life and hospice care patients. In these patients these algorithms may not apply.
A major example of an Early Warning Score is the PEWS or Pediatric Early Warning System. This system was studied at a children's hospital and studied to assess prevention of code blue events which are sudden major decompensations in patients. PEWS uses nursing charted heart rates, respiratory rates, work of breathing, capillary refill, and mental status to generate a score 0-9 with 9 being the most clinical severe state. A score of 5 or higher was selected based on clinical expertise as the trigger point. This score was applied to all inpatient pediatric patients and scored every 4 hours with nursing collected vital signs. If the score was 5 or high a rapid response was performed which is a process where an Intensive Care team comes to evaluate the patient for needs to changes and escalations in care to prevent significant events, in this case code blue events.
Results of this study showed increases in numbers of rapid response events, particularly at night, but did not show improvement in code blue events (although code blue events are rare events at baseline).
Despite this unimpressive result, many sites continue to use PEWS, often with local modification.
EWS - Early Warning Score - an early scoring system using vital signs.
MEWS - Modified Early Warning Score - an updated version of the EWS based on Respiratory rate, Heart rate, Systolic blood pressure, Conscious level, Temperature, and Hourly urine output (for previous 2 hours).
Rothman Index - System based on multiple components of EHR nursing documentation to predict clinical decompensation with EHR user interface.
A 2014 review found that EWS and MEWS were studied in 7 studies that had mixed methods. The studies had varied thresholds and interventions so general conclusions were not able to be made but 6 out of 7 studies had positive results.
Another 2014 review demonstrated accuracy in these algorithms in predicting death or cardiac arrest within 24 hours but was unable to assess the impact of the interventions on outcome or cost.
When implementing an early warning system, it is important to select a system that is valid and implement the workflows to meet the local needs and processes.
There are a number of challenge components of using and choosing these early warning scoring algorithms which are discussed further here.
Submitted by Kevin O'Bryan