Prediction of Critical Illness During Out-of-Hospital Emergency Care
Seymour CW, Kahn JM, Cooke CR, Watkins TR, Heckbert SR, Rea TD
JAMA. 2010;304:747-754
Study Summary
The Institute of Medicine has recommended the creation of a regionalized, coordinated system of emergency care for high-risk patients. In such a system, patients most in need of acute care would be distributed to centers with the greatest expertise in critical care. In comparison with triage criteria for acute myocardial infarction, stroke, and traumatically injured patients, the current prehospital triage criteria for other critically ill patients are inadequate and subjective. Seymour and colleagues sought to develop a prediction tool for critical illness on the basis of prehospital presentation of noninjured, noncardiac arrest patients. Patients who used emergency medical services in King County, Washington (excluding metropolitan Seattle) from 2002 to 2006 were included, and their prehospital information was linked to hospital records to determine clinical conditions and outcomes.
The cohort was randomly split into 87,266 patients for derivation of the predictors, and 57,647 patients for validation of the identified predictor variables. Critical illness occurred in 5% of all patients, and predictors of critical illness included older age, lower systolic blood pressure, abnormal respiratory rate, lower Glasgow Coma score, lower oxygen saturation by pulse oximetry, and nursing home residence. The predictive value of the critical illness prediction score varied according to the selected cutoff value: at a cutoff of 4, the sensitivity was 0.22 and specificity was 0.98, whereas at a cutoff of 1, the sensitivity was 0.98 but the specificity fell to 0.17. The investigators concluded that a prediction score using prehospital information significantly predicts the development of critical illness during hospitalization.
Viewpoint
Aside from the primary goal of developing a scoring system that could help to implement regionalization of critical care services, this study immediately provides a system by which we can use available information to predict the development of critical illness during hospitalization. A distinct advantage of this tool is the use of clinically available information: predictors were derived from age, gender, blood pressure, heart rate, pulse oximetry, consciousness, and location prior to hospital arrival. However, the current iteration of this tool requires refinement to better predict critical illness with both high sensitivity and high specificity. In particular, depending on the cutoff value applied with the score, the tool could over-predict critical illness for high risk patients and under-predict critical illness at low levels of risk. In addition, the tool requires validation with a broader definition of critical illness, aside from the administrative definitions of sepsis and respiratory failure that were used in this study. Finally, although regionalization has great potential to reduce the variability in care and outcomes that presently characterize critical care, it is a long way from implementation and its true potential in critically ill patients is yet unknown.
Abstract