Prediction of critical illness during out-of-hospital emergency care

Abstract

Context: Early identification of nontrauma patients in need of critical care services in the emergency setting may improve triage decisions and facilitate regionalization of critical care.

Objectives: To determine the out-of-hospital clinical predictors of critical illness and to characterize the performance of a simple score for out-of-hospital prediction of development of critical illness during hospitalization.

Design and setting: Population-based cohort study of an emergency medical services (EMS) system in greater King County, Washington (excluding metropolitan Seattle), that transports to 16 receiving facilities.

Patients: Nontrauma, non-cardiac arrest adult patients transported to a hospital by King County EMS from 2002 through 2006. Eligible records with complete data (N = 144,913) were linked to hospital discharge data and randomly split into development (n = 87,266 [60%]) and validation (n = 57,647 [40%]) cohorts.

Main outcome measure: Development of critical illness, defined as severe sepsis, delivery of mechanical ventilation, or death during hospitalization.

Results: Critical illness occurred during hospitalization in 5% of the development (n = 4835) and validation (n = 3121) cohorts. Multivariable predictors of critical illness included older age, lower systolic blood pressure, abnormal respiratory rate, lower Glasgow Coma Scale score, lower pulse oximetry, and nursing home residence during out-of-hospital care (P < .01 for all). When applying a summary critical illness prediction score to the validation cohort (range, 0-8), the area under the receiver operating characteristic curve was 0.77 (95% confidence interval [CI], 0.76-0.78), with satisfactory calibration slope (1.0). Using a score threshold of 4 or higher, sensitivity was 0.22 (95% CI, 0.20-0.23), specificity was 0.98 (95% CI, 0.98-0.98), positive likelihood ratio was 9.8 (95% CI, 8.9-10.6), and negative likelihood ratio was 0.80 (95% CI, 0.79- 0.82). A threshold of 1 or greater for critical illness improved sensitivity (0.98; 95% CI, 0.97-0.98) but reduced specificity (0.17; 95% CI, 0.17-0.17).

Conclusions: In a population-based cohort, the score on a prediction rule using out-of-hospital factors was significantly associated with the development of critical illness during hospitalization. This score requires external validation in an independent population.

Figure 1

Participant Accrual Missing covariate data do not include participants who were missing Glasgow Coma Scale or pulse oximetry data.

Figure 2

Observed vs Expected Probability of Critical Illness Across Scores in the Development and Validation Cohorts Error bars represent 95% confidence intervals (CIs). Data points are staggered for clarity. If patients with a critical illness score of 6 or higher are collapsed into a single group, the probability of critical illness is 0.76 (95% CI, 0.69-0.82) in the development cohort and 0.78 (95% CI, 0.69-0.85) in the validation cohort.