Evaluations of quality and outcomes of care in emergency medical services for children (EMSC), and comparisons of outcomes between EMSC systems or components of systems, require a method of adjusting for important differences in acuity between patient populations. No satisfactory method of risk adjustment currently exists that is broadly applicable to all EMSC patients. This study aims to develop a multivariable model, using predictor information available at the time of patient triage, to predict the probability of the need for different levels of emergency care. We plan a cohort study of 10,000 visits by children age 18 years or less to any of the eight acute care hospital emergency departments in Delaware, randomly selected from all visits over a 12 month period. Information will be obtained from review of ED records. Predictor variables of interest will include data collected at the time of patient triage and registration: age; date and time or visit; gender; race/ethnicity; chief complaint; triage vital signs; presence of significant past medical history (coded as positive or negative); current medications; mode of transportation; insurance coverage. A three level outcome will reflect the level of care provided: routine-standard nursing and physician care only; ED treatment-diagnostic tests performed or treatment provided in the ED, and patient discharged to home; and admission to the hospital (including transfer to another hospital or death in the ED). Using multivariate analytic techniques, including multinomial logistic regression, a predictive model will be developed in a subsample of 75 percent of the subjects, and validated in the remaining 25 percent. The creation of a new pediatric emergency assessment tool will permit risk stratification of patients according to the probability of requiring increasing intensity of services. Such a tool, after being validated in other settings in a future study, will permit adjustment for difference in patient acuity when evaluating outcomes of EMSC, or when comparing the performance of different systems of EMSC. It can also be used to generate expected distributions of levels of care for quality improvement purposes.
