Vital Signs Data and Probability of Hospitalization, Transfer to Another Facility, or Emergency Department Death Among Adults Presenting for Medical Illnesses to the Emergency Department at a Large Urban Hospital in the United States.

ABSTRACT

BACKGROUND: Vital signs are routinely measured from patients presenting to the emergency department (ED), but how they predict clinical outcomes like hospitalization is unclear. OBJECTIVES: To evaluate how pulse, respiratory rate, temperature, and mean arterial pressure (MAP) at ED presentation predicted probability of hospitalization, transfer to another center, or death in the ED (as a composite outcome) vs. other ED dispositions (discharged, eloped, and sent to observation or labor and delivery), and to assess the performance of different modeling strategies, specifically, models including flexible forms of vital signs (as restricted cubic splines) vs. linear forms (untransformed numeric variables) vs. binary transformations (vital signs values categorized simply as normal or abnormal). METHODS: We analyzed the data of 12,660 adults presenting for medical illnesses to the ED at the University of California, San Francisco Medical Center, San Francisco, California, throughout 2014. We used flexible forms of vital signs data at presentation (pulse, temperature, respiratory rate, and MAP) to predict ED disposition (admitted, transferred, or died, vs. other ED dispositions) and to guide binary transformation of vital signs. We compared performance of models including vital signs as flexible terms, binary transformations, or linear terms. RESULTS: A model including flexible forms of vital signs and age to predict the outcome had good calibration and moderate discrimination (C-statistic = 71.2, 95% confidence interval [CI] 70.0-72.4). Binary transformation of vital signs had minimal impact on performance (C-statistic = 71.3, 95% CI 70.2-72.5). A model with linear forms was less calibrated and had slightly reduced discrimination (C-statistic = 70.3, 95% CI 69.1-71.5). CONCLUSIONS: Findings suggest that flexible modeling of vital signs may better reflect their association with clinical outcomes. Future studies to evaluate how vital signs could assist clinical decision-making in acute care settings are suggested.