Development and Initial Implementation of a Machine-Learning-Based Predictive Index for Critical Deterioration among Hospitalized Children

Background: Inpatient critical deterioration is a severe worsening of a hospitalized child’s condition that warrants initiation of life support systems to prevent death. Many episodes of critical deterioration are preventable, yet objective indicators of impending deterioration are often unrecognized or underappreciated. Structured data continuously captured in the electronic health record (EHR), such as laboratory values, vital signs and nursing assessments, can be used to ascertain the severity of a patient’s condition. Recognizing that clinicians are only able to synthesize EHR information streams episodically, we aimed to augment our institution’s existing surveillance systems by creating an acuity index to provide a continuous, probabilistic assessment of the likelihood of critical deterioration among inpatient children. Methods: Conceptualization of the acuity index adhered to the Littenberg framework …