Predicting patient outcomes via neural network estimation of discharge APACHE scores for traumatic brain injury

It is highly desirable to be able to predict the likely outcome of critical patients admitted to the intensive care unit (ICU) for traumatic brain injury (TBI). Vital signs, laboratory values, and clinical assessments from throughout a patient’s ICU stay were collected retrospectively in an IRB-approved protocol from a Level I Trauma-Military Medical Center in the Southwest. ICU patients were included if they had been admitted for TBI during a five-year period ending in October 2007. Data were collected for 139 ICU patients with TBI. Admission and discharge APACHE IV scores were then derived from the collected data for each patient. A static back propagation neural network was developed to predict a patient’s ICU outcome vis-a-vis discharge APACHE IV scores. The resulting network, trained using leave-one-out methodology, was able to predict the discharge APACHE score on average within 12.9% of the actual score.