Application of Data Sciences in Traumatic Brain Injury

点击翻译按钮获取中文摘要

ABSTRACT Traumatic brain injury (TBI) is a leading cause of morbidity and mortality, and patients with moderate-severe traumatic brain TBI often require urgent/emergent surgical and anesthesia care. Patients with TBI who have surgery have poor outcomes, attributed to a high (>50%) prevalence of perioperative second insults such as hypotension and hypocarbia, which reduce cerebral perfusion and cause cerebral ischemia. Anesthesiologists provide analgesia, sedation, immobility, and amnesia, and aim to confer physiological stability, expected patient response, real-time physiological data, and professional judgement but are unfortunately unable to accurately predict in real time which patients with TBI will have hypotension and hypocarbia. Yet, avoidance of these second insults increases discharge survival among patients with TBI. Predicting and preventing hypotension and hypocarbia during TBI care is, therefore, vital, and avoidance of hypotension and hypocarbia are key performance indicators for perioperative TBI care. Small data science studies suggest that machine learning (ML) techniques can model and predict TBI pathophysiology and help reduce unwanted second insults after TBI. The project goal is to use ML methods to prevent second insults (hypotension and hypocarbia) during urgent/emergent perioperative TBI care. In response to PA-16-161, we propose 2 Specific Aims: 1) To construct and identify the TBI physiological ML model that most accurately predicts perioperative hypotension and hypocarbia, and 2) To develop ML derived personalized prescriptions for prevention of perioperative hypotension and hypocarbia. This project is innovative and will be impactful because the approach is grounded in strong data science, and acute care, and implementation science frameworks, because it develops ML derived prescriptions to prevent hypotension and hypocarbia, and because we use ML solutions to improve care quality and outcomes after TBI.