Machine learning to inform health services and policy for traumatic brain injury

项目总结

Project Summary Traumatic brain injury (TBI) is recognized as the leading cause of death and disability in all parts of the world and costs the international economy approximately US$400 billion annually, which, given an estimated standardized gross world product of US $73.7 trillion, is a striking 0.5% of the entire annual global output. To address the profound issues related to a drastic increase in emergency department visits and hospitalizations for TBI over the past decades, the United States Congress highlighted injury surveillance as a federal priority. The Centers for Disease Control and Prevention defines surveillance as “use of health-related data that precede diagnosis and signal a sufficient probability of a case or an outbreak to warrant further public health response”. To prevent TBI, it is essential to understand its distribution and patterns, in addition to having strong knowledge of clinical disorders, characteristic, or other definable entity, that differentiates TBI from other clinical populations. A critical barrier to the progress of the NIH-funded program “Comorbidity in traumatic brain injury and risk of all-cause mortality, functional and financial burden: a decade-long population based cohort study” was the presence of complex and multifaceted comorbidities in a patient with TBI before and at the time of the injury, and their links to patients’ frailty, injury circumstances, severity, and outcomes. This resulted in a shift in the research paradigm, and development of a novel data mining approach used in genomics to sequence more than 70,000 clinical diagnosis codes in a TBI population, and compare them to a matched population. The developed data mining approach allowed not only the validation of previously known risk factors of TBI, but also the identification of associations previously unknown, without any preconceived human biases. This project will continue advancement of a non-hypothesis driven scientific approach, which will: (1) Characterize patients with TBI at three different time periods in relation to the TBI event – before, at the time of, and after the injury; (2) Develop individual and population level models to study the transitions between the different time states; and (3) Construct and validate predictive models of susceptibility to TBI events, adverse outcomes, and high healthcare resource use at the individual and population level. Decades- long population-based health administrative data from the publicly-funded healthcare system in Ontario, Canada is ready to be further analysed for clinical and technological advancement, to support human thinking in categorizing personal, clinical, and environmental exposure data preceding TBI.