Comparative Effectiveness in Traumatic Brain Injury: Problem Rich/Solution Poor?

Severe traumatic brain injury is a major public health problem accounting for considerable mortality and disability, especially in the young.   Non severe or mild-TBI is especially common and has received greater attention in recent years. As is true in multiple areas of medical practice, despite the large numbers of patients there is relatively little evidence-base upon which to choose optimal care.  Yet decisions are made on a daily basis to administer one or another treatment to patients. Consensus based guidelines provide direction for patient care and studies have demonstrated that adherence to the guidelines improve functional outcome. However none of the major components of the care of severe head injury are supported by definitive, or so called (LEVEL1) evidence. https://www.braintrauma.org/pdf/protected/Guidelines_Management_2007w_bookmarks.pdf .  As a result the standard of care is local, and there is substantial variation which makes discovery research difficult.  Any signal due to the beneficial effects of a new treatment may be lost in the variation due to differences in practice patterns.  The problem of comparing therapies is daunting when none have initially been shown to be definitively efficacious but there is reluctance not to utilize them. In order to make any progress a number of steps need to be taken: 1) develop standard measures of data collection ie. common data elements or CDEs.  http://www.commondataelements.ninds.nih.gov/TBI.aspx.  2a) utilize CDEs, including those that risk stratify patients, for pragmatic trials to identify which treatments in the guidelines afford benefit and which do not.  This usually requires randomized trials, where either the individual or the site are the unit of randomization.  2b) explore new methodology to detect a signal of benefit or harm from large non-randomized data sets.  Heretofore the latter has been treacherous due to sample bias (ie. patients who get treatment A may be very different from those who do not). The problem of signal detection in large data sets is currently a topic of intense interest for medical research in which there are a large number of variables but also extremely large data sets coming from electronic medical records and monitoring devices. If successful the next stage is to move from evidence to standardized treatment protocols and document improved patient outcome.