Brain injury severity due to direct head contact from near-side motor vehicle collisions

Abstract

Objective The objective of this study was to generate functional forms of brain injury risk curves using the National Automotive Sample System Crashworthiness Data System’s (NASS-CDS) database for the years of 2001–2015. The population of interest was near-side occupants who experienced a direct head impact with an injury source located lateral to a typical seated position. Methods Brain injuries were restricted to Abbreviated Injury Scale (AIS) 2005 Update 2008 defined concussions and internal organ injuries of the head. Near-side occupants comprised two major groups, both of which were required to have evidence of head contact (i.e., a head injury with DIRINJ = 1 and SOUCON = 1 or 2): brain injured occupants (MAIS1, MAIS2, MAIS3+) and non-brain injured occupants with some other direct contact head injury (MAIS0). Analyzed cases were required to have an indication of a reasonable crash reconstruction. Injury sources allowed within the final sample consisted of A-pillars, B-pillars, roof/roof rails, impacting vehicles/exterior objects, other components of the vehicle’s side interior, and other occupants or otherwise unspecified interior objects. Risk curves for occupants with brain injury severities of MAIS0, MAIS1+, MAIS2+, and MAIS3+ were generated using multivariate stepwise logistic regressions. Investigated predictors involved vehicle change in velocity, seat belt use, principal direction of force (PDOF), and injury source type (B-pillar and side window). Results Multivariate stepwise logistic regressions identified significant predictors of lateral change in velocity (dvlat) for all injury severity categories, and side window injury source (INJSOU = 56, 57, 58, 106, and 107) for MAIS0 and MAIS1+ risk curves. Although model sensitivity decreased for more severe injury predictions, risk curves dependent on only dvlat yielded accuracies of 70% for all presented models. Conclusions Real world crashes are often complex and lack the benefit of real time monitoring; however, NASS-CDS post-crash investigations provide data useful for injury risk prediction. Further analysis is needed to determine the effect of data confidence, injury source, and accident sequence restrictions on NASS-CDS sampling biases. The presented models likely favor a more conservative risk prediction due to the limitations of NASS-CDS data collection, AIS code conversion, and unweighted sample analysis.