Concussion Risk Between Individual Football Players: Survival Analysis of Recurrent Events and Non-events

• Published in: Annals of biomedical engineering Vol. 48; no. 11; pp. 2626 - 2638
• Main Authors: Rowson, Steven, Campolettano, Eamon T., Duma, Stefan M., Stemper, Brian, Shah, Alok, Harezlak, Jaroslaw, Riggen, Larry, Mihalik, Jason P., Brooks, Alison, Cameron, Kenneth L., Svoboda, Steven J., Houston, Megan N., McAllister, Thomas, Broglio, Steven, McCrea, Michael
• Format: Journal Article
• Language: English
• Published: Cham Springer International Publishing 01.11.2020; Springer Nature B.V
• Subjects: Biochemistry; Biological and Medical Physics; Biomedical and Life Sciences; Biomedical Engineering and Bioengineering; Biomedicine; Biophysics; Classical Mechanics; Concussion; Concussion Biomechanics in Football; Data analysis; Distribution functions; Football; Head; Impact analysis; Players; Risk analysis; Survival; Survival analysis; Biomechanics; Injury; NCAA-DOD CARE Consortium; Risk function; Sensors; Rotational; Angular
• ISSN: 0090-6964; 1573-9686; 1573-9686
• DOI: 10.1007/s10439-020-02675-x

Concussion tolerance and head impact exposure are highly variable among football players. Recent findings highlight that head impact data analyses need to be performed at the subject level. In this paper, we describe a method of characterizing concussion risk between individuals using a new survival analysis technique developed with real-world head impact data in mind. Our approach addresses the limitations and challenges seen in previous risk analyses of football head impact data. Specifically, this demonstrative analysis appropriately models risk for a combination of left-censored recurrent events (concussions) and right-censored recurrent non-events (head impacts without concussion). Furthermore, the analysis accounts for uneven impact sampling between players. In brief, we propose using the Consistent Threshold method to develop subject-specific risk curves and then determine average risk point estimates between subjects at injurious magnitude values. We describe an approach for selecting an optimal cumulative distribution function to model risk between subjects by minimizing injury prediction error. We illustrate that small differences in distribution fit can result in large predictive errors. Given the vast amounts of on-field data researchers are collecting across sports, this approach can be applied to develop population-specific risk curves that can ultimately inform interventions that reduce concussion incidence