Rowson et al. (2012) use data from 301,034 head impacts among US football players (57 of which resulted in concussions) to estimate a logistic risk function for concussion as a function of rotational acceleration. There were three issues in the estimation of this logistic risk function. First, it did not account for uncertainty in finite sample of the head impacts. This might seem minor, given that over 300,000 observations were made, however, because only 57 resulted in concussions, there is still substantial uncertainty in the relationship between risk of concussion and rotational acceleration. (A secondary issue is the assumption of linearity between the log-odds of concussion and rotational acceleration ? which the small number of concussions, it is unlikely that small deviations from linearity could be determined, but large deviations could be tested for.) A second issue is that the measure of rotational acceleration is subject to non-trivial levels of error (Allison et al. 2013), which will impact the estimates of the relationship between concussion risk and rotational acceleration (by attenuating it if this error is non-differential, as Allison et al. suggest it is). A third issue is misclassification of the concussion outcome itself, with underreporting believed to be as high as 50% (suggesting there may be have been more than 100 or more injuries in the Rowson et al. dataset, even though only 57 were reported).
Funding:
The Children's Hospital of Philadelphia
Funding Period:
08/01/2014 to 03/31/2015