Players from three National Collegiate Athletic Association (NCAA) football programs (Brown University, Dartmouth College, and Virginia Tech) were provided the opportunity to participate in this IRB-approved observational study after informed consent was obtained. During the 2007 and 2008 fall football seasons, a total of 254 male players from the three teams, denoted arbitrarily as Team A (n = 85 players), Team B (n = 83 players), and Team C (n = 86 players), participated in this study. Of these players, 116 were monitored in both seasons. This participant turnover was expected, and due primarily to typical roster fluctuations on a collegiate team (e.g., graduation, incoming freshman, and injuries). Each player was assigned a unique identification number and categorized in one of eight position units defined by the team staff as the player’s primary position: defensive line (DL, n = 39), linebacker (LB, n = 38), defensive back (DB, n = 46), offensive line (OL, n = 60), offensive running back (RB, n = 29), wide receiver (WR, n = 26), quarterback (QB, n = 10), and Special Teams (ST, n = 6).
All players wore Riddell (Riddell, Chicago, IL) football helmets instrumented with the HIT System (), a device capable of recording the acceleration–time history of an impact from six linear acceler-ometers at 1000 Hz. Impact data from all participating institutions were uploaded to a secure central server with a consolidated database, and subsequently exported for statistical analysis. Data were reduced in postprocessing to exclude any impact event with a peak resultant linear head acceleration less than 10 g
(Mihalik et al., 2007
) to eliminate events that had been determined during initial system development to be inconsequential, nonimpact events (e.g., running and jumping). Any impact event in which the acceleration–time history pattern of the six linear accelerometers did not match the theoretical pattern for rigid body head acceleration (Crisco et al., 2004
), such as a spike in a single accelerometer signal that can occur when a player removes his helmet and throws or kicks it, was also excluded. These data reduction methods have been previously verified (Brolinson et al., 2006
; Duma et al., 2005
; Funk et al., 2007
; Manoogian et al., 2006
), as was the accuracy of the HIT algorithm (Crisco et al., 2004
). Laboratory tests have determined that the linear and rotational accelerations measured by the HIT system were within ± 4% of a helmet-equipped Hybrid III dummy (Duma et al., 2005
Figure 1 Football players wore helmets instrumented with the HIT system that was specifically designed to record head accelerations as a result of an impact to the helmet without interfering with play. The HIT System (A) comprises an in-helmet unit containing (more ...)
A team session (session) was defined as either a formal team practice (players wore protective equipment with the potential of head contact) or a game (competitions and scrimmages). An individual player was defined to have participated in a session when at least one head impact was recorded for that given player. Impacts that were recorded outside the time of the team session, as defined by the team staff, were excluded from the analysis.
Head impact magnitude was quantified by peak linear acceleration (g
) and peak rotational acceleration (rad/s2
). Each recorded impact event was processed using a simulated annealing optimization algorithm to solve for the linear acceleration magnitude at the head center of gravity (CG) (Crisco et al., 2004
). Peak rotational acceleration was calculated as the vector product of peak linear acceleration and a point of rotation estimated to be 10 cm inferior to the CG of the head. Laboratory testing has confirmed that this location is consistent with the impact response of the Hybrid III dummy (Duma et al., 2005
). Helmet impact location for each impact was computed as azimuth and elevation angles in an anatomical coordinate system relative to the CG of the head (Crisco et al., 2004
) and then categorized into one of five helmet impact locations: front (F), left (L), right (R), back (B), and top (T) (). Four equally spaced regions centered on the anatomical midsagittal and coronal planes defined front, left, right, and back impact locations. All impacts occurring above an elevation angle of 65°, where 0° elevation was defined as a horizontal plane through the center of gravity of the head, were defined as impacts to the top of the helmet. In addition, a nondimensional measure of head impact severity, HITsp (Greenwald et al., 2008
), was computed. HITsp transforms the computed head impact measures of peak linear and peak angular acceleration into a single latent variable using principal component analysis, and applies a weighting factor based on impact location (Greenwald et al., 2008
The 50th and 95th percentile values of the peak linear and peak rotational acceleration were first calculated across the entire study, independent of player. For analysis, individual players’ 50th and 95th percentiles were calculated for each impact location (front, left, right, top, and back) within all their practices and within all their games. For HITsp, the 50th and 95th percentiles were calculated for practices and for games without consideration of location. HITsp was not analyzed among impact locations because impact location is a factor in computing HITsp values. The 50th and 95th percentile values were positively skewed (normality test failed, p < .05), making general linear models that assume normally distributed variances inappropriate. Therefore, generalized estimating equations for log-normally distributed data were used to model the 50th and 95th percentiles, with repeated measures within players treated as having correlated error with a heterogeneous compound symmetrical variance–covariance matrix for session type × location, block diagonal by season. For peak linear and rotational accelerations, the predictive factors were team, season, season × team, session type, impact location, player position, and the two- and three-way interactions among session type, impact location, and player position. The interactions between season and team were also included to allow for differences in the changes across seasons between institutions. For HITsp, the factors were team, season, team × season, session, player position, and session × player position.
Statistical significance was set at α = .05. Given the large number of hypotheses, to minimize type II error, α was only adjusted for multiple comparisons within families of effects (e.g., differences by player position were adjusted without consideration of differences by helmet impact location). These post hoc tests used the Holm-simulated adjustment procedure. All statistical analyses were performed using SAS version 9.2 (SAS Institute, Cary, NC).