Although no set of variables yet provided ideal sensitivity to concussive injuries, in vivo examinations of head kinematics of injury tolerance have yielded progressively more sophisticated findings with a progression from a single variable to a multi-variable analysis. Early work by Pellman et al (30) reported on data collected through Hybrid III crash dummy reconstruction of open field impacts in a small subset of concussed and non-concussed elite athletes. The findings indicated linear acceleration to be most directly related to concussion with a mean threshold for injury to be 98g and an impact generating a minimum 70–75g necessary to cause injury. These impact levels corresponded to a 75% and 50% injury risk, respectfully. Application of these thresholds to other levels of play however, has fallen short in accurately predicting concussions. In fact, in a dataset generated from high school level football athletes 271 impacts exceeded the 70g threshold and 78 impacts exceeded the 98g magnitude with only five reported concussive injuries(7).

Guskiewicz (14) later reported on data collected in collegiate level football players equipped with the Head Impact Telemetry System (HITS). This instrumentation permits the tracking of all impacts (concussive and not) in all athletes equipped with the device. The authors reported the mean linear acceleration of the 13 concussions recorded to be 102.8 g with a rotational acceleration of 5311.6 rad/s². The value for linear acceleration appears similar to that of professional athletes, although injury tolerance levels were not estimated. Notably, when the athletes were evaluated following a diagnosed concussion there appeared to be no relationship between the magnitude of impact and injury severity as measured by decrements to postural control, neurocognitive functioning, and increased symptom reports. This finding was counter to the commonly held belief that larger magnitude impacts would result in more severe injuries as indicated by worse performance on clinical tests.

Similarly, Greenwald and colleagues (12) calculated the risk for concussion from HITS biomechanical data collected in a large sample of collegiate and high school level football. Using a weighted principal component analysis from 17 concussive impacts, the investigators reported a combination of variables that included linear acceleration, rotational acceleration, Head Impact Criteria (HIC) (27), and impact location provided the highest sensitivity to concussive injuries. The analysis indicated that athletes exceeding the calculated principle component value of 63 were at 75% risk for injury. This multi-variable equation extends the work of previously published findings, but the ability to consistently and accurately predict concussion based on biomechanical data remains elusive.

Others have tried to improve upon injury diagnostics by predicting the relationship between cranial kinematics and brain tissue mechanics from impacts associated with injury. Zang (34) applied the National Football League study data to a finite element (FE) model of a human skull and brain in an attempt to predict both injury occurrence and severity. The authors reported the greatest level of shear stress was located in the brainstem and provided the highest predictive value of concussive injury. The predictive value of shear stress was surpassed by only the combined use of both the resultant linear and angular acceleration values. Concussion risk levels were also calculated and deviated little from previous reports, whereby a 106g and 79000 rad/s² impact resulted in an 80% injury risk. Implementing a more extensively validated FE model, Kleiven (20) later evaluated the relationship between head kinematics and the simulated cerebral tissue response in 58 impacts (25 concussions and 33 non-concussive impacts) recorded from professional football. This analysis suggested that brain stem shear stress did not provide the highest predictive values to injury, but rather a combination of HIC and rotational velocity correlated with maximal brain strain and thusly provided the highest injury prediction values. A separate evaluation of simulated brain motion predicted from HITS recorded impact kinematics (without concussions) in collegiate level football demonstrated that measures of cerebral motion (eg maximal tissue strain) were most highly correlated to angular acceleration and velocity, but not with linear acceleration(32). Collectively, these findings suggest that rotational motion of the head following impact may be the best predictor of cerebral tissue strain and subsequent injury.

The HITS is a novel telemetry system designed to identify the location and magnitude of head impacts in real time. The HITS is capable of monitoring and recording the location and magnitude of impacts to the head that occur during sport. The system consists of two components: an encoder that is retrofitted into the football helmet and a sideline computer for data recording and storage. The encoder consists of six thimble sized single-axis accelerometers recording at 1000Hz, a telemetry unit and data storage device, and an onboard battery pack. The memory device can record and store up to 100 impacts when out of range of the sideline computer. The components are sealed in waterproof plastic and inserted between the internal padding of any standard Riddell football helmet (Figure 1). Helmets equipped with the HIT System look and function identically to other helmets and continue to meet National Operating Committee on Standards for Athletic Equipment (NOCSAE) standards for safety.

A sideline computer receives impact data continuously from up to 100 players simultaneously at ranges up to 137m. Consistent communication between the helmet and the sideline receiver is achieved through a continually changing, Federal Communications Commission approved radio frequency range. The sideline computer is housed in a waterproof plastic case and cooled with an internal fan. Within the case, the computer is padded with foam to lessen the effects of any ancillary impacts. The system has been successfully used in stadiums of more than 100,000 fans, with TV, radio, and other 2-way communication systems all simultaneously functioning.

In order for an impact to be recorded, a minimum threshold of 15g must be exceeded on at least one accelerometer. Following impact a total of 40ms worth of data are recorded as representative of the impact. This consists of the 12ms prior to impact and the 28ms following. Importantly, those athletes who are involved with the majority of plays (i.e. linemen), but whose impacts are obscured from video capture, can be monitored. A novel algorithm that determines the location and magnitude of each impact to the head (9) was validated against Hybrid III crash dummies fitted with a 3-2-2-2 accelerometer array (28). The correlation between the HITS and Hybrid III dummies was high (r=0.98) with a 4% absolute error rate when estimating both linear and angular accelerations (10). Likewise, impact location accuracy was validated using the National Operating Committee on Standards for Athletic Equipment (NOCSAE) testing protocol, whereby the HITS accurately identified impact location within ±0.41cm(10).

For the purposes of this study the definition of concussion was taken from the American Academy of Neurology which states: “Concussion is a trauma-induced alteration in mental status that may or may not involve loss of consciousness” (1). All diagnoses where made by a certified athletic trainer and/or physician and the same physician determined return to play following injury. Injuries were not graded because of the lack of support from medical organizations for the use of grading scales (13) and emerging evidence that current grading scales are not accurately representative of severity (22).

The 166 data collection sessions were spread across 128 practices and 38 games and accounted for 54247 recorded impacts. Table 2 presents mean biomechanical data for practice and game sessions throughout the investigation. Across the duration of the investigation, mean linear acceleration was 25.1(15.4) g from all sessions with 24.2(14.3) g impacts occurring during practices and 26.1(16.6) g impacts games. Rotational acceleration during all sessions was 1627.1(1182.9) rad/s², 1554.3(1082.8) rad/s² in practices, and 1711.2(1284.0) rad/s² in games. Impact duration was longest during practices at 10.2(3.6) ms, shortest in games at 10.1(3.7) ms, with an average of 10.2(3.7) ms from all sessions. Across the four year investigation 13 concussions were diagnosed with specific demographic and biomechanical information about each injury presented in Table 3.

To explore the relationships between “concussion” and biomechanical variables, we fit a classification tree to the data. The R package “tree” was used and the addition of nodes was stopped when no less than five observations were present. Using the complete dataset, the tree suggests that the best predictors for the risk of concussion are: resultant rotational acceleration, resultant linear acceleration, and impact location (Figure 2). The accuracy of injury prediction improves with the addition of each variable, indicating that impacts resulting in a rotational acceleration in excess of 5582.3 rad/s² (n=684) will be associated with 1.9% chance of injury for impacts above this level. When resultant linear acceleration in excess of 96.1 g (n=159) is included in the decision making process the chance rises to 6.9%. Further inclusion of impact location to the front, side, or top of helmet (n=82) elevates the chance to 13.4%. If we further examine the risk set for impacts that exceed 5582.3 rad/s², but fall below 8445 rad/s² (n=47), the risk of injury increases to 21.3%. Impacts exceeding 8445 rad/s2 are associated with a lower injury risk, with concussions occurring once out of 35 impacts or 2.9% of impacts meeting these criteria. Further evaluation of resultant linear acceleration exceeding 96.1 g, but less than 102.6 g (n=9), increases the risk of injury to 55.6%. Impacts exceeding 102.6 g are associated with a lower injury risk with concussions occurring five out of 38 times or 13.2% of impacts meeting these criteria. Because the number of impacts meeting the criteria at the lower end of the tree (Figure 2) represents a small percent of all impacts, the risk calculations there are less reliable than those at the higher end of the tree.

To check the stability of the fitted classification tree, we repeatedly sampled 80% of the data randomly from the complete dataset and constructed the tree for each random sub-sample. After examining ten sub-samples, we found that the basic structure of the risk set from those trees remained stable.

Our analysis demonstrates that rotational accelerations in excess of 5582.3 rad/s² may be the bottom threshold in increasing the probability of concussion in the high school athlete. This value and our range of values (5582 to 9515 rad/s²: Table 3) that resulted in concussion is similar to findings generated from adult athletes (17;30). Indeed, biomechanical data collected with the Head Impact Telemetry System on concussed collegiate athletes identified the mean rotational acceleration at 5311.6 rad/s² (17). Likewise, video reconstruction of concussed professional athletes found the mean rotational acceleration to be only slightly higher at 6432 rad/s² (30). It is unknown if these differences are clinically meaningful, but they all exceed previously proposed concussions thresholds (26).

The evaluation of linear acceleration values collected in this investigation indicated that impacts generating acceleration in excess of 96.1 g increased the sensitivity of our decision making tree. This value is nearly identical to the mean linear acceleration of 98g reported in professional athletes (30) and similar to 102.8 g mean reported in collegiate athletes (17). The range of linear accelerations (77.8 to 146.0 g: Table 3) also fell within the range of previous reports.

A complex interrelationship exists between impact location, linear and rotational acceleration and concussion. Temporal impacts, particularly when the athlete is unaware of the impending blow, are thought to generate the greatest concussion risk (16). This may be in part to the higher rotational component seen here (Table 3) and elsewhere (17) that result from lateral forces and its effect on brain stem integrity and subsequent loss of consciousness (25). Anatomically, brain stem fibers have an increased susceptibility to rotational loads due to their linear alignment (2) and computer simulation models indicate higher brain shearing with temporal impacts (34). Further, an analysis of Australian rules football athletes revealed the majority concussions (61 of 97) resulted from temporal region impacts that generated a high rotational component (24). Others have reported that linear acceleration is the single best predictor of concussion (30) and more tightly linked with severe injuries such as cerebral contusion and hemorrhaging (16). Our data do not support the use of linear acceleration as the prime variable of interest as rotational acceleration was the chief predictor within our classification tree.

The combined interpretation of rotational acceleration, linear acceleration, and impact location significantly improved the sensitivity of biomechanical variables in making injury diagnoses over previous estimates. In particular, when the single variable concussion threshold proposed by the Pellman and colleagues (30) was applied to our dataset, 259 impacts exceeded the 98g level, nine of which resulted in concussions, providing a 3.5% sensitivity. An alternative multivariable threshold estimate implementing linear acceleration, rotational acceleration, HIC, and impact location in excess of 63(12) also identified nine concussive episodes, but with only 150 false positive results (6.0% sensitivity). Our CART analysis however, accurately identified ten concussions among the 47 impacts exceeding our injury tolerance level (21.3% sensitivity). The improved sensitivity may have resulted from a more sophisticated data analysis technique and/or the development of an injury profile specific to the high school athlete. Importantly, the analysis supports a multi-variable approach in determining the biomechanical components of the concussion mechanism.

An additional concern related to injury diagnostics is the tolerance athletes have in sustaining blows that do not result in concussion. That is, individual variability results in not all impacts exceeding the injury predictor levels to result in concussion. Thus, there is utility in determining the percent risk for injury given a pre-determined set of head acceleration criteria. As such, evaluation of concussion data collected in the NFL suggests that a 98g acceleration places the athlete at 75% injury risk (34). These same data were later re-evaluated on the premise of exposure bias during the original data collection period (11). The authors’ statistical evaluation suggested that similar impact values (ie 107g, 6619 rad/s², and HIC of 191) lowered the injury risk in the professional athletes to 1%. Further evaluation of collegiate level impacts indicated that a similar set of acceleration criteria (ie 109g, 6714rad/s², and HIC of 232) resulted in the same 1% injury risk level (11). Application of these estimates to our high school dataset revealed 71 impacts exceeded the 1% risk estimate for the professional athlete, three of which were concussive impacts (4.3%). Where as application of the 1% risk estimates generated for collegiate athlete athletes generated 60 impacts exceeding the threshold with one reported concussion (1.7%). These findings suggest that the tolerance level in the high school football athlete is slightly lower than that of the collegiate and professional athlete, placing him at an increased risk for injury compared to his older counterparts given equivalent head kinematics following impact.

The clarification of concussion biomechanics and injury tolerance levels will ultimately yield better injury prevention equipment. American football helmets that were initially designed to reduce the risk of skull to skull contact resulting in fracture are now engineered to reduce concussion incidence. The Riddell (Elyria, OH) Revolution uses thickened padding in the temporal region that protects the athlete from lateral blows (8). Revolution helmets were solely employed in this investigation and we found only 2 of 13 concussions resulted from side impacts. It may therefore be reasonable to believe that this innovation has been successful at reducing concussion incidence resulting from blows to the lateral helmet, but other impact areas should be evaluated more closely. For example, consistent with impact descriptives from high school football (7), concussions resulting from top of head impacts were the result of the highest magnitude linear acceleration values (108.8 g: Table 3), but the majority of concussions resulted from front and back impacts (9 of 13).

Further, not all athletic endeavors require or allow the use of protective headgear, necessitating the exploration of alternative prevention methods. Neck strengthening protocols are one suggested, but unproven, concussion deterrent intervention. The underlying justification for this type of training is to tighten the neck musculature at the time of impact to create a single rigid unit of both the head and body to increase effective mass and decrease the resulting post-impact head acceleration. Animal based research highlights this principle when the impacted head is and is not secured by external means. Identical force application generating a concussion when the head is allowed to rotate freely did not yield injury when the head was externally secured (25). Others have shown the same effect by using electrical stimulation to generate neck musculature contractions forceful enough to prevent head motion and consequently concussion (19). Studies of soccer heading yield similar findings with sternocleidomastoid activation just prior to impact to stabilize the head against the ball (4). In combination these investigations support the idea that tensing the neck musculature in an effort to increase effective mass at the moment of impact and lessen head acceleration following impact may be an effective concussion prevention strategy.