In recent years, dramatic reductions in the size and cost of triaxial accelerometers, gyroscopes, global positioning systems, magnetic field detectors, and heart rate monitors have led to the development of integrated microelectromechanical systems (MEMs) that can be worn during participation in sports (22). When the output of interest is limited to the magnitude and frequency of linear accelerations, such devices are widely designated as inertial measurement units (IMUs). Wearable MEMs are believed to have the potential to dramatically advance training methods and injury prevention initiatives, but very little research evidence currently exists to support utilization of the data that they provide for making decisions relating to performance optimization and injury risk reduction (4,6).
Player “load” is a term used to refer to the sum of instantaneous change in linear acceleration recorded by an IMU during a practice session or game, which provides a numeric representation of the magnitude and frequency of the effects of ground reaction forces and collisions on the body mass inertia (3,11,17). Inertial load quantified by an IMU has been shown to provide a valid representation of the physical demands imposed by volitional sport-related movements through comparison to ground reaction force data that was derived from a set of 6 force plates (17). Also, a very strong correlation has been reported between inertial load measurements derived from the same IMU model and the number of collisions experienced by rugby players (r = 0.96), which were related to one another through temporal synchronization of the IMU output with video recordings that were obtained during practice sessions (13).
Before the development of wearable IMUs, the overall physical stress imposed by a sport-related activity was estimated by the product of an athlete's rating of perceived exertion times the duration of the activity session (7). Accumulated training load over a period of one or more weeks was represented by the sum of session load values, and training monotony was defined as an individual athlete's weekly mean training load divided by the SD for the set of training intensity values (8). Because lack of variability in training intensity may interact with total training load to create a chronically fatigued state, referred to as overtraining syndrome, the term strain was used to designate the product of training load and monotony.
To date, only 2 published studies have directly related IMU data to injury risk (5,14). Both of these studies documented an association between accumulated player load and the incidence of noncontact overuse or overexertion injuries among contact sport athletes (Australian football and rugby). A series of previous investigations used rating of perceived exertion values to quantify the training load that preceded noncontact injury occurrence among contact sport athletes (9–12,26). The results of these studies have demonstrated that overtraining can induce a state of chronic fatigue that increases risk for the occurrence of overuse and overexertion injuries.
Rugby injuries caused by contact mechanisms were included with noncontact injuries in a study that failed to establish an association between RPE training load and injury incidence (19), which supports the contention of some researchers that most injuries caused by collisions are unavoidable (26,11). The possibility that exposure to a high volume of collisions may have a similar fatigue-inducing effect as a high volume of running emphasizes the potential for prevention of noncontact injuries through avoidance of excessive accumulation of inertial load (11). Conversely, a low inertial load value for a given volume of participation in sport-related activities could represent a suboptimal capacity to generate body mass accelerations that reduce risk for contact injuries. The findings of previous research strongly suggest that optimal neuromuscular function reduces risk for both noncontact and contact sprains and strains among American football players (29,30).
Movement variability (MV) is a factor that may contribute to optimal neuromuscular function (16). Accumulating evidence supports the importance of MV as an indicator of sensorimotor responsiveness to rapidly changing situational demands (24,25,27), which may reflect superior athletic performance capabilities and a lesser level of injury risk (1). Because movement of the body mass results from internal and external forces that change its inertia, IMUs may provide data that can estimate an athlete's MV during performance of sport-related activities. Although MV is typically related to changes in kinematic variables between repetitions of a cyclical movement pattern that is performed within a single test session, an individual athlete's inertial load variability (ILV) between multiple activity sessions that are completed over an extended period may have relevance to sport-related injury risk.
The purpose of this study was to retrospectively analyze archived data derived from IMUs worn by college football players during practice sessions to identify any associations of average inertial load or intraindividual ILV metrics with the occurrence of musculoskeletal sprains and strains over the course of the season. We were particularly interested in the possibility that ILV could be an important injury risk factor, which guided a comparative analysis of different metrics for its quantification.
Experimental Approach to the Problem
A cohort of 45 National Collegiate Athletic Association Division I-BCS football players (age: 20.1 ± 1.3 years; age range: 18–24 years; height: 187.4 ± 5.7 cm; weight: 104.1 ± 15.6 kg) wore an IMU (Optimeye S5; Catapult Sports USA, Chicago, IL) on the upper back throughout the duration of monitored practice sessions that were conducted over a period of 15 consecutive weeks.
Linear accelerations associated with both voluntary body movements and collisions with other players were sampled at 100 Hz, which have been shown to provide a valid and reliable representation of the magnitude and frequency of changes in body mass inertia (3,11,17). The Player Load value provided by the Catapult Sprint 5.1 software program is the square root of the sum of the squared instantaneous rate of change in acceleration in each of 3 spatial planes divided by 100 (3,17). The football program's athletic training staff supervised the acquisition of all sensor data and documented all injuries that occurred during a 17-week period from the initiation of preseason practice sessions to the end of a 12-game regular season. An injury was operationally defined as any joint sprain or muscle strain that required the attention of an athletic trainer and that limited football participation to any extent for at least 1 day. The analysis was limited to injuries most likely to result from an insufficient neuromuscular response to dynamic loading of muscles and joints. Injuries affecting the wrist, hand, or fingers were excluded, as well as fractures, contusions, lacerations, and abrasions. Injuries affecting the brachial plexus were classified as neck strains, and dislocations or subluxations of the glenohumeral joint were classified as shoulder sprains. The participants provided preparticipation consent for access to records pertaining to sport-related injuries and performance measurements. Approval for retrospective analysis of the archived data was granted by the institutional review board of the University of Tennessee at Chattanooga, which recognized precautions for protection of confidentiality of the data and did not require the acquisition of an additional informed consent document for the project.
Monitored practice sessions varied from 1 to 4 per week, with as many as 4 consecutive days of data collection and as many as 5 days elapsing between monitored sessions. The total number of practice sessions that a given player was monitored over the 15-week period ranged from 21 to 44. The football positions for the 45 players were predominantly those that require fast running speed and a high degree of agility. Among 1,980 possible values for average practice session load (i.e., 45 players X 44 monitoring sessions), data were missing for 10.4% of player sessions (206/1,980). Missing IMU values were primarily due to periods of injury-related activity restriction, but there were some instances of wireless data transmission failure. Intraindividual practice session load was averaged for the total number of sessions completed by each player, along with 3 different metrics for representation of between-session ILV for each player (i.e., SD, interquartile range, and coefficient of variation). Because a high volume of exposure to game conditions is widely recognized as a possible confounding factor in the assessment of injury risk factors, the total number of game plays during the 12-game season was documented for each member of the cohort.
Receiver operating characteristic (ROC) analysis of injured vs. uninjured status was used to identify the optimum cut point for dichotomized classification of high-risk vs. low-risk status, thereby permitting 2 × 2 cross-tabulation analysis and calculation of sensitivity, specificity, and the odds ratio (OR) for each potential injury predictor. Logistic regression analysis was used to assess the relative contributions of the binary predictor variables to the discriminatory power of a multivariable model. A Cox regression analysis was also performed to model the instantaneous probability for injury occurrence from initiation of preseason practice sessions to the last game of the regular season, represented by the hazard ratio (HR). A 90% confidence interval (90% CI) was calculated to permit assessment of OR and HR point estimate precision. A 90% CI lower limit >1.0 was interpreted as a statistically significant association.
A total of 79 upper extremity, core, and lower extremity sprains and strains were sustained by 32 of 45 individual players during 3,543 player exposures (22.3 injuries per 1,000 player exposures). A single injury was sustained by 11 of the players, 2 or 3 injuries were sustained by 16 of the players, and 5 of the players sustained between 4 and 7 injuries. For each football position, Table 1 presents injury frequency data, averaged intraindividual player load values for all monitored practice sessions, and averaged intraindividual ILV represented by 3 different metrics. The results of the ROC and 2 × 2 cross-tabulation analyses of exposure-outcome association for each potential injury predictor are presented in Table 2. A 2-factor categorical prediction model derived from logistic regression analysis demonstrated an exceptionally good fit to the data (Hosmer-Lemeshow χ 2 = 0.568; p = 0.753; Nagelkerke R 2 = 0.330). Inertial load coefficient of variation (IL CoV ≤0.15) had an adjusted OR = 6.73 (90% CI: 1.02, 44.25) and total number of game plays (plays ≥289) had an adjusted OR = 6.39 (90% CI: 1.74, 23.39). The close agreement of the adjusted ORs to those derived from the univariable analyses indicated that the 2 predictors exerted independent effects. Having either IL CoV ≤0.15 or plays ≥289, or both of the risk factors, provided strong discrimination between injured and noninjured players (χ 2 = 9.048; p = 0.004), with a good balance of sensitivity (0.78) and specificity (0.69) and an OR = 8.04 (90% CI: 2.39, 27.03). Every player who had both factors (n = 10) sustained an injury (Figure 1).
The number of days of injury avoidance vs. first injury occurrence was more strongly related to plays ≥289 (HR = 2.54; 90% CI: 1.31, 4.90) than IL CoV ≤0.15 (HR = 1.76; 90% CI: 0.94, 3.30), but adjustment for the effect of a high level of exposure to game conditions demonstrated that a low level of player ILV among monitored practice sessions had an effect on cumulative hazard across the season (Figure 2). Neither the logistic regression nor the Cox regression analyses addressed multiple injuries to the same player. The injury rate per 1,000 player exposures for players who had neither risk factor (n = 16) was 9.94 (12/1,207). The corresponding value for players who had both risk factors (n = 10) was 26.4 (22/834). For players who had only 1 of the 2 risk factors (n = 19), the injury rate was 30.0 per 1,000 player exposures (45/1,502).
The injury rate for time-loss injuries among college football players (i.e., complete activity restriction for at least 1 day) has been previously reported to be 9.6 injuries per 1,000 player exposures (18). The higher rate for the cohort in this study may be attributed to a more inclusive injury definition (i.e., participation limited to any extent for at least 1 day) and the intentional selection of players expected to have a high volume of game participation for IMU monitoring. In addition to greater exposure to game scenarios that present high risk for injury, players who are heavily relied upon for achievement of team success are exposed to a greater volume of physically demanding practice session drills and scrimmage plays. Other major challenges to the identification of trends within data provided by IMUs worn by college football players include periodically altered weekly practice schedules, differential physical demands imposed by the various playing positions, temporal variations in practice session intensity, and missing data for players who experience transient injury-related performance limitations. These considerations made the average inertial load across practice sessions the most logical performance metric to analyze as a potential predictor of injury occurrence, which we statistically adjusted for variation in player exposure to game conditions.
Variability in measurements associated with dynamic activities has traditionally been viewed as an unexplainable phenomenon that inflates measurement error, but emerging evidence supports the view of variability as an important aspect of optimal neuromuscular function. Because relatively few studies have assessed MV during the performance of sport-specific movements, no well-established standards exist for selection of the most appropriate metric or the number of repeated measures that should be acquired (25). Nonlinear methods (e.g., approximate entropy and sample entropy) have been advocated for identification of subtle temporal structure within fluctuating time series data (27), but the activity must be cyclical in nature or the time intervals between trials must be consistent. Other options for representation of variability within kinematic and kinetic data that have been used in previous research include the SD, coefficient of variation, and interquartile range. Our study has documented that each of these variability metrics demonstrates a substantial association between ILV and injury occurrence, with the IL CoV providing the strongest association.
Football presents exceptionally demanding physical challenges that require an infinite variety of neuromuscular responses for control of the adjacent segments of the kinetic chain. The availability a wide repertoire of movement and inertial loading patterns may be essential for injury prevention because they offer greater adaptability to potential joint perturbations and impending collisions. Optimal variability is probably essential to balance the mobility required for successful execution of sport skills with sufficient joint stability to avoid injury (21). Furthermore, low variability after injury may be an indication of inability to rapidly adapt to changing task demands that is not spontaneously recovered as other symptoms resolve (23). This phenomenon may relate to the occurrence of multiple injuries within a single season among many football players, which was documented for 47% (21/45) of the players in our study. A relatively low amount of variability in movement and inertial loading patterns might concentrate load distribution within the kinetic chain, which could elevate risk for joint or tendon degeneration (1). Promoting complex MV and ILV might be achieved through training and injury rehabilitation methods that use strategies for enhancement of environmental awareness and responsiveness (28). Activities that promote integration of visual, proprioceptive, and cognitive processes may be critical for development of a sufficient level of sensorimotor control to meet the demands of an individual athlete's sport and playing position (15). Our finding of a strong association between IL CoV and injury occurrence may facilitate identification of individual athletes who are likely to derive the greatest benefit from such training activities, thereby improving sport performance and reducing injury risk.
Rather than contradicting the findings of previous research that has associated a high value for accumulated inertial load with injury risk, we believe that the results of our study provide a complimentary perspective on the interpretation of data derived from IMUs worn by contact sport athletes. Accumulated training load derived from ratings of perceived exertion has been viewed as an indicator of overtraining syndrome, which has also been identified through reduced beat-to-beat variability in heart rate (2,20). The use of IMUs to monitor inertial load accumulation as a strategy for avoidance of chronic fatigue and elevated susceptibility to noncontact injury is certainly supported by the available evidence. We found that a high value for average inertial load across 15 weeks of practice sessions decreased the likelihood for occurrence of either contact or noncontact musculoskeletal sprains and strains. Players who are capable of extremely rapid acceleration, deceleration, and change of direction will generate high inertial load values, which may correspond to a level of neuromuscular responsiveness that provides protective benefits. Furthermore, a high level of ILV may be an indicator of a broad repertoire of responses to changing environmental conditions. We believe that between-session IL CoV may ultimately prove to be equally important as a guide for implementation of individualized preventive interventions for contact sport athletes.
To our knowledge, ours is the first study to investigate the potential value of data derived from IMUs for prevention of injuries resulting from participation in American football. We interpret the association of low variability in average load values with a high rate of injury occurrence as an indication of a deficiency in the ability to rapidly alter the direction of body mass movement in response to changing environmental conditions. Further research is needed to assess within-session variability of ILV and the extent to which targeted training interventions may increase ILV and reduce injury risk.
Both accumulated inertial load and low variability in average inertial load seem to be important indicators of elevated risk for injury among contact sport athletes. Data derived from IMUs may ultimately prove to be exceptionally valuable for guidance of individualized performance enhancement, injury prevention, and injury rehabilitation program design, and also confirmation of progress toward achievement of an optimal level of neuromuscular function. The findings we have reported may be used as a guide for identification of individual contact sport athletes who are most likely to derive benefit from training activities designed to promote integration of visual, proprioceptive, and cognitive processes.
1. Bartlett R, Wheat J, Robins M. Is movement variability
important for sports biomechanists? Sports Biomech 6: 224–243, 2007.
2. Baumert M, Brechtel L, Lock J, Hermsdorf M, Wolff R, Baier V, Voss A. Heart rate variability, blood pressure variability, and baroreflex sensitivity in overtrained athletes. Clin J Sport Med 16: 412–417, 2006.
3. Boyd LJ, Ball K, Aughey RJ. The reliability of MinimaxX accelerometers for measuring physical activity in Australian football. Int J Sports Physiol Perf 6: 311–321, 2011.
4. Camomilla V, Bergamini E, Fantozzi S, Vannozzi G. In-field use of wearable magneto-inertial sensors for sports performance evaluation. In: Proceedings at the 33rd International Conference on Biomechanics in Sports. Poitiers, France, 2015. Available from: http://isbs2015.sciencesconf.org/59217/document
. Accessed September 5, 2015.
5. Colby MJ, Dawson B, Heasman J, Rogalski B, Gabbett TJ. Accelerometer and GPS-derived running loads and injury risk in elite Australian footballers. J Strength Cond Res 28: 2244–2252, 2014.
6. Dellaserra CL, Gao Y, Ransdell L. Use of integrated technology in team sports: A review of opportunities, challenges, and future directions for athletes. J Strength Cond Res 28: 556–573, 2014.
7. Foster C. Monitoring training in athletes with reference to overtraining syndrome. Med Sci Sport Exerc 30: 1164–1168, 1998.
8. Foster C, Florhaug JA, Franklin J, Gottschall L, Hrovatin LA, Parker S, Doleshal P, Dodge C. A new approach to monitoring exercise training. J Strength Cond Res 15: 109–115, 2001.
9. Gabbett TJ. Influence of training and match intensity on injuries in rugby league. J Sport Sci 22: 409–417, 2004.
10. Gabbett TJ. The development and application of an injury prediction model for noncontact, soft-tissue injuries in elite collision sport athletes. J Strength Cond Res 24: 2593–2603, 2010.
11. Gabbett TJ. Quantifying the physical demands of collision sports: Does microsensor technology measure what it claims to measure?. J Strength Cond Res 27: 2319–2322, 2013.
12. Gabbett TJ, Domrow N. Relationships between training load, injury, and fitness in sub-elite collision sport athletes. J Sport Sci 25: 1507–1519, 2007.
13. Gabbett T, Jenkins D, Abernethy B. Physical collisions and injury during professional rugby league skills training. J Sci Med Sport 13: 578–583, 2010.
14. Gabbett TJ, Ullah S. Relationship between running loads and soft-tissue injury in elite team sport athletes. J Strength Cond Res 26: 953–960, 2012.
15. Grooms D, Appelbaum G, Onate J. Neuroplasticity following anterior cruciate ligament injury: A framework for visual-motor training approaches in rehabilitation. J Orthop Sports Phys Ther 45: 381–393, 2015.
16. Hadders-Algra M. Variation and variability: Key words in human motor development. Phys Ther 90: 1823–1837, 2010.
17. Holville E, Couturier A, Guilhem G, Rabita G. MinimaxX player load as an index of the center of mass displacement? A validation study. The 33rd International Conference on Biomechanics in Sports. Poitiers, France, 2015. Available from: http://isbs2015.sciencesconf.org/59234/document
. Accessed September 5, 2015.
18. Hootman JM, Dick R, Agel J. Epidemiology of collegiate injuries for 15 sports: Summary and recommendations for injury prevention
initiatives. J Athl Train 42: 311, 2007.
19. Killen NM, Gabbett TJ, Jenkins DG. Training loads and incidence of injury during the preseason in professional rugby league players. J Strength Cond Res 24: 2079–2084, 2010.
20. Kiviniemi AM, Hautala AJ, Kinnunen H, Tulppo MP. Endurance training guided individually by daily heart rate variability measurements. Eur J Apply Physiol 101: 743–751, 2007.
21. Latash ML, Scholz JP, Schöner G. Motor control strategies revealed in the structure of motor variability. Exerc Sport Sci Rev 30: 26–31, 2002.
22. Marin F, Fradet L, Lepetit K, Hansen C, Manshour KB. Inertial measurement unit
in biomechanics and sport biomechanics: Past, present future. The 33rd International Conference on Biomechanics in Sports. Poitiers, France, 2015. Available from: http://isbs2015.sciencesconf.org/71878/document
. Accessed September 25, 2015.
23. Moseley GL, Hodges PW. Reduced variability of postural strategy prevents normalization of motor changes induced by back pain: A risk factor for chronic trouble?. Behav Neurosci 120: 474–476, 2006.
24. Pollard CD, Heiderscheit BC, van Emmerik REA, Hamill J. Gender differences in lower extremity coupling variability during an unanticipated cutting maneuver. J Appl Biomech 21: 143–152, 2005.
25. Preatoni E, Hamill J, Harrison AJ, Hayes K, Van Emmerik RE, Wilson C, Rodano R. Movement variability
and skills monitoring in sports. Sports Biomech 12: 69–92, 2013.
26. Rogalski B, Dawson B, Heasman J, Gabbett TJ. Training and game loads and injury risk in elite Australian footballers. J Sci Med Sport 16: 499–503, 2013.
27. Stergiou N, Decker LM. Human movement variability
, nonlinear dynamics, and pathology: Is there a connection?. Hum Mov Sci 30: 869–888, 2011.
28. Stergiou N, Harbourne RT, Cavanaugh JT. Optimal movement variability
: A new theoretical perspective for neurologic physical therapy. J Neurol Phys Ther 30: 120–129, 2006.
29. Wilkerson GB, Colston MA. A refined prediction model for core and lower extremity sprains and strains among collegiate football players. J Athl Train 50: 643–650, 2015.
30. Wilkerson GB, Giles JL, Seibel DK. Prediction of core and lower extremity strains and sprains in collegiate football players: A preliminary study. J Athl Train 47: 264–272, 2012.