Differences in lower extremity coordination patterns as a function of sports specialization

Abstract

The practice of early sport specialization, defined as intense year-round training in a single sport at the exclusion of others, is increasing in youth athletics. Despite potential benefits, sport specialization may be detrimental to the health of young athletes, as specialization may increase the risk of musculoskeletal injuries—particularly overuse injuries. However, there remains limited knowledge about how sports specialization uniquely alters underlying sports-related motor behavior. The purpose of this study was to compare the variability of movement patterns exhibited by highly sports specialized youth athletes to that of nonspecialized athletes during performance of a sport-specific, virtual reality based cutting task. It was hypothesized that highly specialized athletes would display different patterns of movement coordination compared to nonspecialized athletes during both the run-up phase and cut-and-decelerate phase. In support of the hypothesis, specialized athletes exhibited both intra- and inter-limb coordination that were significantly different than unspecialized athletes. Overall, the results indicate that the highly specialized athletes tended to exhibit greater degrees of coordination but also the ability to break the coordinated patterns of joint angle changes to execute a cutting maneuver, which requires asymmetric demands on the lower extremities while planting on one leg and changing direction.

In the United States, approximately 30 to 45 million young athletes (ages 6 to 18 years) participate in organized and recreational activities (Brenner, 2007). As the number of participants in youth athletics continues to increase, so too does the practice of early sport specialization (Wiersma, 2000), a term commonly defined as intense year-round training in a single sport at the exclusion of other sports (N. A. Jayanthi, Pinkham, Dugas, Patrick, & LaBella, 2013; N. A. Jayanthi, Pinkham, Durazo-Arivu, Dugas, & Luke, 2011; Malina, 2010). The increasing trend of specialization is accompanied by a general decrease in the amount of time that youth spend in unstructured physical activity (i.e., free play) (Sturm, 2005). In addition, athletes may be encouraged to specialize at younger ages with the belief that early sport focus can help them achieve elite levels of performance and obtain the related benefits that accompany elite athletes (e.g., academic scholarships, professional contracts, and positive feedback from peers, parents, and coaches) (Gould, 2009; LaPrade et al., 2016a; Malina, 2010). Despite potential benefits, sport specialization can also be detrimental to both the physical and mental health of young athletes, as specialization may result in increased psychological stress, social isolation, and burnout in young athletes (Giusti et al., 2020; Myer et al., 2015). Sport specialization may shunt the development of diverse movement competency and thus increase the risk of musculoskeletal injuries (Bell, Post, Biese, Bay, & McLeod, 2018; Post et al., 2017)—particularly overuse injuries (Bell, Post, Biese, et al., 2018; Brenner, 2007; DiFiori et al., 2014; Hecimovich, 2004; Post et al., 2017). However, there remains limited knowledge about how sports specialization uniquely alters underlying sports-related motor behavior.

Sports Specialization and Overuse Injuries

Sports specialization entails that youth athletes partake year-round in intensive, highly technical training (Brenner, 2007; DiFiori et al., 2014) that has been associated with an increased risk of overuse injury (Bell et al., 2016; Hall, Foss, Hewett, & Myer, 2015; N Jayanthi, LaBella, Fischer, Pasulka, & Dugas, 2015; McGuine et al., 2017). This specialized training—notably involving the homogeneous movements of training regiments that repeatedly target musculoskeletal tissues through a small set of sports-specific skills or activities—maladaptively stresses the neuromuscular development of youth athletes (Hamill, Palmer, & Van Emmerik, 2012). This repetitive utilization of particular movements to refine motor skills, combined with the focalized stress placed on the developing bone and connective tissue of youth athletes, may lead to greater fatigue, movement dysfunction, and/or injury in these athletes (Abrams, Renstrom, & Safran, 2012; Olsen, Fleisig, Dun, Loftice, & Andrews, 2006; Sheppard, Nicknair, & Goetschius, 2020; Sweeney, Howell, Seehusen, Tilley, & Casey, 2020). Alternatively, youth who participate in multiple sports (i.e., nonspecialized athletes) exhibit greater and more diverse strength, speed, endurance, and gross motor control compared to specialized athletes (Fransen et al., 2012), and these benefits potentially carry over into adulthood for adolescent multisport athletes (Confino, Irvine, O’Connor, Ahmad, & Lynch, 2019; Herman et al., 2019). Discovering how motor control and biomechanics are altered because of sports specialization can then be used to develop interventions aimed at reducing the deleterious effects of specialization in athletes who choose to specialize at an early age. Analyses of movement pattern variability (Davids, Bennett, & Newell, 2006; Lockhart & Stergiou, 2013; Newell & Slifkin, 1998; Riley & Turvey, 2002; Stergiou & Decker, 2011; van Emmerik & van Wegen, 2000) have been previously used to characterize a variety of movement deficits (Hamill et al., 2012; Howell, Bonnette, Diekfuss, Grooms, Myer, & Meehan III, 2020; Li et al., 2019), and may likewise be useful for understanding sports-related motor alterations associated with specialization.

Cutting Tasks and Injury Risk

Biomechanical analyses of cutting or change of direction tasks are one of the most commonly utilized indicators of lower extremity injury risk (Almonroeder, Garcia, & Kurt, 2015; Fox, 2018). Neuromuscular control of the lower extremity during cutting movements requires intricate coordination of multiple joints and body segments (e.g., thigh and shank). The exact technique used to perform a cut arises from a combination of biomechanical, neuromuscular, and environmental factors (Dowling, Corazza, Chaudhari, & Andriacchi, 2010; Meinerz, Malloy, Geiser, & Kipp, 2015; Sigward & Powers, 2006) that constrain an athlete’s execution of the behavior. Moreover, exercise-based injury prevention programs can effectively alter cutting biomechanics in order to reduce lower extremity injury risks (Pappas et al., 2015); however, it is unknown how sports specialization influences frequently performed sports movements such as change of direction movements (i.e., cutting tasks). Quantifying how lower extremity coordination emerges during a cutting task as a function of sports specialization is an important step in understanding how specialization influences the coordination during sports movements—especially in ecologically valid athletic environments that are beyond the methodological reach of most laboratory-based testing protocols.

Capturing Biomechanical Variability in Virtual Reality

One difficulty in quantifying athletes’ motor variability in the laboratory is the lack of realistic tests or scenarios that mimic “on-field” competitive behavior (DiCesare, Kiefer, Bonnette, & Myer, 2019). This limits the understanding of how specialized, sports-specific training affects movement patterns in athletic settings for which the training was intended. However, a recent trend towards increasing affordability and ease of use of virtual reality (VR) equipment has provided an opportunity to evaluate the biomechanics of specialized athletes in simulated sport-specific environments (Diekfuss et al., 2020). Recent research has demonstrated that VR may be an optimal way to maintain precise control over experimental conditions (such as the timing of a stimulus or scaling task difficulty) while presenting participants with realistic, sport-specific scenarios that evoke movement profiles similar to those seen during competitive play (Diekfuss et al., 2020). Our recent work demonstrated that a VR-based jump-landing task (i.e., jumping to head a virtual corner kick) elicits distinct movement patterns and landing biomechanics in soccer athletes than sterile and unrepresentative standard biomechanical tests such as the drop vertical jump (DVJ) (DiCesare, Kiefer, et al., 2019). The differences in the biomechanical profiles elicited by the VR compared to the “sterile,” non-contextual DVJ task may arise from numerous factors such as a greater sense of goal-directedness or an altered focus of attention for athletes in the more realistic VR task (Diekfuss et al., 2020; Wulf, McNevin, & Shea, 2001).

Physiological and Behavioral Variability: Health, Performance, and Sports Training

Certain time-dependent patterns of physiological and behavioral (i.e., motor) variability can indicate the presence as well as the severity of a pathological state (Stergiou & Decker, 2011; Stergiou, Harbourne, & Cavanaugh, 2006). For example, changes in the structure of participants’ gait variability following a concussion are associated with an increased risk of future musculoskeletal injuries (Howell, Bonnette, Diekfuss, Grooms, Myer, Wilson, et al., 2020), and postural sway variability patterns are correlated with low back pain ratings and predict the occurrence of future falls in elderly people (Johansson, Nordström, Gustafson, Westling, & Nordström, 2017; Ruhe, Fejer, & Walker, 2011). While a vast number of analysis techniques exist to quantify various aspects of physiological and behavioral variability, cross recurrence quantification analysis (CRQA) is a general-purpose, bivariate, (nonlinear) correlational-based technique that quantifies the organizational structure within time series data (Wallot & Leonardi, 2018; Webber Jr, Marwan, Facchini, & Giuliani, 2009) (e.g., an athlete’s variation in the stride-to-stride coordination of the feet during gait). Due to CRQA’s lack of assumptions about the underlying statistical properties of data and its effectiveness in the case of nonlinearity, nonstationarity, noise, and time series outliers (Fusaroli, Konvalinka, & Wallot, 2014; Marwan, Romano, Thiel, & Kurths, 2007; Webber Jr & Zbilut, 2005), it is ideally situated to quantify dynamic movements performed during the current study’s under-constrained, sports-specific scenario described below.

Current Study

The purpose of this study was to compare the variability of movement patterns exhibited by highly sports specialized youth athletes to that of nonspecialized athletes during performance of a sport-specific, VR-based cutting task. A soccer-specific VR cutting scenario—designed to elicit similar motor activity to that displayed in real competitive environments—was designed to mimic the on-field experience of high intensity movement in response to a soccer ball pass occurring during a real-life soccer competition. We focused on patterns of lower-extremity movement coordination, as quantified using CRQA. We separately analyzed the run-up phase (moving forward to the point in space where the cut was required, as described later in the Procedure) and the cut-and-decelerate phase. We included the run-up phase because a cutting maneuver is more than the biomechanical event of planting the foot and changing direction—the run-up prior to the cut is also important for planning and controlling the moment of the cut as the external events precipitating the cut unfold (e.g., the pass of the ball to an NPC), and therefore specialization-related changes in motor planning and execution could be apparent during this phase. In both the run-up and cut-and-decelerate phases, we focused on the interlimb coordination of the left and right frontal plane knee angles and on the intralimb coordination between ipsilateral hip and knee joint sagittal plane angles of the plant leg. The degree of coordination between left and right knee frontal plane angles was of interest because of the known relationship between increases in frontal plane knee angles and dynamic knee valgus and resultant increased lower extremity injury risk (Dempsey, Lloyd, Elliott, Steele, & Munro, 2009; Hewett et al., 2005; McLean, Huang, & Van Den Bogert, 2005). Likewise, during the cut-and-decelerate phase, we focused on intralimb coordination between ipsilateral hip and knee joint sagittal plane angles of the plant leg, which was of interest because of its relation to lower extremity injury risk during sports movement (Paterno et al., 2015; Weir, van Emmerik, Jewell, & Hamill, 2019). We hypothesized that highly specialized athletes would display different patterns of movement coordination compared to nonspecialized athletes during both the run-up phase and cut-and-decelerate phase.

Method

Participants

Ninety-one adolescent female soccer athletes participated in the current study. This data set was a subset of a larger dataset from an ongoing project (baseline testing) investigating an ACL injury risk reduction intervention (NIH: 1U01AR067997–01A1). The current study was approved by an institutional review board and each participant, as well as their legal guardian, provided informed, written consent and assent prior to data collection.

Sports Specialization Survey

Participants were asked to complete a short survey in order to determine each athlete’s level of specialization (N Jayanthi et al., 2015). The survey consisted of the following three questions:

1. Could you pick a main sport that you participate in?

2. Did you quit other sports to focus on a main sport?

3. Do you train for a single sport greater than or equal to 8 months a year?

The number of ‘yes’ responses to these questions was used to determine the degree of an athlete’s specialization: 0 or 1 ‘yes’ answer was assigned as nonspecialized athlete, 2 ‘yes’ answers resulted in moderate specialization assignment, and 3 ‘yes’ answers were classified as highly specialized athletes. The moderate specialization participants (n = 22) were not included in the current analysis. This survey was chosen to determine the level of sports specialization in the current study as it is a relatively new area of research interest and there is no widespread agreed on definition of what constitutes sports specialization (Neeru Jayanthi et al., 2020; N Jayanthi et al., 2015). Specifically it was chosen for the 3 key components of this definition mentioned above—choosing a main sport (question #1), quitting other sports (question #2), and year-round training (question #3)—have become the basis for one of the most commonly used methods of classifying specialization. This survey has been utilized in numerous past investigations (Bell et al., 2016; Neeru Jayanthi et al., 2020; N Jayanthi et al., 2015; N. A. Jayanthi, Post, Laury, & Fabricant, 2019; McGuine et al., 2017) and it has been shown to differentiate the effects of single sport specialization training related to risk for musculoskeletal injury, when accounting for age and total time spent in sports related activity (N Jayanthi et al., 2015).

Participants were classified as belonging to either the non-specialized or highly specialized group depending on their self-reported answers to the sports specialization survey described above. Four participants had incomplete data (i.e., missing/poorly recorded motion capture data) and were excluded from analyses. Due to a significant difference in age (p < .05) between the groups, all nonspecialized and highly specialized participants over the age of 16.0 years were removed (total removed, n = 16; nonspecialized athlete removed, n = 1; highly specialized athletes removed, n = 15). Therefore, the final sample was composed of 16 nonspecialized participants (age = 14.65 ± 1.19 yrs; height = 160.56 ± 7.16 cm; and weight = 54.65 ± 9.30 kg) and 33 highly specialized participants (age = 15.00 ± 0.53 yrs; height = 162.61 ± 5.11 cm; and weight = 56.77 ± 6.92 kg). There were no significant differences in age (t[47] = 1.43, p = .16), height (t[47] = 1.15, p = .26), or weight (t[47] = 0.90, p = .37) between the groups after removal of participants over the age of 16. All participants were healthy and had no recent history of musculoskeletal injury that excluded them from participation in the larger ACL injury risk reduction clinical trial.

Procedure

Virtual Reality Task

Each participant was outfitted with 45 retroreflective markers, distributed across all major body segments at the following anatomical landmarks: four markers on the head (i.e., on the VR head-mounted display [HMD]), sternum, sacrum, offset, and bilaterally at the acromioclavicular joint, elbow, mid-wrist, medial and lateral hands, anterior superior iliac spine, greater trochanter, mid-thigh, medial and lateral femoral condyles, tibial tuberosity, lateral shank, distal shank, medial and lateral malleoli, heel, toe, lateral foot, and posterior foot. Participants also wore a Vive Pro HMD (HTC Corporation; Taoyuan City, Taiwan) which was used to present the virtual cutting scenario. A 44-camera passive motion capture system (Motion Analysis Corp, Santa Rosa, CA) was used to capture the motion of the reflective markers during each trial. Three-dimensional marker trajectories were captured at 120 Hz and were streamed into Unity3d (Unity Technologies, San Francisco, CA) for presentation of a participant’s full-body avatar in the virtual reality environment.

Prior to the cutting task, participants first underwent a brief introduction to familiarize themselves with the layout of the virtual environment—specifically, the boundaries of the virtual space were emphasized to ensure participants stayed within the physical constraints of the real environment (i.e., the motion capture space). After this period of familiarization, participants engaged in the sports specific virtual cutting task (see Figure 1). The virtual cutting task required participants to defend against three non-player characters (NPCs). The task was designed to elicit natural cutting moves from the participants following the pass of a virtual soccer ball from a central, offensive NPC. The cutting task began with three NPCs simultaneously approaching the participant (as indicated by the yellow dashed lines in Figure 1A and B). The central offensive NPC was directly in front of the participant with an additional NPC flanking the left and right side each of the central NPC. The participant was required to run directly towards and meet the central offensive NPC (as noted by the black dashed line in Figure 1B). The central NPC would then randomly turn to the left or right and pass the virtual soccer ball to the appropriate NPC (see Figure 1B for an example where the orange dashed line indicates the pass of the ball to the left NPC). The participant was then required to cut in the direction of the passed soccer ball (as shown by the white dashed line in Figure 1B). The participant was not informed of which NPC the ball would be passed to and the passing direction (left vs. right) was presented at random. Additionally, no directions were given on how to perform the cutting movement beyond what had been described. The cutting task ended once two cuts had been successfully accomplished on both the right and left sides. A successful trial was operationalized as the participant completing the run-up to the central NPC and making a cutting movement towards the NPC that the ball was passed to.

Figure 1.

Panel A displays what participants viewed during the cutting scenario. The grey dashed lines in panel A indicate the path the non-player characters (NPC) took with the central NPC passing the ball to the left or right NPC. Panel B is a top-down view of the cutting scenario. The black dashed line going through the player-controlled avatar (i.e., the avatar at the top of panel B) indicates the run-up phase with the white dashed line coming directly off that line towards the soccer ball representing the cutting towards the NPC. The grey dashed line going from the middle NPC to the left indicates the ball path.

Data Processing and Analysis

Raw marker trajectories were first low-pass filtered at a cutoff frequency of 12 Hz using a fourth-order Butterworth filter. Lower extremity joint angles were then calculated using a 6 degree-of-freedom skeletal model in Visual 3D (C-Motion Inc, Germantown, MD). Using a custom written MATLAB script (MathWorks, Inc, Natick, MA), each cutting trial was automatically segmented into 2 separate sections: 1) the run-up phase to the cut and 2) the cut-and-decelerate phase. The run-up section was defined as the time between the onset of participants’ anterior-posterior (AP) center of mass (CoM) acceleration up to the onset of their medial-lateral (ML) CoM acceleration that marked the beginning of the cut and the end of the run-up period. The beginning of the cut-and-decelerate phase was defined as the beginning of participants ML CoM acceleration to when participants completed both AP and ML CoM deceleration (i.e., no motion). This procedure resulted, on average, in 4.5–7.0 s (approximately 540–840 data points in total) of data per trial: 2–4 s for the run up to the cut (240–480 data points) and approximately 2.5–3 s to cut and decelerate (300–360 data points). Data collected before the participants initiated movement and following complete deceleration after the cut (i.e., participants were no longer moving) were not included in the subsequent analysis. No filtering beyond the initial filtering of marker trajectories or down-sampling methods were performed. After the data were processed, the time series were subjected to further analysis (i.e., CRQA) in order to investigate the coordination dynamics between the right knee and hip sagittal plane angles during cuts made to the left (two trials), the left knee and hip sagittal plane angles during cuts made to the right (two trials), and the right and left knee frontal knee angles during the run-up phase for both right and left cuts (four trials total including two left cuts and two right cuts).

Cross recurrence quantification analysis

CRQA is an extension of recurrence quantification analysis and involves the quantification of cross-recurrence plots. Cross-recurrence plots are two dimensional representations of “recurrences” (shared locations) between trajectories (e.g., of left and right knee frontal angles for analysis of the run-up phase) within a reconstructed phase space. The analysis has been used to index the coordinative patterns between human physiological signals across a variety of behaviors including gait and postural sway following a concussion (Bonnette et al., 2020; Howell, Bonnette, Diekfuss, Grooms, Myer, & Meehan III, 2020; Howell, Bonnette, Diekfuss, Grooms, Myer, Wilson, et al., 2020). Several parameters must be selected for CRQA prior to the analysis—the radius, the delay, and the embedding dimension. In the current analysis, a variable radius value was utilized to ensure each recurrence plot maintained a fixed recurrence rate of 10.0%. The values for the delay and the embedding dimension parameters were chosen using the average mutual information (Roulston, 1999) approach and global false nearest neighbors analysis (Kennel & Abarbanel, 2002), respectively. A mean delay of 26 samples and a mean embedding dimension of 3 were selected for the analysis. The distance matrix was rescaled using the maximum normalized distance. The required minimum line length for vertical and diagonal lines was set to two consecutive points. CRQA was performed using a combination of open source functions(Marwan & Kurths, 2002; Marwan, Wessel, Meyerfeldt, Schirdewan, & Kurths, 2002) and custom-written MATLAB scripts (MathWorks, Inc, Natick, MA). Four CRQA dependent variables were returned with two variables each describing the structure of diagonal and vertical lines.

The two reported diagonal line measures were percent determinism and average diagonal line length. Percent determinism indexes the percentage of recurrent points that make up diagonal lines in the cross-recurrence plot. A higher percentage of recurrent points that fall on diagonal lines indicates greater predictability in the coupling of the two signals (i.e., more predictable and regular patterns of coordination) (Shockley, Butwill, Zbilut, & Webber Jr, 2002; Webber Jr et al., 2009). Average diagonal line length indicates the average number of recurrent points that make up the diagonal lines of a recurrence plot. Longer average line lengths indicate that the two signals spent, on average, more time in the same regions of reconstructed phase space and may be an indication of stronger coupling (i.e., more stable coordination) between two time series (Shockley et al., 2002; Webber Jr et al., 2009). The two vertical line measures were laminarity and trapping time. Laminarity indicates the percentage of recurrent points that constitute vertical lines. While the diagonal line measures indicate the level of coordination (coupling) between two time series, the vertical line measures index how often the signals exhibit laminar behavior—a type of activity where one signal remains relatively fixed while the other varies about that value (Bonnette et al., 2018; Davis, Pinto, & Kiefer, 2017; Marwan et al., 2007). Trapping time is the average length of recurrent points that make up vertical lines and provides an indication of the average amount of laminar behavior.

Statistical Analysis

Potential outliers in the dependent measures were identified by any value ± 2.698 SD from the median. If an outlier was statistically identified, the original time series was visually inspected for artifacts. If artifacts were detected (e.g., participants cutting too soon or starting to run then stopping before the cut), that specific trial’s dependent variables were replaced by the median values of the dependent variables for that condition. Following this procedure, less than 1.0 % of all data points were replaced (4 of 1224 total values). For the right and left frontal knee angle coordination comparison the dependent measures were averaged across all four cutting trials to result in one averaged value per dependent measure for each participant. Cutting entails planting on the leg contralateral to the cut direction. Therefore, for the two left cut trials the right hip and knee sagittal plane data were averaged to result in a single value for each dependent variable of the left cuts for each participant. Likewise, the left hip and knee sagittal plane data were averaged for the two right cut trials to result in a single value for each dependent variable of the right cuts for each participant. To evaluate differences in knee coordination patterns between groups, a series of independent samples t-tests with a Welch’s correction were performed. Statistical significance was set at α < 0.05 and all tests were two-sided. Statistical analyses were performed in JASP (Team, 2018).

Results

Run-Up Phase

There were no significant differences in biomechanical coordination between the two groups of athletes during the run-up to the cut (all p > .05). This includes no differences in the left knee and hip sagittal plane angle coordination (right-cut trials), nor differences between the groups in the right knee and hip sagittal plane angle coordination (left-cut trials). Lastly, there were no significant differences between the groups of athletes in left and right knee frontal plane angle coordination (all p > .05).

Cut-and-Decelerate Phase

There were significant differences in trapping time values between the two groups of athletes for the left (t[29.07] = 2.07, p = .047, d = 0.63) and right (t[45.78] = 2.45, p = .018, d = 0.64) knee and hip sagittal plane angle coordination during the right and left cuts, respectively. For the right leg (left cut), the highly specialized group (M = 16.42, SD = 2.72) had greater trapping time values than the nonspecialized group (M = 14.67, SD = 2.80). Likewise, for the left leg (right cut) the trapping time values were greater for the highly specialized (M = 15.66, SD = 3.56) group compared to the nonspecialized group (M = 13.93, SD = 1.37).

For the right leg only (left cut), there was also a significant difference in average diagonal line length, t(39.43) = 2.88, p = .006, d = 0.83, between the non- and highly specialized groups. Like the trapping time results, the highly specialized group had greater average diagonal line length values (M = 41.19, SD = 10.82) than the nonspecialized group (M = 33.32, SD = 7.90). There was no difference between groups in mean diagonal line length for the left leg (right cut; p > .05) or group differences for percent determinism and laminarity for any comparison (all p > .05).

For the left and right knee frontal plane angle there was one significant difference between the groups. Trapping time was significantly greater, t(45.20) = 2.69, p = .01, d = 0.69, in the high specialization group (M = 11.72, SD = 2.37) than in the nonspecialized group (M = 10.51, SD = 0.81). There were no differences between the groups for percent determinism, laminarity, and average diagonal line length (all p > .05).

Discussion

To the authors’ knowledge, this study is one of the first investigations to provide evidence that sports specialization affects motor control during a sport-specific VR scenario. Consistent with our hypothesis, there were group differences between nonspecialized and highly specialized athletes in the coordination of their hip and knee sagittal plane angles of each leg and of their left and right knee frontal plane angles during the cut-and decelerate phase. Specifically, the highly specialized athletes exhibited greater trapping time values in the coordination of the hip and knee sagittal plane angles in both cutting directions (i.e., from left and right legs) and in the coordination of the knee frontal plane angles during the averaged four cuts. The highly specialized athletes also exhibited greater mean diagonal line length values in the coordination of the right hip and knee sagittal plane angles during the left cut. In contrast to our hypothesis, there were no differences associated with specialization between the two groups during the run-up phase.

Taken together the results indicate that the highly specialized athletes demonstrated distinct lower extremity coordination strategies from the nonspecialized athletes during a dynamic sport-specific cutting scenario. Specifically, the highly specialized group exhibited greater mean diagonal line length values for right knee and hip sagittal plane coordination during the cut-and-decelerate phase for the left cut than the nonspecialized group. This indicates that the highly specialized athletes were able to produce similar right knee and hip sagittal angle movements (i.e., the hip and knee were moving in a similar manner) longer than the nonspecialized athletes during the cut-and-decelerate phase. The trapping time results revealed that during the cut-and-decelerate phase the highly specialized athletes furthermore exhibited the ability to effectively dissociate sagittal plane knee and hip angles and frontal plane knee angles—effectively demonstrating the ability to occasionally interrupt coordinated patterns of joint angle changes. This behavior may allow the specialized athletes to respond more successfully and efficiently to the cutting scenario through more dynamic, independent movement of joint angles. Overall, the results indicate that the highly specialized athletes tended to exhibit greater degrees of coordination but also the ability to intentionally break the coordinated patterns of joint angle changes to execute a cutting maneuver, which requires asymmetric demands on the lower extremities while planting on one leg and changing direction.

Overall, the reported coordination differences between the nonspecialized and highly specialized athletes in their lower extremity movement patterns may arise from two different scenarios (or a combination of both). The first explanation is that highly specialized athletes may possess an inherent physical ability to produce the coordinative patterns expressed during the sport-specific scenario, and this led them to specialize in the sport where those talents are critical to success. The second is that the coordinative patterns expressed by the highly specialized athletes were a result of the training regimens that repeatedly target motor behavior through a small set of sports-specific skills or activities (i.e., sports specialization). While it is beyond the scope of this report to determine if either scenario is correct, it does provide evidence that specialized athletes express different patterns of lower extremity coordination than nonspecialized athletes during the sport-specific VR cutting scenario.

Combined with past evidence of increased overuse injury risk (Bell et al., 2016; Hall et al., 2015; N Jayanthi et al., 2015; McGuine et al., 2017) it is possible that the coordinative patterns express by the highly specialized athletes may contribute to increased risk of injury, as cutting or sudden deceleration with a cut have been identified as one of the most common mechanisms for lower extremity injuries in soccer (Alentorn-Geli et al., 2009; Brophy, Stepan, Silvers, & Mandelbaum, 2015; Faude, Junge, Kindermann, & Dvorak, 2005; Sheppard et al., 2020; Waldén et al., 2015). Future investigations should prospectively investigate the current reported coordination metrics related to sports specialization in order to effectively establish the link between specialization and future musculoskeletal injury risk. While the current results cannot be directly associated with injury risk or outcomes, the difference in coordination patterns supports past evidence showing that differential sports training and specialization affect motor control and biomechanics (Cowley, Ford, Myer, Kernozek, & Hewett, 2006; DiCesare, Montalvo, et al., 2019; Herman et al., 2019; Smith et al., 2007) and is related to increases in self-reported symptoms and dysfunction during activities of daily living (Sheppard et al., 2020). Additionally, a generalized 10-week neuromuscular training program resulted in changes in the dynamics of muscle activation as measured by a univariate analysis similar to CRQA (Kiefer & Myer, 2015) (recurrence quantification analysis [RQA]). Specifically, participants exhibited significant reductions in RQA dependent measures (related to the dependent variables reported in the current manuscript) compared to baseline levels and to control participants after completing the training. In order to reduce the possible elevated injury risk faced by highly specialized athletes, future investigations should also evaluate the prophylactic effect of general neuromuscular training programs in altering the movement patterns of specialized athletes to a pattern that is similar to nonspecialized athletes (i.e. a decrease in the CRQA dependent measures).

Clinical Implications for Virtual Reality Testing and Analysis

While specialization is not inherently detrimental, more targeted approaches are needed to determine when specializing in a sport “too early” as determined through any combination of a number of factors (e.g., prior to sufficient musculoskeletal system development, motor skill development, and/or psychological readiness to commit to a single sport, etc. (Feeley, Agel, & LaPrade, 2016; LaPrade et al., 2016b; Malina, 2010)). The results of the present study may support the use of biomechanical or coordinative criteria, in addition to other possible recommendations (e.g., age, activity level, etc. (N. A. Jayanthi et al., 2019)) to establish when it is appropriate for young athletes to specialize in sport. Establishing objective movement-related criteria is especially important because many coaches and parents are unaware of the inherent increased risk of musculoskeletal injury in specializing too soon (Bell et al., 2020; Post et al., 2020). The use of more targeted criteria (as opposed to simply using age as a barometer for safe sport specialization, for example) may signal when, from a motor perspective, young athletes may be better able to accommodate a higher or less heterogeneous training load.

Another consideration that may be taken from the current work is the potential to utilize VR testing scenarios to identify athletes at risk for overspecialization and therefore sports related injuries (N Jayanthi et al., 2015; Myer et al., 2015, 2016). Moreover, sports related VR training scenarios could improve movement variability and potentially decrease injury risk. This could be used in combination with traditional on-field training in order to increase injury resilience through learned prophylactic biomechanical movement patterns. Lastly, for athletes undergoing injury rehabilitation or younger athletes who are developing neuromuscular control, sports related VR may be a safe alternative to scenarios associated with the undesirable risks of live sports, such as contact with other athletes. To this end, VR equipment and software have become drastically more accessible to everyday consumers, including clinicians, looking to improve athlete readiness for sport. Current consumer-grade VR systems can be purchased for under a thousand dollars and are accompanied by a range of software options that address sensorimotor behavior, cognitive abilities, and overall fitness. In combination with recent advances in VR accessibility, the current study highlights the potential to employ VR training strategies to augment traditional training methods. Especially for athletes who are not ready to return to full contact sports exposure after an injury.

Limitations

It should be recognized that there is a strong relationship between age and sports specialization. This may be particularly relevant in the current data set where participants were recruited from a major US metropolitan area and it may be assumed that varsity team sports require greater specialization (Bell, Post, Trigsted, et al., 2018)—due to competition—than in lower populated rural areas, which are not represented in the current study. The competitiveness (i.e., the number of available roster spots for a varsity team versus the number of athletes total) of the surrounding area may have contributed to the high levels of specialization after the age of 15. In fact, in the current study only one nonspecialized participant was over the age of 16 compared to 15 specialized participants. The current study’s design is unable to address this relationship. Future longitudinal studies may help explain the relation between age and specialization and possibly identify when sports specialization may begin to detrimentally affect motor coordination.

The VR avatars (the NPCs) used in the current study were not ultra-realistic (see Figure 1A); however, our previous research has shown that the movements of the ball and NPCs do evoke appropriate motor responses from the participants (Kiefer et al., 2017). The authors expect that the motor responses—cutting towards the direction of a randomly passed ball—would be similar to live “in-real-life” cutting scenarios and that the findings of the study are generalizable beyond laboratory testing. Nonetheless, future research should determine the appropriate level of realism in virtual environments—especially related to dynamic sports scenarios—that creates immersion and elicits appropriate behavior by participants.

Conclusions

By utilizing a sports-specific VR cutting scenario, we found that sports specialized athletes exhibited altered lower extremity coordination patterns compared to their nonspecialized athletic peers. Specifically, the highly specialized athletes exhibited longer periods of dissociated coordination (i.e., the angles did not change in a similar manner at the same time) of the hip and knee sagittal plane angles and in the knee frontal plane angles. The highly specialized athletes also exhibited longer periods of coordination of the right hip and knee sagittal plane angles (i.e., the angles moved in a similar manner for a longer amount of time). The differential coordination pattern exhibited during the cuts by the high specialization group may permit the athletes to respond more effectively—in terms of the sports-related appropriateness of the movement—to the cutting scenario than the nonspecialized athletes. However, this increased ability to respond with more adaptive movement coordination patterns may place the athletes at greater risk for musculoskeletal injuries.