Energy flow in men’s javelin throw and its relationship to joint load and performance

Participants

Ten right-handed javelin throwers (body height: 189.2 ± 7.2 cm; body mass: 92.4 ± 9.3 kg; age: 21.8 ± 3.6 years; personal best: 78.23 ± 11.38 m) of the German (junior-) national team participated in the study. All subjects were free from injuries at the time of the investigation. The study was approved by the Leipzig University ethics committee (ethical approval nr: 462/18-EK) and performed according to the declaration of Helsinki. All participants gave written informed consent to participate in the study prior to the investigations.

Material and experimental protocol

The three-dimensional position data of the markers attached to the skin of the athletes were recorded at 300 Hz using 12 infrared cameras (Qualisys AB, Gothenburg, Sweden). In addition, two perpendicular video cameras (Qualisys AB, Gothenburg, Sweden) recorded the throws at 150 Hz. The camera system was set up in an oval around the last 10 m before and 2 m behind the foul line, with half of the infrared cameras on each side of the approach. The video cameras were positioned in such a way that one camera was orthogonal to the approach at the release point about two meters before the foul line. The second video camera stood 10 m away from the foul line, recording the release from the back. The average residual of the calibration was 0.75 millimeter.

To record the movements of the thrower’s torso and upper extremities, 16 markers (metacarpophalangeal joint of the 2^nd and 5^th finger; ulnar and radial styloid; lateral and medial epicondyle of the humerus; left and right acromion; 7^th cervical vertebrae and 12^th thoracic vertebrae; processus xiphoideus; incisura jugularis; left and right spina iliaca anterior superior; left and right spina iliaca posterior superior) and two clusters (upper arm, forearm) were placed on each subject. Five markers were attached to the javelin (GETRA Kinetic, 800 g, 70 m), which was modified for indoor use by replacing the sharp metal with a dull carbon tip. The investigations were carried out indoors, with the athletes throwing into a net.

After an individual warm-up, each participant threw at least three trials using their preferred approach (mean approach speed: 5.05 ± 0.62 ms⁻¹). The best three throws were selected for further analysis based on the highest release speed of the javelin (v₀).

Data processing

Prior to further data analysis three critical events were identified visually from the recorded videos: (1) touchdown of the rear leg, (2) touchdown of the bracing leg and (3) release of the javelin. Afterwards, marker trajectories were filtered using a fourth order, zero-lag Butterworth filter. The cut-off frequencies (10–13 Hz) for each marker were determined via residual analysis (Winter, 2009).

To calculate the kinematics and kinetics, a six-segment model (javelin, right hand, forearm, upper arm, thorax, abdomen) was built using Visual 3D (Ver. 2020.11.2; C-motion, Germantown, USA). The shoulder joint center was determined using functional methods implemented into Visual 3D (Schwartz & Rozumalski, 2005; Howenstein, Kipp & Sabick, 2019). The joint centers of the elbow and the wrist were respectively determined as the midpoints between the medial and lateral humeral epicondyles and the midpoints between the ulnar and radial styloid (Wu et al., 2005). In dynamic movements the pose of each segment was estimated using the six degrees of freedom algorithm. The kinetics were calculated using Newton-Euler equations of motion implemented in Visual 3D. All derived net joint forces (NJF) and torques (NJT) were calculated as internal torques and forces. For the inverse dynamics calculations, body segments inertial parameters reported by De Leva (1996) were used. The BSIP of the javelin were estimated using a torsion pendulum and a reaction board (Sommerfeld, 1950).

Using the calculated kinematics and kinetics, a segment power (SP) analysis was carried out for all segments of the different joints(shoulder, elbow, wrist) of the throwing arm using MATLAB (Ver. 9.11.0; The Mathworks Inc., Natick, MA, USA). The joint force power (JFP) and the segment torque power (STP) for the proximal and the distal segment of a joint were calculated as:

${\displaystyle \mathrm{JFP}={\mathit{F}}_{\mathit{ij}}\cdot {\mathit{v}}_{\mathit{j}},}$ ${\displaystyle STP={\mathit{T}}_{\mathit{ij}}\cdot {\dot{\theta }}_{ij}}$ where F_ij is the NJF vector and T_ij the NJT vector acting on the i^th segment of the j^th joint. v_j is the linear velocity vector of the j^th joint, $\dot{\theta }$ represents the angular velocity vector of the i^th segment at the j^th joint. The NJF as well as the NJT vector of both segments connected by a joint are equal in magnitude and opposite in direction. While the linear velocity is the same at the joint for both segments and therefore the rate of energy loss of one segment equals the rate of energy gain of the second segment connected with the joint, the angular velocity of the two connected segments is not necessarily the same. Hence, the power produced by the NJT can not only transfer energy, mechanical energy can also be absorbed or generated by the muscles (Robertson & Winter, 1980) (Table 1). To further estimate power transfer, generation and absorption (TGA), the previously calculated terms of the SP analysis were used. To account for the net rate of energy transfer (PT) at the different segments, the sum of the JFP of the distal segment of the joint and the portion of the distal STP which represents transfer by the NJT according to the conditions outlined in Table 1 was calculated (Wasserberger et al., 2021). Additionally, the time series for the rate, at which energy is generated (PG) and absorbed (PA) at the corresponding joint was calculated accordingly.

Beside the EF between the different segments, the kinematics and the kinetics of the center of mass (CoM) of the javelin were also calculated. The position of the CoM of the javelin was calculated based on data obtained from markers attached to the javelin. The velocity (v_j) of the javelin’s CoM was calculated via differentiation of the CoM’s position. The release speed (v₀) of the javelin was defined as the magnitude of the javelin’s velocity one frame after release. Furthermore, the acceleration force (F_acc) and acceleration power (P_acc) acting on the CoM of the javelin were calculated. The F_acc was calculated as: ${\displaystyle {\mathit{F}}_{\mathit{acc}}={\mathit{a}}_{\mathit{j}}{m}_j={\dot{\mathit{v}}}_{\mathit{j}}{m}_j,}$ where a_j is the acceleration vector of the javelin’s CoM and m_j is the mass of the javelin. The P_acc of the javelin was calculated as: ${\displaystyle P_{acc}={\mathit{F}}_{\mathit{acc}}{\mathit{v}}_{\mathit{j}}.}$

Using the SP power-time series, peak segment torque power (pPST) and peak joint force power (pJFP) acting on the proximal end of the distal segment of a joint were extracted. From the TGA power-time series, peak transfer (pPT), generation (pPG) and absorption (pPA) were extracted for each joint. In addition, by integrating the different power-time series from the touchdown of the rear leg until the release of the javelin, the energy that was transferred, generated or absorbed by the TGA power terms and the energy delivered by the NJT and NJF to the distal segment was calculated. The peak shoulder internal rotation (T^IR) and elbow varus torque (T^{V AR}) were extracted from the joint moments. Furthermore, the peak values were extracted for both F_acc and P_acc.

Statistics

All extracted variables were tested for normal distribution using the Shapiro–Wilk test. A repeated measures analysis of variance (rmANOVA) was conducted for the different peak power and energy terms of the shoulder, elbow and wrist joint to investigate joint differences. In addition, Mauchly’s test of sphericity was carried out to test for equal variances of differences. If the assumption was violated, Greenhouse-Geisser correction was used to adjust the degrees of freedom. If a main model effect (p < 0.05) was found, Bonferroni-corrected post-hoc comparisons were calculated.

To test for relationships between the power variables and v₀, T^IR and T^{V AR}, a correlation analysis was calculated. Afterwards, the extracted peak Power variables, separated by TGA and SP, were entered as independent variables into a multiple stepwise regression analysis to determine the linear model that best predicts v₀, T^IR and T^{V AR}, respectively. Power variables were normalized using body mass (BM), the T^IR and T^{V AR} were normalized by BM × body height (BH). As for the rmANOVA, the normalized data were tested for normal distribution. To address multicollinearity between predictor variables, the variance inflation factor (VIF) was observed. In case of collinearity (VIF > 10), the most distal terms were removed first in order to receive the earliest predictor possible. The model was chosen by an information theory approach using corrected Akaikes Information Criterion (AICc), which was corrected for small sample sizes (DelSole & Tippett, 2021). The model with the lowest AICc was chosen for TGA and SP inputs. To compare models between TGA and SP, the goodness of fit for each model was described by the adjusted coefficient of determination (R²). Unstandardized regression model parameters (B) as well as their 95% confidence intervals (95% CI) were calculated. All statistical analyses were conducted using MATLAB.

Energy flow and performance

The results indicate that mechanical energy is primarily transferred across the joints of the upper arm, which is in concordance with findings from baseball studies (Aguinaldo & Escamilla, 2019; Howenstein, Kipp & Sabick, 2019; Wasserberger et al., 2020). While the production of small amounts of energy at the different joints cannot not be denied, it cannot be stated conclusively whether this energy is reused elastic energy or energy generated from the muscles. However, while the correlation analysis could not reveal relations between the v₀ and the peak energy generation at the different joints, the best TGA regression model, chosen by the AICc, indicates that the peak power generation at the shoulder is related to v₀. This observation is redundant with the SP regression model, where peak joint force power and peak segment torque power at the shoulder entered the regression, the latter containing transfer and generation of energy by the NJT. Therefore, the results indicate that the shoulder is the last joint inserting energy into the kinetic chain, which partly confirms the previous assumptions made in javelin throwing, but neglects the proposed need for energy generation at the elbow and wrist joint (Bartonietz, 2000; Tidow, 2008; Strüder, Jonath & Scholz, 2013).

The need for high rates of energy transfer is also supported by the different time series, as well as the differences in the energy transferred by the different power terms. The P_acc peaks shortly after reaching the timepoint of the maximum external rotation(MER) at the shoulder, therefore the energy flow to the javelin up to MER seems to be realized by the transfer from the proximal segments. As a shortening of the phase up to MER, where the muscles contract eccentrically, would lead to a reduced release speed, the amplitude needs to be maintained (Roach & Lieberman, 2014). To enhance the rate of energy transfer in this phase the eccentric angle-torque relationship should be improved as this would lead to higher joint torques which are basic requirements to do more work. Also, due to eccentric training not only the torque–angle relationship could be improved, fascicle lengthening could also be reached. This could be advantageous as longer fascicles are generally linked to higher power output (Kumagai et al., 2000; Blazevich et al., 2007; Butterfield, 2010; Lee et al., 2021).

If one looks at the energy generated and transferred at the shoulder, one can see that the transfer of mechanical energy accounts for a significantly larger fraction. The question is, therefore, whether and to what extent the generation of mechanical energy at the shoulder contributes to the release speed and whether a focus on the transfer of mechanical energy would be advantageous. However, this cannot be conclusively clarified with the current data.

Interestingly, the different joints do not differ in terms of peak energy transfer, but they do if STP or JFP and thus NJT or NJF are used. While both the STP and the JFP are utilized to transfer mechanical energy at the shoulder joint, the more distal joints rely on the energy transfer by the JFP. As the TGA regression models show, it is most important to raise the rate of energy transfer into the distal segment of the shoulder: for each W/kg increased peak transfer, the v₀ increased by 0.13 ms⁻¹. But as the SP model indicates, this should mainly be done by raising the peak segment torque power, or more specifically the transfer components thereof. Each unit raise leads to an improvement in the v₀ of about 0.16 ms⁻¹, while the unit raise of the peak joint force power leads to an improvement of about 0.09 ms⁻¹. Therefore, the rotations of the upper body seem to be an important factor in transferring energy through the shoulder utilizing NJT, which was also found in baseball (Aguinaldo & Escamilla, 2019; Aguinaldo & Escamilla, 2022). For javelin throwing it could therefore be assumed that the rotation of the upper body about its longitudinal axis and its forward tilt are essential motions needed to transfer mechanical energy. It is therefore very important to generate a stable support, especially by means of the bracing leg, to allow the upper body to act in the desired manner (Bartonietz, 2000; Tidow, 2008).

While the peak energy transfer is the same at all joints, only the energy transferred by the wrist decreases. This could be due to the close temporal proximity of the peak energy transfer to the release and therefore the short period of time available for energy transfer. To allow for more time to transfer energy, the path travelled by the javelin within the acceleration phase should be as long as possible. As Bartonietz (2000) stated, the active work of the upper body towards the throwing direction that is often observed in elite throwers may help to preserve a long acceleration path and therefore allow for sufficient time to transfer mechanical energy into the javelin.

Energy flow and joint load

The calculated joint loads for the internal shoulder rotation and elbow varus torque show values which are higher compared to recently reported values from professional baseball pitching (T^IR: 100 ± 16 Nm [59.8 ± 8.8% BM×BH]; T^{V AR}: 99 ± 17 Nm [59.8 ± 7.8% BM×BH]) although the release speed is comparatively lower (22.73 ± 1.28 ms⁻¹ vs. 38.1 ± 1.6 ms⁻¹) (Oi et al., 2019). Even if a comparison should be considered carefully between different research works due to the body-segmental inertia parameters and the modelling framework that were used (Gasparutto et al., 2021; Sterner et al., 2022; Köhler et al., 2023), it seems that the heavier javelin puts higher loads on the throwers’ joints. It should also be taken into account that in competition significantly higher release speeds are achieved and with an increase in release speed, an increase in the joint load is to be expected (Slowik et al., 2019; Köhler, Hepp & Witt, 2022). This is supported by the correlation between segment torque power at the shoulder and T^IR and T^{V AR} as well as v₀. Furthermore, the SP regression models for the three dependent variables contain the same predictor (peak segment torque power). Thus, athletes with greater peak segment torque power have a faster release speed but are also exposed to bigger T^IR and T^{V AR}, and hence, a higher load on the shoulder and elbow. Therefore, an appropriate preparation for high loads is urgently needed in order to prevent injuries and to avoid long-term absences from training and competition. Particular attention should also be paid to the specifics of the joint loads and the type of muscle contraction, since general intervention strategies seem to be ineffective in avoiding overuse injuries (Achenbach et al., 2022).

When using the TGA regression model, the increase in joint loads also seems unavoidable. But as the differentiation in peak energy transfer and peak energy generation at the shoulder shows, there is the opportunity to make the increase of the joint load more efficient. While the effect of peak energy generation is minor compared to peak energy transfer relating to the raise in v₀, this relationship is reversed for both joint loads. For each W/kg raise in peak energy generation, there is an increase of almost double for both joint loads compared to peak energy transfer. This leads to the assumption that the same v₀ can be reached with lower joint loads, or higher release speeds at the same load, if the thrower is able to transfer more energy from the proximal segments rather than generating it at the shoulder joint. This is in line with the findings of Howenstein, Kipp & Sabick (2019) for baseball pitching and Martin et al. (2014) for tennis. Both showed that players who took advantage of energy transfer from the proximal segments had lower joint loads or used their joint loads more efficiently. Furthermore, other studies reported that improper timings within the kinetic chain lead to reduced v₀ and increased joint loads (Aguinaldo, Buttermore & Chambers, 2007; Urbin et al., 2013). It was proposed by Urbin et al. (2013), that the disruption in energy transfer caused by improper timing of pelvis and trunk rotation either caused a reduction in release velocity, or resulted in increased T^IR and T^{V AR} in an attempt to compensate for the lost energy. Therefore, our data also indicate that the throwing mechanics should be focused on an effective use of the joint loads by focusing on the transfer of mechanical energy within the kinetic chain rather than the generation in the upper extremity joints. This would increase performance while reducing joint loads and, therefore, reduce injury risk as well.

Limitations

The results of the present study should be considered in the context of the following limitations. Firstly, we must take into consideration the sample size. Several statistically significant and relevant (practical and clinical) results could be found. Although the prerequisites for the selected statistical tests were checked and fulfilled, the results must be viewed with a certain degree of caution due to the group size. Nevertheless, the results show important new findings, which are also in line with the results of other sports. Therefore, they can be considered quite plausible.

Secondly, the release velocities were relatively low compared to results from competitions. This could be due to different reasons. Firstly, we should mention the timepoint of the investigation, which was several months before the competition season. Secondly, the investigation was carried out indoors, which does not meet the current conditions in javelin throwing. Furthermore, the indoor experiment leads to the fact that not the throwing distance but the release speed was used as a performance parameter. Since the competition performance depends to a large extent, but not only, on the release speed, a transferability of the results may not be possible without restrictions.

Finally, motion capturing and multi body modelling has several limitations. Errors may arise from marker motion, estimation of the body segment inertia parameters as well as the computation of joint centers. However, in the context of the chosen methods, everything possible was done to minimize their influence.