Relationships between throwing mechanics and shoulder anterior force in high school and professional baseball pitchers

Abstract

Background

Excessive shoulder anterior force has been implicated in pathology of the rotator cuff in little league and professional baseball pitchers; in particular, anterior laxity, posterior stiffness, and glenohumeral joint impingement. Distinctly characterized motions associated with excessive shoulder anterior force remain poorly understood.

Methods

High school and professional pitchers were instructed to throw fastballs while being evaluated with 3D motion capture (480 Hz). A supplementary random forest model was designed and implemented to identify the most important features for regressing to shoulder anterior force, with subsequent standardized regression coefficients to quantify directionality.

Results

130 high school pitchers (16.3 ± 1.2 yrs; 179.9 ± 7.7 cm; 74.5 ± 12.0 kg) and 322 professionals (21.9 ± 2.1 yrs; 189.7 ± 5.7 cm; 94.8 ± 9.5 kg) were included. Random forest models determined nearly all the variance for professional pitchers (R² = 0.96), and less than half for high school pitchers (R² = 0.41). Important predictors of shoulder anterior force in high school pitchers included: trunk flexion at maximum shoulder external rotation (MER) (X.IncMSE = 2.4, β = −0.23, p < 0.001), shoulder external rotation at ball release (BR)(X.IncMSE = 1.7, β = −0.34, p < 0.001), and shoulder abduction at BR (X.IncMSE = 3.1, β = 0.17, p < 0.001). In professional pitchers, shoulder horizontal adduction at foot contact (FC) was the highest predictor (X.IncMSE = 13.9, β = 0.50, p < 0.001), followed by shoulder external rotation at FC (X.IncMSE = 3.6, β = 0.26, p < 0.001), and maximum elbow extension velocity (X.IncMSE = 8.5, β = 0.19, p < 0.001).

Conclusion

A random forest model successfully selected a subset of features that accounted for the majority of variance in shoulder anterior force for professional pitchers; however, less than half of the variance was accounted for in high school pitchers. Temporal and kinematic movements at the shoulder were prominent predictors of shoulder anterior force for both groups.

Clinical relevance

: Our statistical model successfully identified a combination of features with the ability to adequately explain the majority of variance in anterior shoulder force among high school and professional pitchers. To minimize shoulder anterior force, high school pitchers should emphasize decreased shoulder abduction at BR, while professionals can decrease shoulder horizontal adduction at FC.

Introduction

Anterior shoulder force has been reported to be a source of anterior shoulder pain and injury in pitchers, with prior studies establishing that most cases of anterior shoulder pain were during the acceleration and early follow-through phases of the pitching motion.^1,2 Excessive shoulder anterior force in baseball pitchers has been implicated in the pathology of the rotator cuff; in particular, anterior laxity, posterior stiffness, and glenohumeral joint impingement. ³ However, the relationships between pitching motion as a function of segment kinetics, mechanical stress, and shoulder injury remain elusive.^2,3 Several studies have investigated particular components of the pitch in isolation, reporting that shoulder horizontal abduction at maximum external rotation,^4–6 degree of glove arm extension, ⁷ and trunk rotation^8,9 are associated with the magnitude of stress incurred at the shoulder during pitching among various playing levels. Therefore, it is imperative to understand the combination of kinematic and temporal variables associated with changes in the forces experienced by the shoulder. Identifying these variables would allow for optimization of throwing mechanics and to minimize injury risk in pitchers. To this end, a predictive model capable of identifying the combination of variables with the greatest influence on shoulder anterior force could be of significant clinical benefit.

Although distinct models have been developed that characterize distinct upper extremity kinetics such as elbow varus torque,^10–14 shoulder internal rotation torque, ¹⁵ and shoulder distraction force,^16,17 no such model has been attempted to account for the variance in shoulder anterior force in any playing level, even given the clinical implications discussed. The purpose of this study was to create a predictive model to identify kinematic pitching variables that explain the greatest variance in anterior shoulder force in high school and professional pitchers. The authors hypothesized that the statistical model would be able to identify a combination of features that explain a high degree of variance in anterior shoulder force during pitching.

Methods

A total of 130 high school and 322 professional pitchers were included in this study. Inclusion requirements for professional pitchers were: (1) at the time of testing, pitchers were on the Major League or Minor League (Low A, High A, AA, and AAA) roster; (2) pitchers had no record of serious injury (requiring > 2 weeks of rest or rehabilitation) in the past six months; and (3) cleared by team clinicians to participate in baseball activities. Inclusion requirements for high school pitchers were: (1) at the time of testing, pitchers were on a high school or club team, (2) had no record of serious injury (requiring > 2 weeks of rest or rehabilitation) in the past six months; and (3) cleared to participate in baseball activities by their physician. Motus Global (Rockville Center, NY USA) de-identified all data pre-distribution, qualifying for exempt institutional board review under federal guidelines.

Pitching evaluations were conducted as previously described.^18,19 Pitchers reported to the test site and a privacy waiver was administered with consent provided. For underage pitchers, parents provided consent and the players provided assent. Demographic data were reported by the pitcher and included age, preferred throwing arm, years of play, and history of injury. Pitchers with prior elbow or shoulder surgery were included given they were not considered injured and were actively throwing for at least 6 months. Researchers measured and recorded the pitchers’ height with a standard tape measure and weight using a medical scale. The pitcher was given unlimited time to warm-up with his preferred routine of pitching at maximum effort (i.e. arm bands, stretching, plyocare, long-toss). Once the pitcher indicated that he was ready to pitch, 46 reflective markers were placed on anatomical landmarks. ¹⁸ The 8-camera Raptor-E motion analysis system (Motion Analysis Corp, Santa Rosa, CA, USA) was used to record the markers at 480 Hz. Prior to pitching, a single static calibration was collected with the pitcher standing in the capture volume, the hip-width legs apart, the shoulders abducted at 90°, and the elbows flexed at 90°. The static test was conducted to align the pitcher with the laboratory coordinate system and to define the local coordinate systems. The global coordination system was established based on the International Society of Biomechanics standards: Y was vertically upward, X was perpendicular to Y (positive to home plate), and Z was the cross product of X and Y.

Pitchers were instructed to pitch 8–12 fastballs from a standard dirt mound with game-like effort to a catcher behind home plate at a regulation distance (18.4 m). Pitchers pitched at their own set rate and were given the option to pitch from the stretch or the wind-up. Ball velocity was collected with a radar gun located behind the pitcher (Stalker Sports Radar, Richardson, TX, USA).

All data processing for building full body kinematics and throwing arm kinetics was performed in MATLAB scripts (The Mathworks, Natick, MA, USA) as previously described by Luera et al. ¹⁸ Data from the markers were filtered by a low-pass filter (fourth order, zero-lag Butterworth filter, 13.4 Hz cutoff frequency). ²⁰ Three time points of the pitch were used in this study: foot contact, maximum external rotation, and ball release. Foot contact was defined as the first frame when the lead toe or heel had reached the minimum in the Y axis. Maximum external rotation of the shoulder was established as the frame in which the throwing arm achieved maximum external rotation with respect to the trunk. Note the scapula is not immobilized during the pitch and thus, the external rotation angle is influenced by scapular motion and is not an independent representation of the glenohumeral joint. Ball release was determined to be an instant of 0.01 s after the wrist passed the elbow in the forward direction.^19,20 Maximum shoulder internal rotation was identified during the frame when the throwing arm reached maximum internal rotation. The horizontal adduction of the shoulder was defined as the angle between the upper throwing arm and the upper trunk vector in the transverse plane of the upper trunk, so that a negative value was considered to be abduction. The joint reaction force exerted on the shoulder can be broken into vectors of the anterior, proximal (compressive), and superior forces whereby the anterior, proximal, and superior forces on the shoulder reflect positive values and posterior, distal (distractive), and inferior forces represent negative values. ² Peak shoulder anterior force was calculated and subsequently normalized by player weight (kg·m/s²).^18,19

As the number of pitches thrown per pitcher was not uniform, the median of all independent variables and the dependent variable was calculated per pitcher. The median was chosen rather than the mean because it is less partial to noise and other outlying data.

A random forest model was developed and implemented to identify the most important features for regressing to shoulder anterior force. Absolute variable contribution to the mean square error (IncMSE) was chosen as the feature selection metric. IncMSE is defined as the increase in MSE of prediction, given each of the variables is subject to permutation. Using this method, each variable is assessed for relative importance as it relates to anterior shoulder force through the various decision trees comprising the random forest. The corresponding change to the MSE is subsequently derived for each variable.

The 10 most influential variables were subsequently included in regression calculations with normalized shoulder anterior force as the dependent variable. Multivariate regression was employed to determine whether potential associations existed between kinetic or kinematic variables and anterior shoulder force. The primary metric of model performance was Pearson correlation coefficient (R²) used to describe the strength of the linear relationships between the independent variables and dependent variable (anterior shoulder force).

Univariate regression was then employed to more robustly examine the effect of shoulder anterior force on measured kinetic forces and torques. Similar to the aforementioned multivariate regression analysis, R² was utilized as a primary metric of model performance. Additionally, an adjusted R² was calculated to aid model evaluation. Whereas the R² metric is derived under the assumption that every variable contributes to the dependent variable, an adjusted R² indicates the proportion of variation explainable by independent variables that affect the dependent variable. In line with our assessment of variable importance with respect to shoulder anterior force, p-values were included to quantify the importance of independent variables. Pearson correlation coefficients were interpreted as weak (<0.10), moderate (0.10–0.30), or strong (>0.50). ²¹ Significance was set at p < 0.001. Data manipulation and analyses were performed in R statistical computing software (version 4.0.3, The R Foundation, Vienna, Austria). The random forest implementation was unloaded from the Caret package in R.

Results

Demographics for the high school and professional pitchers are listed in Table 1. High school pitchers significantly differed compared to professional pitchers on the basis of age, height, and weight (p < 0.001). Professional pitchers had greater shoulder anterior force (392.3 ± 14.1 N) compared to high school pitchers (252.8 ± 10.4 N) (p < 0.001), even when normalized by pitcher weight (34.6 ± 6.3 vs. 42.2 ± 1.5%BW respectfully, p < 0.001).

Table 1.

Pitching population demographics.

	High school (n = 36)	Professional (n = 322)	P value
Age, years	16.3 ± 1.2	21.9 ± 2.1	<0.001
Height, cm	179.9 ± 7.7	189.7 ± 5.7	<0.001
Weight, kg	74.5 ± 12.0	94.8 ± 9.5	<0.001
Handedness	88 R, 42 L	245 R, 77 L	-

The top variables that could predict anterior shoulder force and their mean squared errors are listed in Table 2. Multivariate regression was performed on anterior shoulder force, resulting in a model that could account for less than half of the variance (R² = 0.41). Significant variables included in this model are shown in Table 3.

Table 2.

Most predictive variables for shoulder anterior force in high school pitchers.

Kinematic variables	X.IncMSE
Back hip internal rotation at HS, °	1.769
Trunk flexion at MER, °	2.358
Lead hip flexion at MER, °	1.976
Shoulder abduction at MER, °	1.941
Elbow flexion at MER, °	1.839
Shoulder abduction at BR, °	3.116
Shoulder external rotation at BR, °	1.657
Lead knee flexion at BR, °	1.824
Trunk flexion at MIR, °	1.778

Note: HS, hand separation; MER, maximum shoulder external rotation; BR, ball release; MIR, maximum internal shoulder rotation; X.IncMSE, increase of the mean squared error.

Table 3.

Significant variables for shoulder anterior force regression in high school pitchers.

Kinematic variables	B	β	P value
Trunk flexion at MER, °	−0.226	−0.277	<0.001
Shoulder abduction at BR, °	0.144	0.168	<0.001
Shoulder external rotation at BR, °	−0.114	−0.335	<0.001
Trunk flexion at MIR, °	0.177	0.293	<0.001

Note: HS, hand separation; MER, maximum shoulder external rotation; BR, ball release; MIR, maximum internal shoulder rotation.

A total of 322 professional pitchers were also included in this study (21.9 ± 2.1 years; 189.7 ± 5.7 cm; 94.8 ± 9.5 kg). The top variables predictive of anterior shoulder force and their mean squared errors are listed in Table 4. A similar model that was used for the high school players was created to describe anterior shoulder force. Unlike the model for high school players, this model was capable of accounting for most of the variance in anterior shoulder force (R² = 0.96). Significant variables included in this model are shown in Table 5.

Table 4.

Most predictive variables for shoulder anterior force in professional pitchers.

Kinematic variables	X.IncMSE
Shoulder horizontal adduction at FC, °	13.939
Maximum elbow extension velocity, °/s	8.51
Shoulder external rotation at BR, °	7.763
Maximum elbow extension velocity, % pitch time	7.675
Forearm pronation at MER, °	6.848
Trunk flexion at MER, °	5.229
Maximum shoulder internal rotation velocity, °/s	5.187
Shoulder external rotation at FC, °	3.585
Pelvis rotation velocity, % pitch time	3.487
Shoulder horizontal adduction at BR, °	2.676

Note: FC, foot contact; MER, maximum shoulder external rotation; BR, ball release; X.IncMSE, increase of the mean squared error.

Table 5.

Significant variables for shoulder anterior force regression in professional pitchers.

Kinematic variables	B	β	P value
Shoulder horizontal adduction at FC, °	0.296	0.5	<0.001
Elbow extension maximum velocity, °/s	0.005	0.187	<0.001
Elbow extension maximum, % pitch time	0.493	0.126	<0.001
Forearm pronation at MER, °	−0.048	−0.087	<0.001
Shoulder external rotation at FC, °	0.074	0.258	<0.001
Shoulder horizontal adduction at BR, °	0.119	0.145	<0.001

Note: FC, foot contact; MER, maximum shoulder external rotation; BR, ball release.

Kinematics of interest, denoted as a function of time, are listed as follows for the high school and professional pitchers. At foot contact, high school pitchers had 86 ± 2° of shoulder abduction, 93 ± 1° at maximum external rotation, and 91 ± 1° at ball release. High school pitchers’ trunks were slightly flexed at foot contact (10 ± 4°), proceeded to a neutral position at maximum external rotation (−2 ± 2°), and proceeded to flex the trunk at ball release (15 ± 2°). At foot contact, high school pitchers’ shoulders were externally rotated (43 ± 8°) and continued to externally rotate to a maximum peak (162 ± 2°). Pitchers’ shoulders then began to internally rotate after this peak was achieved, and at the time of ball release, finished with the shoulder parallel to the vertical axis (96 ± 4°). Pitchers began with contralateral trunk flexion at foot contact (−4.8 ± 3.7°), proceeded to move away from the throwing arm at maximum external rotation (−19.4 ± 1.8°), and continued in this direction at ball release (−37.4 ± 1.7°).

At foot contact, professional pitchers’ shoulders were horizontal abducted (−37.9 ± 2.2°), then moved into horizontal adduction at maximum external rotation (7.8 ± 1.3°), and at the time of ball release finished in a neutral position (1.9 ± 1.5°). At foot contact, pitchers’ shoulders were externally rotated (31.1 ± 7.6°) and continued to externally rotate to a maximum peak (165.2 ± 1.5°). Pitchers’ shoulders then began to internally rotate after this peak was achieved, and at the time of ball release, finished with the shoulder parallel to the vertical axis (83.8 ± 4.0°). Pitchers began with their forearms pronated at foot contact (24.8 ± 5.7°), which then progressed to neutral positions at maximum external rotation (8.8 ± 3.7°) and ball release (3.4 ± 4.5°). Pitchers began at foot contact with their trunks slightly flexed (10.0 ± 3.5°), progressing to slight extension at maximum external rotation (−4.8 ± 1.9°), and finally, back to trunk flexion at ball release (14.6 ± 2.2°).

Discussion

The main findings of the current study were that a random forest model was successfully able to select a subset of kinematic features that accounted for the majority of variance in shoulder anterior force for professional pitchers (R² = 0.96), though less than half of the variance could be accounted for in high school pitchers (R² = 0.41). Additionally, temporal and kinematic motions at the shoulder were prominent predictors of shoulder anterior force for both groups. For high school pitchers, in particular, trunk flexion at maximum external rotation, shoulder external rotation, and shoulder abduction at ball release, and trunk flexion at maximum shoulder internal rotation derived the greatest standardized regression coefficients. For professional pitchers, shoulder horizontal adduction at foot contact, shoulder external rotation at foot contact, and elbow extension maximum velocity derived the greatest standardized regression coefficients. Shoulder anterior force was a significant, but weak, correlate with several throwing arm kinetics at the shoulder and elbow in high school and professional pitchers.

A random forest model was able to select a combination of features that predicted the majority of variance in shoulder anterior force in professional pitchers; however, the random forest model for high school pitchers only accounted for less than half of the variance in anterior shoulder force. This may reflect a greater discrepancy in pitching kinematic variables between high school pitchers, who may be less precise, consistent, or refined in their pitching technique, making them more difficult to characterize. Interestingly, other than shoulder external rotation at ball release, the variables identified as being important by the random forest models differed between the two groups. These differences may reflect variations in pitching technique, or even structural changes to the shoulder, such as progressive changes from microtrauma accumulation. ³ Compensatory mechanical adaptations over time may allow the generation of greater anterior forces observed in pitchers at higher playing levels, even when accounting for differences in mass. ³

Understanding optimal pitching biomechanics may help to both minimize the risk of injury and identify the ranges of parameters leading to maximum performance. Deviation from these ideal ranges can lead to performance degradation. Anterior shoulder pain is rather common in overhead athletes, particularly during the acceleration and early follow-through phases of pitching. ¹ Biomechanical studies have suggested anterior capsule laxity^22–24 as well as posterior capsular tightness^25,26 as important contributors to shoulder pain. Repetitive loading can cause structural changes or adaptations that may allow improved performance in the interim; however, in excess, these can lead to injury and performance degradation over time. Fleisig et al. ²⁷ demonstrated that successful pitchers have greater joint laxity and flexibility allowing for greater shoulder horizontal abduction and external rotation, ultimately generating greater ball velocity and throwing arm forces and torques. However, excessively increased laxity can conversely contribute to instability and shoulder dysfunction. ²⁷ Repetitive high magnitude loading of the anterior shoulder tissues may lead to anterior shoulder instability through stretching of the anterior capsule.^4,6 Moderate anterior laxity can improve pitching performance, but excessive anterior instability is thought to be the central factor leading to pitching-related shoulder injuries. ²²

Shoulder horizontal adduction at foot contact and ball release were important components in predicting anterior shoulder force as determined by the random forest model for professional pitchers. For every 10° increase in shoulder horizontal adduction at foot contact for professional pitchers, shoulder anterior force increased by 2.96%BW (p < 0.001). Tanaka et al. ² examined shoulder horizontal abduction position and joint forces in a group of adolescent Japanese baseball pitchers. The authors observed the magnitude of anterior–posterior force at ball release strongly correlated with shoulder horizontal adduction-abduction (R² = 0.72, p < 0.001). Their regression analysis showed that 10.7° of shoulder horizontal adduction at ball release minimized the magnitude of anterior–posterior force exerted on the shoulder, and that deviations of >5° from these angles significantly increased the magnitude of forces. Takagi et al. ²⁸ similarly reported that excessive horizontal abduction at maximal external rotation during pitching significantly increased magnitude of anterior shoulder force, a surrogate for joint loading with a theoretical injury risk. The differences between these reports and the current study findings could be attributed to differences in pitching levels (homogeneous vs. mixed cohort evaluations), the analysis of shoulder horizontal abduction at a specific time point in the pitch rather than at its maximum value, and the evaluation of peak throwing arm kinetics versus using the value at a defined moment of interest (i.e. maximum shoulder external rotation). Ultimately, more research is needed to better understand the role of shoulder horizontal abduction and its definitive role in injury risk of the throwing arm.

Internal impingement is one of the outcomes by which increased horizontal adduction raises potential risks for shoulder joint injury.^4,27,29 Increased abrasion between the posterosuperior component of the glenoid and the greater tuberosity of the humeral head may result from repetitive excessive horizontal adduction when pitching.^4,30 The supraspinatus and infraspinatus tendons are also affected by this repetitive activity given the increased degree of interaction between these tendons and the glenoid. Impingement not only raises the risk of experiencing shoulder pain in pitchers but it may also be a means to additional damage.^31–34 Indeed, superior labrum anterior posterior (SLAP) tears^4,34 and rotator cuff tears^4,33 are two widely observed pathologies in pitchers that have been attributed to impingement.

Assessing the timing of kinematics most crucial in generating shoulder anterior force is of interest. Fortenbaugh et al. ²⁹ reported maximum anterior shoulder force is generated during the arm-cocking phase, ending at maximum shoulder external rotation. Although maximum shoulder external rotation is an important contributor to ball velocity, it also is a position of injury susceptibility, particularly for internal impingement of the glenoid, with the inferior surface of the rotator cuff and anterior capsule being subject to significant tension. ²⁷ Laxity of the anterior shoulder can occur when the glenoid labrum and glenohumeral ligaments absorb excessive stresses from the glenoid fossa, increasing instability within the joint. Although it was anticipated the moment of maximum shoulder external rotation would put pitchers at the highest susceptibility to increased shoulder anterior force, the timing of shoulder kinematic measures did not preferentially occur at any specific time point of the pitch for high school or professional pitchers.

There are a few limitations that should be discussed in the context of the above findings. First, only a finite number of pitches were collected for each pitcher, which does not allow for the ability to understand the effect of player fatigue on performance and anterior shoulder force. Second, only fastballs were assessed, limiting the generalizability of our models to other pitch types. Third, while these random forest models of kinematic variables predict anterior shoulder force, other forces of the shoulder were not specifically examined, which may also influence anterior shoulder force. However, random forest algorithms use rigorous statistical methodology to determine feature subsets with high predictive value, and a large pool of initial features that may contribute to anterior shoulder force were examined. Additionally, the range of values for variables that best minimize anterior shoulder force was not estimated given the statistical approach used. However, the purpose of this study was to identify variables that influenced anterior shoulder force. Lastly, several studies have used different criteria for which pitches to include (i.e. fastest pitches, strikes only, etc.). Although no method has been deemed superior, it is worth mentioning the inability to generalize our results to other pitch analysis methodologies.

Conclusion

A random forest model was successfully able to select a subset of features that accounted for the majority of variance in shoulder anterior force for professional pitchers; however, less than half of the variance could be accounted for in high school pitchers. Temporal and kinematic motions at the shoulder were prominent predictors of shoulder anterior force for both cohorts. To minimize shoulder anterior force, which may put pitchers at risk of injury, pitchers should emphasize proper shoulder motions when pitching.