Phys. Ther. Korea 2020; 27(2): 140-148
Published online May 20, 2020
© Korean Research Society of Physical Therapy
1Kinetic Ergocise Based on Movement Analysis Laboratory, 2Department of Physical Therapy, The Graduate School, Yonsei University, 3Department of Physical Therapy, College of Health Science, Yonsei University, 4Department of Ergonomic Therapy, The Graduate School of Health and Environment, Yonsei University, Wonju, Korea
This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/4.0)which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.
Background: Scapular winging (SW) could be caused by tightness or weakness of the periscapular muscles. Although data mining techniques are useful in classifying or predicting risk of musculoskeletal disorder, predictive models for risk of musculoskeletal disorder using the results of clinical test or quantitative data are scarce.
Objects: This study aimed to (1) investigate the difference between young women with and without SW, (2) establish a predictive model for presence of SW, and (3) determine the cutoff value of each variable for predicting the risk of SW using the decision tree method.
Methods: Fifty young female subjects participated in this study. To classify the presence of SW as the outcome variable, scapular protractor strength, elbow flexor strength, shoulder internal rotation, and whether the scapula is in the dominant or nondominant side were determined.
Results: The classification tree selected scapular protractor strength, shoulder internal rotation range of motion, and whether the scapula is in the dominant or nondominant side as predictor variables. The classification tree model correctly classified 78.79% (p = 0.02) of the training data set. The accuracy obtained by the classification tree on the test data set was 82.35% (p = 0.04).
Conclusion: The classification tree showed acceptable accuracy (82.35%) and high specificity (95.65%) but low sensitivity (54.55%). Based on the predictive model in this study, we suggested that 20% of body weight in scapular protractor strength is a meaningful cutoff value for presence of SW.
Keywords: Decision tree, Musculoskeletal disease, Physical examination
Scapular winging (SW) is an abnormal scapulothoracic posture that refers to prominence of the medial border or inferior angle of the scapula from thoracic wall . Abnormal scapular position and movement are associated with scapular dysfunction and imbalance of the periscapular muscles, resulting in shoulder pain and impingement syndrome [2,3]. Although multiple factors may cause SW, it can result from weakness of the scapular stabilizers without bone, joint, and neurological factors [1,2].
Soft tissue mechanisms for SW involve tightness or weakness of periscapular muscles . The serratus anterior muscle is considered a primary scapular stabilizer during arm movement, which stabilizes the scapula against the thoracic wall . Weakness of the serratus anterior muscle could lead to insufficient scapular posterior tilt and upward rotation during arm movements, causing dyskinesis [6,7]. In addition, weakness of scapular stabilizers including serratus anterior and lower trapezius may lead to pseudoweakness of the upper extremity muscles because of lack of proximal stabilization to provide a stable origin [8,9]. Inflexibility and stiffness of the pectoralis minor and biceps short head could result in scapular anterior tilt and protraction due to their pull on the coracoid process [4,10]. Moreover, glenohumeral internal rotation deficit due to posterior shoulder tightness could create compensatory scapular anterior tilts . The scapula moves with the humerus as the arm is internally rotated, which creates a “windup” of the scapula on the thorax with reduced humeral internal rotation and horizontal abduction [4,11].
Repetitive stress has been considered as an important factor for musculoskeletal disorders . Previous studies have noted that the dominant extremity often exhibits accentuated dyskinetic findings attributed to overuse, muscle asymmetry, and differences in range of motion [13-15]. The loss of shoulder internal rotation range of motion (ROM) in dominant shoulder has been studied extensively in unilateral overhead athletes such as baseball players and tennis players [13,16,17]. The results of the studies reported that dominant shoulder had significantly lower shoulder internal rotation ROM in unilateral overhead athletes. Also, Cieminski et al.  reported that the dominant shoulder had significantly less shoulder internal rotation ROM compared with the nondominant shoulder in nonathletic persons. Because the unwanted stress caused by the compensatory movements may vary depending on the degree of use of the arm , whether dominant or nondominant side could affect the prevalence of upper extremity musculoskeletal disorders .
Previous studies have proposed various methods in assessing SW, such as visual observations and 3D motion analysis systems. Visual observations are easy and inexpensive to implement but relatively subjective. In contrast, 3D motion analysis systems could provide quantitative data; however, such systems are impractical in clinical settings. Other assessment tools have been designed to measure the posterior displacement of the inferior angle of the scapula with reference to the posterior thoracic cage [20-23]. Weon et al.  designed a scapulometer to measure the posterior displacement of the scapular inferior angle relative to the thoracic wall. They also suggested that a posterior displacement > 2 cm indicates presence of SW. De Oliveira et al.  suggested that a 1.5 cm cutoff showed better sensitivity, specificity, and diagnostic performance than a 2 cm cutoff. Although the results of these tests could provide information on the presence of SW, additional tests were needed to examine the shoulder girdle function of patients.
The decision tree is one of the most popular classification techniques . One of the main reasons is the ability of decision trees to present the results in a simple format, which is easy to interpret for experts . In the decision tree model, for categorical variables, a branch is created for each known category, while, for continuous variables, the best split point is determined as the cutoff point for these variables . Previous studies reported that data mining techniques are useful in classifying or predicting risk of musculoskeletal disorder [25,27-29]. However, most studies used questionnaire data or environmental variables [25,27]. The use of data mining techniques in classifying and predicting risk of musculoskeletal disorder using the results of clinical test or quantitative data is scarce . Therefore, this study aimed to establish a predictive model for presence of SW, and determine the cutoff value of each variable for predicting the risk of SW using the decision tree method.
Fifty female subjects (mean age, 27 ± 4 years; range, 21–35) participated in this study. Subjects were excluded if they had (1) pain in the shoulder, elbow, or wrist joint; (2) previous history of shoulder, elbow, or wrist surgery; (3) problems such as tendonitis, adhesive capsulitis, or neurological pathology including damage to the long thoracic nerve. All variables were measured in both sides of all subjects. The data of both sides were used in the analysis because whether the scapula is in the dominant or nondominant side was one of predictor variables. Before the study, all subjects received explanations on all procedures of the study and signed an informed consent form approved by the Institutional Review Board of Yonsei University Mirae Institutional Review Board (approval No. 1041849-201912-BM-177-01).
In measuring SW in the static condition, each subject stood in a neutral shoulder position with the elbow joint in 90° flexion and the forearm in neutral position; then, a cuff weighing 5% of the body weight of each subject was placed on the distal wrist. Then, we measured the scapular posterior displacement (distance between the thoracic wall and inferior angle of the scapula) in all subjects using a modified scapulometer . Subjects were assigned to the SW group (scapular posterior displacement ≥ 1.5 cm) and without SW group (scapular posterior displacement < 1.5 cm) based on the previous study conducted by de Oliveira et al. .
The Smart KEMA strength sensor (KOREATECH, Inc., Seoul, Korea) was used to measure the maximal isometric scapular protractor and elbow flexor strength [32,33]. Data measured using the sensor were transferred to an Android tablet PC (Galaxy Tab A6 10.1; Samsung, Inc., Seoul, Korea) via Bluetooth network and analyzed using the Smart KEMA application (KOREATECH, Inc.). Data were recorded for 5 seconds, and the mean value of the middle 3 seconds was recorded.
The Smart KEMA motion sensor (KOREATECH, Inc.) was used to measure shoulder internal rotation ROM. The motion sensor contained a tri-axillar gyroscope, magnetometer, and accelerometer and signal converter and signal transmission sensor. Data from the motion sensor were recorded at a 25 Hz sampling frequency and transmitted to an Android tablet PC with the Smart KEMA software .
Subjects were placed in supine position on a table with 90° shoulder flexion and full elbow extension to measure maximal isometric scapular protractor strength (Figure 1). In measuring isometric strength during maximal scapular protraction, the subject grasped a handle connected to the Smart KEMA strength sensor, and then the arm length was adjusted. The Smart KEMA strength sensor was attached to a grounded vacuum lifter using a belt to provide a fixed point during measurements (Figure 1). To adjust the initial tension before strength measurements, the strength sensor maintained a tension of 2 kgf when the handle was first grasped . The maximal isometric scapular protractor strength was measured twice in each position, with a 3 minute rest period between measurements to prevent muscle fatigue. In each trial, the elbows were extended during exertion of maximal force against the grip handle in the forward direction.
Subjects were seated with the shoulder in the anatomical neutral position, elbow flexed to 90°, and forearm supinated to measure maximal isometric elbow flexor strength (Figure 2). To measure maximal isometric elbow flexor strength, a cuff connected to the Smart KEMA strength sensor was attached to the distal forearm of subjects. The Smart KEMA strength sensor was attached to a grounded vacuum lifter using a belt to provide a fixed point during measurements (Figure 2). To adjust initial tension before strength measurements, the strength sensor maintained a tension of 2 kgf with elbow 90° flexion . Subjects performed isometric elbow flexion maximally against the fixed point. The humerus of test side was stabilized to maintain neutral shoulder position using opposite side hand during the measurement. The maximal isometric elbow flexor strength was measured twice in each position, with a 3-minute rest period between measurements to prevent muscle fatigue.
Shoulder internal rotation ROM was measured with the subject lying on the measurement side. The shoulder was flexed to 90° with 0° rotation, and the elbow was flexed to 90° (Figure 3). The Smart KEMA motion sensor was attached to the wrist of subjects with a cuff. The examiner stabilized the head of the humerus to prevent anterior gliding and then passively internally rotated the humerus while maintaining 90° shoulder and elbow flexion (Figure 3). Determination of shoulder internal rotation ROM based on the point of maximal internal rotation, where a distinct, firm end feel was noted by the examiner. When the maximal passive shoulder internal rotation was achieved, the data were collected and transmitted to the tablet.
SPSS version 25.0 (IBM Co., Armonk, NY, USA) was used for the statistical analysis. The intraclass correlation coefficient (ICC 3, 1) model was used to test intra-rater reliability of measurements. Data normality was examined using the Kolmogorov-Smirnov test. Before establishing the prediction model, independent t-test was used to identify statistically significant differences in scapular protractor strength, elbow flexor strength, and shoulder internal rotation ROM between the two groups. The level of statistical significance was set at 0.05.
A classification tree for presence of SW as outcome variable was derived using the rpart function of RStudio (version 1.1.463; RStudio, Inc., Boston, MA, USA) . A classification tree selects the predictor variables that maximizes the decrease in impurity (minimization of the Gini index for categorical variables and the expected sum variances for continuous variables as node impurity criterion) . In the classification with the presence of SW as outcome variable, scapular protractor strength, elbow flexor strength, shoulder internal rotation ROM, and whether the scapula is in the dominant or nondominant side were used as predictor variables. The strength values were divided by body weight of subjects for normalization. The subjects were randomly divided into a training data set (two-thirds of the subjects) and test data set (the remaining third), each group with the same proportion in both data sets. Decision trees also have excellent prediction capabilities but have been criticized for overfitting and poor performance on extremely small data sets . Therefore, fivefold cross-validation was used to evaluate the performance of the classification tree model.
The general characteristics of subjects are presented in Table 1. Of 100 cases, 33 were assigned into the SW group and 67 cases were assigned into the without SW group (without SW). All variables were normally distributed (p > 0.05). The ICC (3, 1) of the modified scapulometer was 0.867 (95% CI, 0.808–0.908). There were no significant differences in age, height, or body weight between the two groups. There were significant differences in scapular protractor strength (p = 0.026) and elbow flexor strength (p = 0.020). However, there was no significant difference in shoulder internal rotation ROM between the two groups (p = 0.062) (Table 2). The ICCs (3,1) were 0.923 (95% CI, 0.887–0.947) for measuring scapular protractor strength, 0.926 (95% CI, 0.892–0.950) for measuring elbow flexor strength, and 0.995 (95% CI, 0.993–0.997) for measuring shoulder internal rotation ROM.
Figure 4 shows the classification tree for presence of SW, which had 4 terminal nodes. The classification tree for presence of SW showed scapular protractor strength as the first predictor of SW. Each node label showed the major subgroup (with SW or without SW). Below the node label, there was the number showing how many the major subgroup was observed in the node. Among the training data set (66 cases), 44 cases were without SW and 22 cases were with SW (Figure 4). The cutoff point of scapular protractor strength for absence of SW was ≥ 20% of body weight. In the subgroup of subjects with scapular protractor strength greater than or equal to 20% of body weight (27 cases), 23 cases were without SW and 4 cases were with SW. On the contrary, in the subgroup with lower scapular protractor strength (39 cases), 21 cases were without SW and 18 cases were with SW. Shoulder internal rotation ROM was the second predictor. The cutoff point of shoulder internal rotation ROM for absence of SW was ≥ 56°. Eleven cases had shoulder internal rotation ROM greater than 56°, and 9 of them were without SW. Twenty-eight cases had shoulder internal rotation ROM less than 56°, and 16 of them were with SW. The last predictor variable was whether the scapula is in the dominant or nondominant side. The scapula of the dominant side showed higher risk compared to the scapula of the nondominant side. In the terminal node, among 12 nondominant cases, 4 cases were with SW. On the contrary, among 16 dominant cases, 12 cases were with SW. Details of the tree divisions, with the respective predictors’ cutoff points and number of cases classified in each subgroup according to selected predictors, are presented in Figure 4.
No information rate (classified by chance) of each data set showed 67% of accuracy due to imbalanced class (33 cases with SW, 67 cases without SW). The classification tree model correctly classified 78.79% (p = 0.02) of the training data set (66 cases). The predicted SW risk levels based on the classification tree were verified against the test data set (34 cases). The accuracy obtained by the classification tree on the test data set was 82.35% (p = 0.04). The confusion matrix between the predicted and actual SW risk levels indicates good overall accuracy in both data sets (Tables 3, 4).
The classification tree showed acceptable accuracy (82.35%) and high specificity (95.65%) but low sensitivity (54.55%). Scapular protractor strength, shoulder internal rotation ROM, and whether the scapula is in the dominant or nondominant side were predictors for presence of SW. Based on the predictive model in this study, we suggested that 20% of body weight in scapular protractor strength is a meaningful cutoff value for presence of SW.
No potential conflict of interest relevant to this article was reported.
Conceptualization: GG. Data curation: GG, YW. Formal analysis: GG, YW. Investigation: GG, JK. Methodology: GG, SA, JK, OK. Project administration: GG, OK. Resources: SA, JK, OK. Software: GG. Supervision: OK. Validation: SA, JK, OK. Visualization: GG. Writing - original draft: GG. Writing - review & editing: OK.
Characteristics of study subjects.
|Variables||With SW (n = 33)||Without SW (n = 67)||p-value|
|Age (y)||28.09 ± 4.79||26.64 ± 3.94||> 0.05|
|Height (cm)||163.64 ± 4.46||162.51 ± 4.12||> 0.05|
|Weight (kg)||60.27 ± 9.96||57.82 ± 8.27||> 0.05|
|SPD (cm)||1.71 ± 0.21||0.85 ± 0.42||< 0.05*|
Values are presented as mean ± standard deviation. SW, scapular winging; SPD, scapular posterior displacement. *Represents significant differences..
Comparison of scapular protractor strength, elbow flexor strength, and shoulder internal rotation range of motion between the two groups.
|Variables||With SW (n = 33)||Without SW (n = 67)||p-value|
|Scapular protractor strength (kgf)||10.30 ± 3.11||12.07 ± 3.96||0.026*|
|Elbow flexor strength (kgf)||7.63 ± 2.64||9.15 ± 3.21||0.020*|
|Shoulder internal rotation ROM (°)||43.60 ± 12.67||49.13 ± 15.70||0.062|
Values are presented as mean ± standard deviation. SW, scapular winging; ROM, range of motion. *Represents significant differences..
Performance of the classification tree as calculated during the model training.
|Calculated evaluation measures on the training data set||Reference|
|Without SW||With SW|
|Prediction (n)||Without SW||40||10|
|Overall accuracy/95% CI||78.79%/66.98%–87.89%|
SW, scapular winging; CI, confidence interval..
Performance of the classification tree as calculated during the model testing.
|Calculated evaluation measures on the test data set||Reference|
|Without SW||With SW|
|Prediction (n)||Without SW||22||5|
|Overall accuracy/95% CI||82.35%/65.47%–93.24%|
SW, scapular winging; CI, confidence interval..