This study is a double-blinded randomized controlled trial. The participants were enrolled from the rehabilitation center of a hospital. The inclusion criteria were (1) newly diagnosed oral cancer subjects with clinical signs of spinal accessory nerve dysfunction (e.g., shoulder droop, limited AROM of shoulder abduction, and insufficient muscle strength of the shoulder abductor against gravity) after neck dissection, (2) the presence of scapular dyskinesis (e.g., asymmetric scapular motion in multiple planes by observation) , (3) had prominent scapular asymmetry i.e., more than 1.5 cm side-to-side difference of the distance between the inferior angle of the scapula and the spinous process of the seventh thoracic vertebra when performing shoulder abduction to 90° with a 1kg load in the scapular plane [19, 20], and (4) age between 20 and 65 years. Participants were excluded if they (1) had bilateral neck dissection, (2) had distant metastasis or recurrence, (3) were unable to communicate or comprehend the questionnaires, (4) had a history of shoulder pain in one year prior to neck dissection, (5) had any disorder that could influence movement performance, or (6) were pregnant or breastfeeding. This study was approved by the Chang Gung Medical Foundation Institutional Review Board (Approval No: 201901788A3) and Clinical Trials (Approval No: NCT04476004). Informed consent was obtained from each participant. This report was in accordance with the Consolidated Standards of Reporting Trials (CONSORT) Statement for randomized trials (Online Resource 1).
The sample size was analyzed priorly using G*Power 3.1.9 based on AROM of the shoulder joint from a previous study , and at least eight participants in each group were required (power=80%, α=0.05). However, previous studies demonstrated that at least ten participants in each group need to be included to explore EMG activity involving the scapular muscles [14, 21]. Twelve participants in each group were recruited, considering the 10% dropout rate. A researcher who was not involved in the intervention and evaluation sessions used computer-generated random numbers to allocate four participants in one block. All participants were randomly assigned to the motor-control with biofeedback group or the motor-control group.
Before the intervention, all participants acquired anatomical and functional education about the trapezius muscle. Both groups received conventional physical therapy (e.g., scar massage, stretching, active and passive ROM exercise of the shoulder joint) and motor-control training integrated into scapular-focused exercises. The scapular-focused exercises were based on the previous studies (Online Resource 2) [7, 21, 22]. For both groups, a physical therapist provided kinesthetic and verbal cues during the exercises to enhance conscious control of scapular position and movement during exercises [22, 23]. For example, the therapist tapped the top of the acromion to instruct clavicle elevation or contacted the posterior acromion to instruct verbally to draw shoulder blades toward the spine for emphasizing scapular posterior tilt, external rotation, and upward rotation. In the motor-control with biofeedback group, additional online EMG biofeedback of the UT, MT, and LT was implemented during scapular-focused exercises (Figure 1), and participants were instructed to increase muscle activities during exercises. The physical therapist instructed the participants to focus on the specific parts of the trapezius muscle shown in Online Resource 2 during each scapular-focused exercise. There were 12 intervention sessions in three months for each participant, and there were 60 minutes of each session.
The scapular position was assessed by the modified lateral scapular slide test (MLSST), which has been proposed as a reliable method for evaluating scapular symmetry . The distance between the inferior angle of the scapula and the spinous process of the seventh thoracic vertebra was measured in centimeter three times for each side using a vernier caliper, and the difference between bilateral sides was averaged. MLSST was measured in three positions: bilateral arms placed by the side (position 1), hands placed on the hips (position 2), and holding a 1kg dumbbell and arms elevated to 90° of shoulder abduction with maximal internal rotation in the scapular plane (position 3). Intraclass correlations (ICCs) for intra- and inter-rater reliability of MLSST is 0.81–0.96 in subjects with shoulder pain, and 95% confidence interval (CI) of minimal detectable change (MDC) for MLSST is 0.67–1.40 cm on the symptomatic side .
The AROM of shoulder abduction was measured in degrees three times using a universal goniometer, and shoulder pain was measured during exercises by a 10 cm visual analog scale (VAS). The internal reliability of the two-arm goniometer is 0.58–0.99 , and the MDC of the AROM of shoulder abduction is 11–16° with good intra-rater reliability (0.91) . The test-retest reliability of the VAS is 0.94 , and the minimal clinically important difference (MCID) is 1.4–1.6 in the shoulder pain .
The Disabilities of the Arm, Shoulder, and Hand (DASH) is a 30-item, reliable and valid assessment of upper extremity function and symptoms  and has been used for patients undergoing neck dissection . The scores range from 1 to 100, with a higher score indicating greater disability. The ICC for test-retest reliability is 0.91 in patients with head and neck cancer after neck dissection . A change in the DASH score exceeding 10.83 points is meaningful in discriminating between improved and unimproved states .
The strength of the maximum voluntary isometric contraction (MVIC) of the UT, MT, and LT was measured in newtons (N) by a hand-held dynamometer (MicroFET®3, Hoggan Scientific, LLC, USA), and the testing position was based on previous studies [7, 31]. The ICC for test-retest reliability of the hand-held dynamometer is 0.85–0.96 , and for MVIC measurement is 0.84–0.98 . The participants were asked to resist a manual force provided by the physical therapist for 5 s in each testing position. Each MVIC task was repeated three times with a 30 s rest between each repetition. There was a 60 s rest between different muscles.
The muscle activities of the UT, MT, and LT were recorded by surface EMG electrodes (Ambu® BlueSensor NF-50-K, Malaysia) with an AC amplifier (cut-off frequency: 10-450 Hz; sampling rate: 1000 Hz; sampling rate: 1000 Hz; Model: QP511, GRASS, USA) when conducting the tasks of MVIC and arm movement, including elevating and lowering arm with a 1kg weight in the scapular plane for three times at a speed of 3 s per movement according to a metronome. The placement of the EMG electrodes (Online Resource 3) was based on previous studies . The root mean square (RMS) values of the EMG data were calculated between 2–5s for each MVIC task. The EMG RMS values of arm elevation and lowering were normalized by the RMS values of MVIC and were represented as %MVIC. The test-retest reliability of the EMG under MVIC is good for the scapular exercises (0.89–0.96) .
All assessments were employed at baseline (Pre-test) and the end of the intervention (Post-test) by a trained physical therapist who was blinded to the subject allocation.
The generalized estimating equation (GEE) procedure was conducted to analyze repeated-measures outcome variables over time, which has the benefit of providing higher power with a small sample size for repeated measurements with complete or missing data [34, 35]. We used a model-based estimator and an exchangeable working correlation matrix. Separate models were run for all outcome measures with post-test as the reference, and each muscle was analyzed separately for each task. Bonferroni adjustment was conducted for multiple analyses. The level of significance was set at p<0.05. Statistical analyses were completed using SPSS version 21.