This study was carried out in the Cardiothoracic Surgical Unit, Auckland City Hospital, a tertiary Epigenetics Compound Library referral hospital in New Zealand. One control group participant inadvertently received physiotherapy intervention as per the experimental group until discharge

from hospital. Another control group participant required physiotherapy input for a postoperative neurological complication, including transfer to a stroke rehabilitation unit, however as the neurological problem was cerebellar, this did not include specific shoulder and thoracic cage exercises. There were no reports of additional shoulder and thoracic cage exercises implemented during the inpatient phase for experimental group participants beyond those in the protocol. Two participants from each group reported that they had independently sought

treatment for problems related to their shoulder on the operated side following discharge from hospital. Data from all these participants have been analysed using intention-totreat principles. Experimental group interventions were provided PI3K inhibitor as scheduled on 81% of occasions during the inpatient phase of the trial. For the experimental group, the median (range) number of physiotherapy treatment sessions received was 6 (1 to 18) and the median (range) total physiotherapy time per participant in 15-minute units of service was 12 (2 to 47) units. For the 76 randomised participants, data on pain, shoulder function and quality of life were obtained 83% of the time. Missing data most frequently resulted from nonreturned or incomplete questionnaires. For the subgroup of 47 participants who were scheduled to participate in measures of range of motion and strength, data were obtained 82% of the time. Missing data most often resulted from unwillingness or inability to attend for measurement. Exercise diaries were completed by only 8 (19%) of the 42 experimental group participants, so data from the diaries have not been reported. The physiotherapists who acted as independent assessors were asked to report any episodes of unblinding to group allocation. Five reports of inadvertent unblinding were received from the 122 follow-up assessment occasions (4%):

2 of these episodes occurred at the time of discharge, and 3 episodes occurred at the 3 months next postoperative follow-up. When unblinding occurred, an alternative blinded assessor performed the outcome measures on all subsequent occasions. Group data at baseline and follow-up are shown in Table 2 for pain and range of motion and in Table 3 for muscle strength, shoulder function and quality of life. Individual data for all outcomes are provided in Table 4 (see eAddenda for Table 4). The experimental group had significantly less shoulder pain at discharge than the control group, by 1.3 units (95% CI 0.3 to 2.2). The experimental group also had significantly less total pain than the control group at discharge, by 2.2 units (95% CI 0.2 to 4.3).