Gracitelli 2016.

Study characteristics
Methods	Method of randomisation: use of computer software with allocation concealed with password‐protected database Assessor blinding: yes, evaluations were performed by a physiotherapist Loss to follow‐up at 1 year: 7/72 (4 deaths)
Participants	University Hospital, São Paulo, Brazil Period of study recruitment: May 2011 to March 2014 72 participants with closed displaced surgical neck proximal humeral fractures, without or with involvement of the greater tuberosity (2‐part or 3‐part surgical neck fractures, abiding by the Neer criteria of displacement ≥ 1 cm or ≥ 45 degrees angulation), aged 50 to 85 years, treated surgically ≤ 21 days of injury. Written informed consent. Exclusion criteria: isolated fracture of the greater or lesser tuberosity, Neer 4‐part fractures, fracture involving the articular surface of the humeral head, fracture‐dislocation of the proximal humerus, neurological injuries in the affected limb, previous surgery on the affected shoulder, associated fractures in the affected limb, pathologic fractures, psychiatric illnesses or inability to understand preoperative questionnaires, active infection or previous infection on the shoulder. (Post‐randomisation exclusion criteria listed: loss to follow‐up before the first clinical assessment at 3 months, irreparable tendon tears of the rotator cuff.) Of 65: 47 female, 18 male; mean age 65 years, range not stated
Interventions	Interventions started: surgery on average 9.7 days after injury (randomisation after anaesthesia). All had general anaesthesia with an interscalene brachial plexus block and were positioned in dorsal decubitus with 30 degrees elevation. 1. Internal fixation using a locking plate: PHILOS (DePuy‐Synthes, Solothurn, Switzerland), deltopectoral approach. The plate was placed 1 cm inferior to the upper portion of the greater tuberosity and 1 cm lateral to the long head of the biceps tendon. Three guidewires were inserted through the plate proximal holes, followed by distal fixation with a cortical or locking screw under radioscopic imaging. At least 5 locking screws were inserted proximally, followed by 2 more distal locking screws, with a minimum of 3 distal screws. Tension‐band sutures were placed in the proximal plate holes, with non‐absorbable sutures passed through the rotator cuff tendons. 2. Internal fixation using a locking nail: Centronail (Orthofix, Verona, Italy), 3 proximal locking screws, longitudinal anterolateral transdeltoid approach. Supraspinatus and infraspinatus tendons sutured with non‐absorbable sutures. Greater tuberosity fractures were reduced before the intramedullary nail was inserted. The humeral head was reduced and provisionally fixed with Kirschner wires. A longitudinal 1.5‐cm incision was made in the supraspinatus tendon. The nail was placed 5 mm below the cartilage and proximally fixed with 3 screws and distally fixed with 2 screws. Tension‐band sutures were placed around the proximal screws, with non‐absorbable sutures passed through the rotator cuff tendons. The supraspinatus tendon was repaired with non‐absorbable sutures. Slings were used for 4 weeks. Physical therapy with passive‐ and active‐assisted exercises was started 14 days after surgery. Active and active resisted exercises were started 30 and 60 days after surgery, respectively. Assigned: 36/36 Completed (at 1 year): 29/32 but analysed at 32/33 (data used from 3‐ or 6‐month follow‐up for 3 and 1 participants)
Outcomes	Length of follow‐up: 1 year (also 3 and 6 months) DASH score Constant score (absolute and relative to opposite shoulder) Pain (VAS) Mortality Complications (overall number of participants with a complication, varus malunion (not reported); osteonecrosis, non‐union, hardware problems, refracture, shoulder stiffness, infection, neurologic injury, rotator cuff tendon tears) Reoperation Radiographic outcomes, including avascular necrosis, union, and loss of reduction of humeral head, loss of reduction of the greater tuberosity, insufficient reduction
Funding and conflict of interest statements	Funding: not reported The authors declared no benefits in any form were received from a commercial entity.
Notes
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"Randomization was generated on the website “http://www.randomization.com”, with random blocks, and stratified in 2 levels corresponding to the Neer classification."
Allocation concealment (selection bias)	Low risk	"Allocation was performed after anesthesia by using a password‐protected database." Adequate safeguards seem likely.
Blinding (performance bias and detection bias) Functional outcomes, pain, clinical outcomes, complications	Unclear risk	"Outcome evaluations were performed by a physical therapist who did not participate in the rehabilitation of patients and who was blinded to their allocation. The evaluations were performed with the patients wearing clothes to preclude the visualization of the surgical wound." However, participants and personnel were not blinded.
Blinding (performance bias and detection bias) Death, reoperation	Unclear risk	Lack of blinding unlikely to affect assessment of these outcomes, but may affect decisions for subsequent surgery.
Incomplete outcome data (attrition bias) Functional outcomes, pain, clinical outcomes, complications	Unclear risk	Participant flow described. Comparable losses to follow‐up in the two groups in terms of the analyses: 4 for the plate group (2 died); and 3 for the nail group (1 died). However, 1 participant in the plate group and 3 in the nail group did not undergo the 12‐month assessment but data from 3 or 6 months were included in the final analysis. Thus, overall 7 (19.4%) versus 4 (11.1%) were lost to follow‐up.
Incomplete outcome data (attrition bias) Death, reoperation	Low risk	Participant flow described and comparable between‐group losses. One conversion from nail to plate analysed as allocated. These outcomes are unlikely to have occurred without detection.
Selective reporting (reporting bias)	High risk	Trial registration before analysis. However, there were differences in outcomes reported: the UCLA score is not reported but described at trial registration; VAS pain is reported but not described at trial registration.
Balance in baseline characteristics?	Unclear risk	Balanced baseline characteristics for those in the analyses. However, baseline characteristics missing for 8 participants.
Free from performance bias?	Low risk	All surgery was done by the principal investigator ‐ no indication of experience with the two operations but expected. Other co‐interventions ‐ anaesthesia and rehabilitation ‐ were similar in the two groups.