**4. Discussion**

The investigated PI approach involves a subdeltoidal access and an internervous split between the infraspinatus and the teres minor, as described by Brodsky et al. in 1987 [12]. Furthermore, it includes a medial T-shaped incision of the capsule with iuxtaglenoidal posterior to inferior capsular release to allow for the posteroinferior dislocation of the humeral head, and thus extended exposure for precise humeral head osteotomy. The study results show that the implantation of an anatomical total shoulder arthroplasty with acceptable implant sizing and positioning can be performed via a PI approach in a cadaveric shoulder.

While no statistically significant difference was observed regarding the neck-shaft angle, the variation and range seemed to be a little wider in the PI approach group, with a tendency towards more valgus positioning of the head in some cases. This can be explained by the fact that after postero-inferior dislocation of the humeral head through the internervous interval, the calcar tends to be covered by the teres minor, which needs to be pushed inferiorly, while the superior insertion of the rotator cuff is more easily exposed (Figure 2).

The COR was slightly but statistically significantly more medial in cases with the PI approach than with the deltopectoral approach, indicating a risk of lateral overstuffing due to insufficient resection of the humeral head. While this could be explained by a lack of exposure, it may also be caused by the presence of the bare area on the posterior side of the humerus, which makes identification of the anatomical neck more difficult. Since the neck-shaft angle is above 90◦, a lack of resection of the anatomical neck tendentially also leads to a superior translation of the COR, which however, was only slightly observable in this study, and did not yield statistically significant differences. The anteroposterior positioning of the COR showed no difference between the groups.

No differences in terms of sizing of the prosthetic head were observed, with slight supero-inferior undersizing but good antero-posterior matching in both groups, as the resection plane is usually oval shaped with a smaller antero-posterior than supero-inferior diameter [20]. The larger variation in the PI approach group is likely explained by the learning curve.

While the Anatomic Reconstruction Score was not statistically different in both groups, there was a trend towards slightly lower scores in the PI group, mostly explained by the larger variation of the neck-shaft angle and the lack of sufficient resection of the humeral head.

Due to the anterior tilt of the scapula, the described approach offers a postero-inferior direct view of the glenoid, which can be changed to an e-face view when the humeral head is retracted anteriorly. This may lead to a tendency of postero-inferior placement and an increased retroversion of the glenoid guide pin in cases with poor exposure. Greiwe et al. point at different advantages of the posterior approach, including easier access to the

retroverted glenoids, as well as facilitated posterior soft tissue balancing [14]. However, it remains unclear as to whether posterior approaches may also weaken the posterior soft tissues, including the posterior capsule and rotator cuff, and therefore, this may possibly aggravate the posterior humeral subluxation in patients with posterior eccentric glenoid wear.

Posterior approaches have already been used in the clinical setting to implant shoulder arthroplasties. Gagey et al., were able to implant 53 hemiarthroplasties through a posterolateral approach with subperiosteal detachment of the posterior part of the deltoid muscle and osteotomy of the external rotator muscles [13]. Although this approach provides a wide range of exposure of the glenoid and humeral head, deltoid release and detachment of the external rotators by means of an osteotomy are the main limitations, as they warrant postoperative immobilization of the arm and pose the risk for deltoid atrophy [13] and the insufficiency of external rotators. In contrast, the PI approach used in this study spares the deltoid and external rotators, and therefore, it allows for immediate postoperative mobilization. Greiwe et al., performed a total shoulder arthroplasty in 31 patients using a posterior rotator cuff-sparing approach, which uses a split of the middle and posterior heads of the deltoid muscle, a lateral based T-shaped capsular incision, and an in-situ humeral osteotomy [14]. Short-term follow up was available for 26 patients, with a significant improvement in clinical scores.

The authors also conducted an anatomic feasibility study to evaluate their approach, and this showed good access to glenoid and humerus, despite the mentioned technical difficulties [15]. While the deltoid split does not seem to affect deltoid integrity [14], the in-situ osteotomy of the humeral head, which is performed without dislocation of the head and via the internervous split, poses a surgical challenge, due to limited exposure and few bony landmarks for reference. As the identification of the anatomic neck for a precise humeral osteotomy may be difficult, there is a risk for an improper humeral cut, which can lead to malpositioning of the prosthetic humeral head, and potentially cause asymmetric loading of the glenoid, resulting in glenoid erosion and loosening [21–23]. According to the present anatomical study, a medial T-shaped incision of the capsule, with iuxtaglenoidal posterior to inferior capsular release, instead of a lateral T-shaped incision, may facilitate posteroinferior dislocation of the humeral head through the internervous interval, and thus allow for a precise identification of the anatomic neck, and easier humeral head osteotomy. However, grea<sup>t</sup> attention must be given not to stretch and harm the axillary nerve with the retractors placed inferiorly between the dislocated humeral head and the teres minor. Finally, it must be mentioned that even though the step of posterior dislocation of the humeral head can quite easily be obtained in cadaveric specimens, it might not be achievable in patients with severe joint stiffness due to advanced osteoarthritis.

A limitation of this study is the fact that the implantation of arthroplasties in cadaveric shoulders is typically easier, due to the reduced tension of the soft tissues. This might have facilitated the exposure and the implantation, especially in the PI approach group, as even in the cadaveric setting, only limited exposure could be obtained in some cases. Furthermore, not whole-body, but rather mere shoulder specimens were used for this study, making the placement easier to handle than what could be expected in clinical practice. While most measurement parameters have proven to be reliable, with acceptable ICCs, two parameters (pre-operative neck-shaft angle and the antero-posterior COR) showed only fair ICCs, thus limiting their interpretabilities. Finally, no conclusions regarding the risk of damage to the axillary nerve when performing the PI approach can be drawn from this study. Despite the apparently sufficient distance to the interval between the teres minor and the infraspinatus, no information on the changes in position and tension on the nerve during the posterior dislocation of the humeral head, humeral and glenoid exposure, as well as motion of the arm were collected.
