The present study included domestic and overseas clinical controlled studies that were published between January 2009 and January 2022. The patients were diagnosed with complex proximal humerus fractures (Neer three-part and four-part fractures) based on the patient's medical history (8), and physical and imaging examinations. All patients required shoulder arthroplasty. Non-case controlled studies, case reports, literature reviews, letters, duplicate reports, studies that included cases with a mean age of <60 years or those that did not provide sufficient relevant data were excluded. Intervention measures were HA replacement and RSA. Outcome indicators were Constant-Murley Shoulder Outcome Score, Disabilities of Arm, Shoulder and Hand (DASH) score, American Shoulder and Elbow Surgeons (ASES) score, Oxford Shoulder score, shoulder range of motion (ROM) and complications (9). The Preferred Reporting Items for Systematic Reviews and Meta-Analyses checklist was followed to perform the present meta-analysis (10).

The Embase (https://www.embase.com), Pubmed Central (https://pubmed.ncbi.nlm.nih.gov/), ProQuest Dissertations and Theses (http://pqdtopen.proquest.com/), Cochrane Library (https://www.cochranelibrary.com/) and Chinese Biomedical databases (www.sinomed.ac.cn) were searched. The directories of periodical and references were manually retrieved and the gray literature (such as chapters in unpublished academic papers, monographs, non-publicly published documents, government documents, dissertations, conference documents, scientific and technological reports, technical archives) was retrieved. Meanwhile, no language constrains were applied to extend the search to all relevant content, and the papers were thereafter translated if necessary. The key words were ‘hemiarthroplasty’, ‘HA’, ‘reverse shoulder arthroplasty’, ‘RSA’ and ‘proximal humeral fractures’, and the search strategy was (reverse shoulder arthroplasty OR RSA) AND (hemiarthroplasty OR HA) AND (proximal humeral fractures).

The included studies were independently analyzed by two physicians, and if disagreements occurred, they were resolved through discussion or handed over to a third senior physician to jointly determine the quality of the literature in strict accordance with the Cochrane risk of bias assessment criteria (11): i) Whether the experimental design adopts the random principle; ii) whether participants, performers and measurers use the double-blind principle; iii) whether the experimental data are complete and credible; iv) whether the allocation concealment method is adopted; v) whether the experiment adopts a selective data reporting method; and vi) other bias factors. According to the Newcastle-Ottawa scale (NOS) (12), the quality of the literature that met the inclusion criteria was evaluated.

Meta-analysis of the extracted data was performed using Review manager 5.4 software provided by the Cochrane Collaboration (https://training.cochrane.org/online-learning/core-software/revman). Dichotomous variables are expressed as the odds ratio and 95% CI. Continuous variables are expressed as the mean difference or standard mean difference and 95% CI. Heterogeneity among studies was tested by I² statistics, with an I² value of >50% indicating significant inconsistency. The reasons for the heterogeneity were analyzed and a random-effect model was used at this time, otherwise, a fixed-effect model was used. Sensitivity analyses were performed by removing certain studies and reanalyzing the data, and funnel plots were made to assess publication bias. P<0.05 was considered to indicate a statistically significant difference.

A total of 210 related publications were retrieved according to the aforementioned search strategy. By reading the title and abstract, 181 publications that were not controlled studies, or were repeated publications and irrelevant to the research purpose were excluded, and so 29 suitable publications passed initial screening. Following full text reading and screening in accordance with the inclusion and exclusion criteria, 16 studies were finally included. Baseline patient information, such as age and course of disease were compared among the included studies, which was comparable (P>0.05). The literature screening process and results are shown in Fig. 1 and the basic characteristics of the included literature studies are shown in Table I.

The present study included 1 randomized controlled trial, 5 prospective studies and 10 retrospective studies. To the best of our knowledge, the meta-analysis of retrospective, randomized controlled and prospective studies together will inevitably lead to certain biases in the results (13). However, when the literature was searched according to the search strategy, the number of randomized controlled studies and sample size was relatively small. Therefore, retrospective and prospective studies were included in the meta-analysis to increase the sample size. Consequently, NOS score was considered the most suitable for evaluating all the studies and the score was divided into low, medium and high quality grades, namely <5, 5-7 and 8-9 points (12). Among the literature, 3 papers scored nine points, 3 scored eight points, 4 scored seven points and 6 scored six points. Although the number of included studies was limited and there was a certain bias, the overall quality was moderate (Fig. 1 and Table I) (14-29).

In the present study, the Constant-Murley Shoulder Outcome Score was chosen as one of the outcomes of treatment. The Constant-Murley Shoulder Outcome Score is one of the most important scores for evaluating shoulder joint function (30); it mainly includes pain, daily activities, strength and shoulder joint ROM (31). A total of 6 studies reported the postoperative Constant-Murley Shoulder Outcome Score for RSA and HA in the treatment of elderly osteoporotic proximal humerus fractures, with scores divided into four subgroups according to pain, activity, strength and ROM. Due to the large heterogeneity (I²>50%) between the results and subgroups of each study, a random-effects model was used for the meta-analysis. The results showed that during the treatment of elderly osteoporotic proximal humeral fractures with joint replacement, the postoperative Constant-Murley Shoulder Outcome Score in the RSA group was improved compared with that in the HA group (95% CI, 1.69-3.76; P<0.001) and the difference was statistically significant. In terms of activity, the RSA group score was significantly improved compared with that of the HA group (95% CI, 2.25-4.70; P<0.001) and similarly, the RSA group outperformed the HA group for ROM (95% CI, 5.50-8.51; P<0.001). However, in the two subgroups of pain (95% CI, -0.17-4.58; P=0.07) and strength (95% CI, -1.10-2.32; P=0.48), the scores of the RSA and HA group were comparable (Fig. 2).

DASH score. A total of 5 studies provided detailed information on the DASH score. The random-effect model was also used for the meta-analysis due to the large heterogeneity among the results (I²=70%). After the meta-analysis, no significance was observed (95% CI, -7.34-2.02; P=0.27) for DASH score (Fig. 3).

ASES score. A total of 5 studies paid close attention to the ASES score for the treatment of elderly osteoporotic proximal humerus fractures. As there was little heterogeneity among the results of the studies (I²<50%), a fixed-effects model was used for the meta-analysis. The mean ASES score of the RSA group was higher than that in the HA group, and the difference was statistically significant (95% CI, 11.81-24.88; P<0.001; Fig. 4).

Oxford Shoulder score. A total of 3 publications compared the postoperative Oxford Shoulder score between the RSA and HA groups. Due to the large heterogeneity (I²>50%) among the results of each study, the random-effects model was used for the meta-analysis. The results showed that the mean Oxford Shoulder score of the RSA group was higher when treating elderly osteoporotic proximal humerus fractures, and the difference was statistically significant (95% CI, 2.89-11.11; P<0.001; Fig. 5).

Shoulder ROM. A total of 9 studies compared the postoperative shoulder ROM of RSA and HA in the treatment of elderly osteoporotic proximal humerus fractures, with the scores divided into four subgroups according to anterior elevation, abduction, external rotation and internal rotation. Due to the large heterogeneity (I²>50%) among the results, the random-effect model was used for the meta-analysis. The results showed that during the treatment of elderly osteoporotic proximal humerus fractures with shoulder arthroplasty, the shoulder ROM in the RSA group was improved compared with that in the HA group (95% CI, 3.41-9.07; P<0.001) with statistically significant differences, especially with regard to the anterior elevation (95% CI, 18.27-39.38; P<0.001) and abduction (95% CI, 24.24-41.26; P<0.001). The meta-analysis for the other two subgroups involving external (95% CI, -5.33-1.12; P=0.20) and internal (95% CI, -1.86-1.06; P=0.59) rotations did not find a significant difference between both methods (Fig. 6).

Complications. There are several possible postoperative complications in the treatment of elderly osteoporotic proximal humerus fractures with shoulder arthroplasty. The 8 included studies compared the postoperative complications of RSA and HA and divided them into 13 complication subgroups, including arthrofibrosis, greater tuberosity osteolysis, malunion greater tuberosity, diaphyseal fracture of humerus, infection, heterotopic bone, loosening, secondary cuff rupture, deep venous thrombosis, instability/dislocation, complex regional pain syndrome, malposition and pulmonary embolism. Due to the large heterogeneity (I²>50%) among the results of each study, the random-effect model was used for the meta-analysis. The results showed that in the deep venous thrombosis subgroup, the HA group had a lower incidence rate than the RSA group (95% CI, 1.21-5.68; P=0.01). No significant differences were observed in the other subgroups or in terms of complications overall (95% CI, 0.59-1.20; P=0.35; Fig. 7).

The Review Manager 5.4 statistical software provided by the Cochrane Collaboration was used to analyze the publication bias of the Constant-Murley Shoulder Outcome Score, DASH score, ASES score, Oxford Shoulder score, shoulder ROM and complications after the shoulder arthroplasty. The results showed that the funnel plots were basically symmetrical, indicating that there was no obvious publication bias (Fig. 8). The analysis showed that there was high heterogeneity in the Constant-Murley Shoulder Outcome Score subgroups (pain, strength and ROM), DASH score, Oxford shoulder score and shoulder ROM subgroups (anterior elevation and internal rotation). The heterogeneity was reduced to 0% after the study by Bonnevialle et al (15) was excluded from the meta-results of Constant-Murley Shoulder Outcome Score-ROM. Due to strict compliance with the search and inclusion criteria, the inclusion of certain results in the literature is limited, and it would be difficult to complete the meta-analysis if certain studies were excluded. After further analyzing the heterogeneous sources using the one-by-one elimination method, the results of the meta-analysis showed no directional changes, indicating that the stability of the research results was good (Figs. 9 and 10).