The study was conducted according to the Preferred Reporting Items for Systematic Review and Meta-Analysis Protocols (PRISMA-P) [15]. It also complies with the guidelines of Meta-Analyses and Systematic Reviews of Observational Studies (MOOSE) [16]. The study is registered in the PROSPERO (International Prospective Register of Systematic Reviews) database, with the identification number CRD42024540037.

A systematic search was conducted in PubMed, Scopus, Web of Science, and Cochrane Library in order to identify all relevant English-language studies published on this topic through April 2024. For the extraction of publications the subsequent terms were used: “transgender”, “FtM”, “female to male”, “trans men”, “transgender men”, “transmen”, “AFAB”, “t-AFAB”, “testosterone”, “gender affirming hormone therapy”, “GAHT”, “androgen”, “bone*”, “bone mass density”, “BMD”, “DEXA”, “hip”, “lumbar”, “femoral neck”, “bone metabolism markers”, “bone turnover markers”, “calcium”, “phosphate”, “vitamin D”, “25OHD”, “PTH”, “parathormone”, “Procollagen type 1 N-terminal propeptide”, “P1NP”, “bone alkaline phosphatase”, “osteocalcin”, “1CTP”, “carboxy-terminal telopeptide of type 1 collagen”, “c-terminal telopeptide of type 1 collagen”, “CTx”. To combine these key terms, Boolean AND/OR operators were used. Finally, eligible studies were identified through a systematic search, supplemented by a manual search of references cited in the retrieved articles. Full texts were obtained for studies with unclear relevance based on the abstracts. The obtained reference lists were also scrutinized to find possible additional pertinent studies.

The article selection process was carried out in several stages. In the first identification phase, database querying determined potentially eligible studies to include in the meta-analysis. Following the removal of duplicated articles tracked across multiple databases, in the second phase potential eligible papers were screened by reading their title and abstract. In the third phase, the remaining articles were evaluated in full-text for eligibility. Both prospective and retrospective observational studies, as well as longitudinal intervention studies, were deemed eligible, while non-experimental descriptive studies, studies conducted in populations other than the one of interest, studies in which endpoints other than those under analysis were evaluated, those with experimental designs other than the one of interest, and studies with incomplete or inaccurate data were excluded. The full-text of all selected studies was evaluated to determine their eligibility. The PRISMA flow-chart [17] was used to schematize the steps of article inclusion.

The methodological quality of the included studies was assessed using the Effective Public Health Practice Project (EPHPP) Quality assessment tool [18]. This assessment tool, used for intervention studies such as randomized controlled trials and case–control studies, has been validated for use in systematic reviews as well [19]. The tool considers the following domains: selection bias, study design, confounding factors, study blindness, data collection method, and loss to follow-up. The quality of each domain can be reported as strong (strong), moderate (moderate) or weak (weak), and in the overall judgment, the quality can be reputed as “strong” if no weak score was assigned, “moderate” if only a weak judgment was assigned to one of the domains, and finally “weak” if two or more weak judgments were assigned to multiple domains.

To minimize bias and ensure the reliability of the review process, two independent reviewers (D.T. and L.M.) were involved in study selection, data extraction, and quality assessment. Any discrepancies were resolved through discussion or consultation with a third reviewer (J.V.).

The primary outcome was to evaluate the differences in lumbar spine, hip, femoral neck and whole-body BMD values before and after one and two years of GAHT with various types of T formulations among TM.

The secondary outcomes were to evaluate the differences of bone metabolism-related parameters, before and after one and two years of GAHT among TM. The investigated biomarkers were the following: calcium, phosphate, PTH, 25OHD, bone-specific alkaline phosphatase (BPA), osteocalcin (OC), procollagen 1 intact n-terminal pro-peptide (P1NP) and C-telopeptide of type 1 collagen (CTx).

Additional information extracted was first author, year of publication, country/geographical region, study design, sample size, mean age, BMI and ethnicity, type of T therapy used, duration of follow-up in months, and parameters investigated in the study.

Changes in BMD values were assessed by calculating mean differences (MD), while standardized mean difference (SMD) was used for metabolic and turnover markers. In the presence of significant heterogeneity, data were combined using random effects models, which assumed that the included studies have varying effect sizes, thus providing a conservative estimate of the overall effect [20]. For nonsignificant heterogeneity, the results were pooled in a fixed effects model.

Publication biases were evaluated by the funnel plot graph [21]. The funnel plot was also subjected to Duval and Tweedie trim-and-fill test, to help detect presumed missing studies to rebalance the funnel distribution in the presence of a skewed shape. In addition, this analysis recalculates the combined estimate considering these putative identified studies corrected for publication bias [22].

To investigate potential moderators (covariates) and to examine the associations between the covariates and the outcomes, meta-regression analyses were conducted. Age and BMI before GAHT, and serum testosterone concentrations before and after 12 and 24 months of GAHT were investigated as potential moderators. A p-value < 0.05 is considered statistically significant.

For both primary and secondary outcomes, Cochran's Chi-square test (Cochran's Q) and I2 test were used for the purpose of analyzing statistical heterogeneity between the outcomes of different studies considering a value of I2 ≥ 50% and/or a value of P < 0.05 indicative of significant heterogeneity [23].

Data analysis was performed using the R statistical software equipped with the metafor package (version 3.6.3, 2020; The R Foundation for Statistical Computing, Vienna, Austria).