跳到正文
原文
Nature 健康护理主题·· 18 天前精选AI 评分79

绝经后骨质疏松治疗药物的相对疗效与安全性网络元分析

Comparative efficacy and safety of pharmacological interventions for postmenopausal osteoporosis: a network meta-analysis

AI 导读

该研究通过网络元分析评估了多种药物治疗绝经后骨质疏松的有效性和安全性,结果显示不同药物在不同骨密度指标上的效果存在差异,为临床选择提供了依据。

推荐理由

该研究通过网络元分析综合评估了多种药物治疗绝经后骨质疏松的有效性和安全性,为临床决策提供了重要依据。

正文

Introduction

The advancing age of global populations has established osteoporosis as a significant concern for public health systems worldwide, primarily due to its high incidence and the severe, costly complications it engenders1. Osteoporosis is a prevalent condition defined by a reduction in bone mass and the degradation of bone microarchitecture, which subsequently elevates the susceptibility to fragility fractures.2. Osteoporosis-related hip fractures are particularly devastating, accounting for 5% of all-cause mortality across both genders, with 21–30% of affected individuals dying within one year3. Following menopause, the cessation of ovarian function and the consequent rapid decline in estrogen levels lead to accelerated bone loss, culminating in postmenopausal osteoporosis (PMOP)4. Currently, research into the pathogenesis of osteoporosis has expanded beyond orthopedics into the fields of neuroscience and gut microbiota. Neural and gastrointestinal influences on skeletal mass regulation, as well as the brain-gut-bone axis regulatory mechanism, have become international research hotspots and challenges in recent years5,6,7, and significant breakthroughs have been made. However, the exact mechanisms remain unclear, and safe and effective treatments are still lacking.

Multiple pharmacological options are available for PMOP, which fall into two main categories: anti-resorptive agents and anabolic agents. These include bisphosphonates, Selective Estrogen Receptor Modulators (SERMs), calcitonin, parathyroid hormone (PTH) analogs, denosumab, as well as calcitriol. Bisphosphonates are selectively absorbed by the skeleton and are slowly released after treatment cessation, allowing for a sustained effect post-discontinuation, making them the most widely used anti-resorptive medications8. Estrogen-based therapies can (1) reduce bone's sensitivity to PTH, decreasing bone resorption; (2) increase calcitonin production, inhibiting bone resorption; (3) accelerate intestinal calcium absorption, thereby increasing serum calcium concentration and promoting bone mass gain; (4) reduce renal calcium excretion, further increasing serum levels; (5) directly modulate bone remodeling, a process mediated through estrogen receptors expressed within bone tissue9. By antagonizing the receptor activator of nuclear factor kappa-B ligand (RANKL), denosumab downregulates the formation and activity of osteoclasts, consequently suppressing bone resorption10. PTH analogs, such as teriparatide, directly stimulate osteoblast formation, thereby increasing bone mineral density11. Concurrently, PTH regulates key anabolic factors such as IGF-1, while also suppressing antagonists like sclerostin, thereby stimulating osteogenesis12. Oral calcitriol (0.25–0.50 mg/d) increases intestinal calcium absorption in women with PMOP13. However, most existing primary studies and meta-analyses compare only two different drugs for treating PMOP.

While numerous drugs are available for PMOP, clinical decision-making is hampered by the lack of comparative evidence. The existing literature compared one drug against placebo, leaving a critical gap in direct comparisons between different active agents. This study utilized a network meta-analysis(NMA) to comprehensively evaluate their relative efficacy and safety and also provided a quantitative ranking of these interventions based on outcome indicators. This aimed to offer evidence to support clinicians in prescription choices and medical decision-making.

Materials and methods

This systematic review and NMA adhered to the Preferred Reporting Items for Systematic Review and Meta-Analyses for Network Meta-Analyses (PRISMA-NMA) guidelines. The completed checklist is provided in Supplementary Material 1. The study protocol was prospectively registered with PROSPERO (CRD420251149474).

Data sources and search strategy

We searched four English databases (PubMed, Embase, the Cochrane Library, and Web of Science) from their inception to March 5, 2026. The search strategy consisted of three main components: (1) Population (e.g., "post-menopausal bone loss", "post-menopausal osteoporosis", "post-menopause bone loss", "post-menopause osteoporosis"); (2) Intervention (e.g., "Alendronate", "Raloxifene", "Calcitonin", "Teriparatide", "Denosumab", "Calcitriol"); and (3) Study type (e.g., "Randomized Controlled Trial", "randomized", "placebo"). Only studies that were published in English were eligible for inclusion. The complete set of search queries is available in Supplementary Material 2.

Inclusion criteria

Publications were considered eligible for inclusion based on the criteria outlined below: (1) Study types: RCTs. (2) Population: Eligible participants were women aged ≥ 40 years with a diagnosis of postmenopausal osteoporosis, defined as either: the presence of a radiographic vertebral fracture or a recent (within 5 years) nonvertebral fracture, accompanied by a bone mineral density(BMD) T-score ranging from − 2.5 to − 5.0 in women ≤ 65 years or from − 2.0 to − 5.0 in those > 65 years; or in women over 65 years without previous fracture, a lumbar spine BMD(LS-BMD) or femoral neck BMD (FN-BMD) T-score between − 3.0 and − 5.0. (3) Interventions: At least one intervention arm involved the use of bisphosphonates, SERMs, calcitonin, parathyroid hormone analogs, denosumab, or calcitriol. Within the evidence network, specific nodes represented concurrent combination therapy, defined as the simultaneous administration of two pharmacological agents during the study period. Table 1 details the composition of each combination node. Sequential therapy was not included in our study. (4) Outcomes: Studies were required to report one or more of the specified outcome measures. Primary outcomes included changes in LS-BMD, FN-BMD, total hip BMD(TH-BMD), Trochanter BMD(TRO-BMD), C-terminal telopeptide of type I collagen (CTX), and N-propeptide of type I collagen(P1NP). Secondary endpoints encompassed the occurrence of adverse events (AEs). Adverse events were defined as any unfavorable medical occurrences arising during the treatment period, regardless of their suspected relationship to the study drug. Our study analyzed only serious adverse reactions and fractures. Serious adverse events refer to adverse medical events such as death, life-threatening, permanent or serious disability or loss of function, hospitalization or prolongation of hospitalization. Vertebral fractures were assessed using sequential lateral spine radiographs (e.g., at baseline and annually) with central review. Incident vertebral fractures were defined as an increase of ≥ 1 grade in the Genant semiquantitative score in vertebrae graded as normal (Grade 0) at baseline. Prevalent vertebral fractures were defined as those with a Genant grade of ≥ 1 at baseline. Nonvertebral fractures included low-trauma fragility fractures, excluding fractures of the skull, face, mandible, metacarpals, fingers, and toes, as well as those resulting from high-energy trauma or underlying pathological conditions.

Table 1 Definitions of the combined therapies included in this study.

Full size table

Exclusion criteria

The exclusion criteria were as follows: (1) Reviews or meta-analyses. (2) Case reports, conference abstracts, clinical guidelines, letters, editorials, short surveys, notes, preprints, or retracted studies. (3) Duplicate publications. (4) Interventions or diseases not relevant to this study. (5) Full text unavailable. (6) Animal studies. (7) Non-English publications. (8) Ongoing or registered studies without results. (9) Lack of relevant outcome measures or data. (10) cohort studies.

Literature screening and data extraction

All retrieved citations were collated into EndNote 2021, and duplicate studies were removed using its deduplication function. A dual independent review process was employed. First, titles and abstracts were screened for relevance. The full texts of potentially eligible studies were then procured and appraised against the inclusion and exclusion criteria by both reviewers. We performed quality evaluation and data extraction only on those studies that satisfied the inclusion criteria, in preparation for the subsequent NMA. Data extraction was performed using a customized template in Excel to systematically collect data from all eligible studies. The following data were extracted: first author, year of publication, country of publication, study design, participant demographics (total sample size, age), details of interventions(sample size and age for each treatment group), duration of intervention, outcomes.

A dual independent review process was employed for both the screening of literature and the extraction of data. Any discrepancies in screening or extraction were resolved by consensus or, when required, by arbitration from a third reviewer.

Risk of bias assessment

Two reviewers (Jiamei Zhuang, Bodi Liu) independently evaluated the methodological quality of each included trial utilizing the Cochrane Risk of Bias tool (RoB 1.0)14. The appraisal encompassed following key domains: selection bias (random sequence generation and allocation concealment), performance bias (blinding of participants and personnel), detection bias (blinding of outcome assessment), attrition bias (incomplete outcome data), reporting bias (selective reporting), and other potential sources of bias. A judgment of 'low risk', 'high risk', or 'unclear' was assigned to each category. All disagreements between reviewers were settled by consensus. To evaluate the methodological rigor of the included trials, the modified Jadad scale was employed15. Key domains such as random sequence generation, allocation concealment, blinding procedures, and handling of withdrawals were appraised. Trials scoring between 1 and 3 were classified as low quality, whereas scores of 4 to 7 indicated high quality.

Statistical analysis methods

All statistical analyses were conducted with Stata17.0 (with metan and metareg) and Rstudio4.3.1(with packages gemtc, rjags, and coda).

Network construction and descriptive analysis

Data preparation and network structuring were conducted using Stata. For dichotomous outcome variables, the relative risk (RR) along with its 95% credible interval (CrI) was used as the effect measure to estimate the relative efficacy between interventions. For continuous variables, the standardized mean difference (SMD) was employed. A network graph was first developed to illustrate the direct and indirect comparisons among interventions and to assess network connectivity.

Network meta-analysis model

The NMA was performed within a Bayesian framework using R. Non-informative prior distributions were specified for all model parameters. For continuous outcomes, a normal likelihood with an identity link was assumed; for binary outcomes, a binomial likelihood with a logit link was used. The models were run with four Markov chains, with 20,000 adaptation iterations, 50,000 posterior sampling iterations, and a thinning interval of 1. Convergence was assessed using the Gelman-Rubin diagnostic, with a potential scale reduction factor (PSRF) < 1.05 considered acceptable for all parameters. A random-effects model was employed to account for anticipated between-study heterogeneity.

The NMA was performed under the assumptions of homogeneity, transitivity, and consistency. Homogeneity was evaluated using the I2 statistic derived from the Bayesian model in R. Transitivity was considered plausible based on the clinical similarity of the included trials in terms of patient populations, interventions, and outcome definitions. Consistency was formally tested as described below.

Consistency assessment

To evaluate the consistency assumption, we first fitted both consistency and inconsistency models (UME: unrelated mean effects) within the Bayesian framework using R. The deviance information criterion (DIC) was used to compare the goodness-of-fit between the two models. A DIC difference of less than 5 between the consistency and inconsistency models was considered to support the use of the consistency model for the primary analysis. When the UME model showed a substantially lower DIC (difference ≥ 5), the inconsistency model was considered to provide a better fit, and the results were interpreted with caution.

Additionally, global inconsistency was assessed using the design-by-treatment interaction model in Stata, with P < 0.05 indicating significant inconsistency. The corresponding forest plot was also generated in Stata. Local inconsistency was evaluated using a loop-specific approach for closed loops. For outcomes with open network structures (i.e., no closed loops), formal inconsistency testing was not applicable; in such cases, the consistency model was assumed for the primary analysis. The results of loop inconsistency were generated by Stata (see Supplementary Material 5). The global inconsistency test was performed using R (Supplementary Material 4), while the forest plot for global inconsistency was constructed in Stata (Supplementary Material 6).

Heterogeneity assessment

Between-study heterogeneity was quantified using the I2 statistic and the between-study standard deviation (tau), both derived from the Bayesian NMA model in R. I2 values ≤ 50% were considered to indicate low heterogeneity, whereas values > 50% suggested substantial heterogeneity. Heterogeneity results were shown in Supplementary Material 4. To explore potential sources of heterogeneity, we performed subgroup analyses stratified by drug class when heterogeneity was substantial or clinically indicated. Detailed subgroup analysis results were provided in Supplementary Material 8.

Ranking and presentation of treatment effects

To evaluate the relative superiority of the different interventions, the SUCRA values were calculated for each intervention, and ranking plots were generated. A greater SUCRA value corresponded to a superior ranking of the intervention's performance in the network comparisons. League tables were also constructed to present the effect sizes and corresponding confidence intervals for all interventions relative to the reference control.

Publication bias assessment

The possibility of publication bias was assessed by generating comparison-adjusted funnel plots and examining their symmetry.

Statistical significance

In this study, statistical significance was defined as a 95% CrI for RR that did not include 1, or a 95% CrI for SMD that did not include 0.

Sensitivity analyses

To assess the robustness of the primary analysis results, we conducted two sensitivity analyses: (1) a fixed-effects model was applied to evaluate the impact of model assumptions; and (2) studies with non-blinding of participants and personnel were excluded to assess the influence of study quality on the findings. The robustness of the results was evaluated by comparing effect estimates, 95% CrIs, SUCRA rankings, and the consistency of the top three interventions for each outcome across the different analyses. Results were considered robust if the direction of effect and statistical significance remained unchanged, with only minor fluctuations in rank order. Conversely, if the statistical significance changed (e.g., from significant to non‑significant, or vice versa), or if the composition of the top three interventions changed substantially, the corresponding outcome was interpreted with caution. The detailed results were displayed in Supplementary Material 9.

Results

Literature search

We identified 10,044 records from the databases and selected 182 potentially eligible records for full-text review. A total of 141 records were excluded for the following reasons: meta-analyses and reviews (n = 47), duplicate records(n = 9), case reports, conference abstracts, guidelines, etc. (n = 2), irrelevant interventions or diseases (n = 41), animal studies (n = 1), cohort study (n = 6), non-English publications (n = 3), trial registrations (n = 18), and lack of relevant data or outcomes (n = 14). Ultimately, 41 articles met the inclusion criteria16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56. The process of study identification, screening, and selection is illustrated in Fig. 1. All included studies are RCTs.

Fig. 1

PRISMA flow diagram.

Characteristics of included studies

A total of 41 RCTs were included. The included studies were conducted across 17 countries. Specifically, 15 studies were conducted in the USA, 4 in Japan, 3 each in China, and Denmark, 2 each in Canada, Poland, the UK and the Netherlands, and 1 each in Australia, Finland, Austria, , Germany, Greece, Israel, New Zealand and Spain. The sample sizes ranged from 22 to 7,868 female participants, all aged ≥ 40 years. The follow-up duration ranged from 1 to 60 months. The definition of osteoporosis was consistent across the trials. The included studies involved 6 categories of therapeutic drugs: bisphosphonates were investigated in 28 studies, SERMs in 4 studies, calcitonin in 5 studies, PTH and PTH analogs in 9 studies, denosumab in 5 studies, and calcitriol in 2 studies. BMD was reported in 29 studies, BTMs in 21 studies, fractures in 10 studies, and AEs in 21 studies. The detailed characteristics of the included studies are presented in Supplementary Material 3.

Risk of bias assessment

For the randomized controlled trials, the Jadad scale results indicated that 2 studies scored 3 points, classifying them as low quality. 39 studies scored > 3 points, classifying them as high quality. The Cochrane Risk of Bias Tool results showed that all included studies described random sequence generation and were rated as low risk for this domain. Allocation concealment was implemented and rated as low risk in 19 studies; 22 studies did not mention allocation concealment and were rated as having an unclear risk. Blinding of investigators and participants was reported in 27 studies, and blinding of outcome assessors was mentioned in 12 studies; these were rated as low risk. 9 studies were identified as open-label for investigators and participants and were rated as high risk for performance and detection bias; other studies that did not mention blinding were rated as unclear risk. 37 studies had complete outcome data and were rated as low risk for attrition bias; 4 studies had insufficient reporting of withdrawals and dropouts, making judgment difficult, and were rated as unclear risk. No study showed evidence of selective reporting or other biases, and all were rated as low risk for these domains. Detailed results of the risk of bias assessment for all included studies are shown in Fig. 2 and Fig. 3.

Fig. 2

Risk of Bias graph.

Fig. 3

Risk of bias summary.

Network meta-analysis

Lumbar spine bone mineral density (LS-BMD)

A total of 30 studies investigated the effects of 7 drug categories on LS-BMD, with 21 studies included in the analysis. These included 16 studies on bisphosphonates (alendronate: 7, ibandronate: 2, minodronate: 3, pamidronate: 1, etidronate: 1, risedronate:2), 2 study on SERMs (raloxifene and bazedoxifene), 6 studies on calcitonin, 2 studies on PTH analogs (abaloparatide and teriparatide), 4 studies on denosumab (include denosumab biosimilar GP2411), and 14 studies involving placebo. Comparing the post-treatment LS-BMD, the highest number of studies involved placebo, followed by alendronate and calcitonin. The comparison between placebo and calcitonin was the most frequently studied. 20 direct comparison pairs were formed between the various drugs. Two closed loops(denosumab–placebo–ibandronate, placebo-alendronate–minodronate-calcitonin_minodronate-calcitonin) formed among the interventions, indicating the presence of indirect evidence. A network of comparisons between interventions is shown in Fig. 4A.

Fig. 4

Network meta-analysis of LS-BMD. (A) Network plot of the network meta-analysis for LS-BMD. (B) Rankogram of overall intervention strategies from the network meta-analysis for LS-BMD. (C) League table of the network meta-analysis for LS-BMD.

Based on the SUCRA values, calcitonin_minodronate (SUCRA = 91%) ranked first for increasing LS-BMD among the 16 interventions, followed by abaloparatide (SUCRA = 83.7%) and teriparatide (SUCRA = 75.5%). Rankogram of overall intervention strategies is displayed in Fig. 4B.

Compared to calcitonin_minodronate, risedronate (SMD = − 1.17, 95% CrI: − 2.21 ~ − 0.13), raloxifene (SMD = − 1.42, 95% CrI: − 2.56 ~ − 0.29), placebo (SMD = − 1.86, 95% CrI: − 2.79 ~ − 0.94), minodronate (SMD = − 1.01, 95% CrI: − 1.89 ~ − 0.13), etidronate (SMD = − 1.66, 95% CrI: − 3.02 ~ − 0.31) resulted in smaller increase in LS-BMD. Relative to calcitonin, greater increase in LS-BMD was observed with teriparatide (SMD = 1.18, 95% CrI: 0.49 ~ 1.87), pamidronate (SMD = 0.96, 95% CrI: 0.02 ~ 1.90), ibandronate (SMD = 0.71, 95% CrI: 0.07 ~ 1.35), denosumab (SMD = 1.00, 95% CrI: 0.47 ~ 1.53), calcitonin_minodronate (SMD = 1.62, 95% CrI: 0.73 ~ 2.52) whereas placebo (SMD = − 0.24, 95% CrI: − 0.55 ~ 0.07) exhibited less. When compared with abaloparatide, risedronate (SMD = − 0.88, 95% CrI: − 1.72 ~ − 0.04), raloxifene (SMD = − 1.13, 95% CrI: − 2.09 ~ − 0.17), placebo (SMD = − 1.58, 95% CrI: − 2.20 ~ − 0.96), etidronate (SMD = − 1.38, 95% CrI: − 2.54 ~ − 0.21), calcitonin (SMD = − 1.34, 95% CrI: − 2.03 ~ − 0.64) were associated with a less pronounced increase in LS-BMD. In comparison to placebo, the increase in LS-BMD was greater for teriparatide (SMD = 1.42, 95% CrI: 0.80 ~ 2.04) and risedronate (SMD = 0.70, 95% CrI: 0.12 ~ 1.27). Compared to teriparatide, raloxifene (SMD = 0.98, 95% CrI: 0.02 ~ 1.94), etidronate (SMD = 1.22, 95% CrI: 0.06 ~ 2.38) resulted in smaller increase in LS-BMD. Using placebo as the reference, pamidronate (SMD = − 1.20, 95% CrI: − 2.09 ~ − 0.32), minodronate (SMD = − 0.86, 95% CrI: − 1.55 ~ − 0.16), and ibandronate (SMD = − 0.95, 95% CrI: − 1.51 ~ − 0.39), denosumab (SMD = − 1.24, 95% CrI: − 1.68 ~ − 0.81), bazedoxifene (SMD = − 0.98, 95% CrI: − 1.89 ~ − 0.07), alendronate (SMD = − 0.99, 95% CrI: − 1.35 ~ − 0.62), PTH (SMD = − 1.35, 95% CrI: − 1.97 ~ − 0.74), GP2411 (SMD = − 1.24, 95% CrI: − 2.02 ~ − 0.47) demonstrated a increased effect on LS− BMD gain. Compared with calcitonin, the LS-BMD increase was reinforced with alendronate (SMD = − 0.74, 95% CrI: − 1.16 ~ − 0.33), PTH (SMD = − 1.11, 95% CrI: − 1.80 ~ − 0.42) and GP2411 (SMD = − 1.00, 95% CrI: − 1.84 ~ − 0.17). The SMD values are shown in Fig. 4C.

A non-significant global inconsistency was observed (P = 0.615), accompanied by small heterogeneity (I2 = 5%). Of the three closed loops in the evidence network, two (alendronate–calcitonin–minodronate, alendronate–calcitonin- placebo) exhibited 95% CrI that included zero, indicating no significant loop inconsistency. The forest plot for global inconsistency is presented in Supplementary Material 6 (Figure S1.1), while the forest plot for LS-BMD is shown in Supplementary Material 7 (Figure S2.1).

Femoral neck bone mineral density (FN-BMD)

There were 24 studies that examined the impact of 7 medications on FN-BMD, among which 15 were incorporated into the analysis. These included 12 studies on bisphosphonates (risedronate: 2, alendronate: 5, minodronate: 2, etidronate: 1, ibandronate: 1, pamidronate: 1), 4 studies on denosumab (include GP2411), 3 studies on PTH analogs (teriparatide: 1, abaloparatide: 1, PTH: 1), 5 studies on calcitonin, 1 study on SERMs (raloxifene), and 9 studies involving placebo. Comparing the post-treatment FN-BMD, the highest number of studies involved placebo, followed by alendronate and calcitonin. The comparisons between placebo and calcitonin was the frequentest. 17 direct comparison pairs were formed between the various drugs. A closed loops (placebo- calcitonin- alendronate) formed among the interventions, indicating the presence of indirect evidence. A network of comparisons between interventions is shown in Fig. 5A.

Fig. 5

Network meta-analysis of FN-BMD. (A) Network plot of the network meta-analysis for FN-BMD. (B) Rankogram of overall intervention strategies from the network meta-analysis for FN-BMD. (C) League table of the network meta-analysis for FN-BMD.

As indicated by the SUCRA rankings, among the 15 interventions, calcitonin_minodronate (SUCRA = 87.7%) was the most effective intervention for improving FN-BMD, with denosumab (SUCRA = 86.3%) and GP2411 (SUCRA = 84.7%) ranking second and third, respectively. Figure 5B presents the corresponding rankogram.

Relative to calcitonin, teriparatide (SMD = 0.60, 95% CrI: 0.20 ~ 1.01), risedronate (SMD = 0.58, 95% CrI: 0.24 ~ 0.93), raloxifene (SMD = 0.51, 95% CrI: 0.11 ~ 0.90), ibandronate (SMD = 0.76, 95% CrI: 0.13 ~ 1.40), denosumab (SMD = 1.01, 95% CrI: 0.42 ~ 1.61), calcitonin_minodronate (SMD = 1.16, 95% CrI: 0.43 ~ 1.90) were associated with a greater increase in FN‑BMD. When benchmarked against GP2411, the increase of FN-BMD was lower for risedronate (SMD = − 1.27, 95% CrI: − 1.82 ~ − 0.72), placebo (SMD = − 1.01, 95% CrI: − 1.97 ~ − 0.05), and minodronate (SMD = − 1.27, 95% CrI: − 2.24 ~ − 0.31), alendronate (SMD = − 0.68, 95% CrI: − 1.10 ~ − 0.26) but higher for denosumab (SMD = − 1.01, 95% CrI: − 1.67 ~ − 0.36). Relative to placebo, a larger increase in FN-BMD was observed with teriparatide (SMD = 0.86, 95% CrI: 0.66 ~ 1.07), risedronate (SMD = 0.85, 95% CrI: 0.45 ~ 1.24), and raloxifene (SMD = 0.77, 95% CrI: 0.33 ~ 1.21). In a head-to-head comparison with teriparatide, etidronate (SMD = 0.86, 95% CrI: 0.05 ~ 1.68) exhibited smaller increase in FN-BMD, whereas abaloparatide (SMD = − 0.30, 95% CrI: − 0.50 ~ − 0.10) exhibited more. In comparison to placebo, ibandronate (SMD = − 1.02, 95% CrI: − 1.55 ~ − 0.50), denosumab (SMD = − 1.27, 95% CrI: − 1.75 ~ − 0.80), calcitonin_minodronate (SMD = − 1.42, 95% CrI: − 2.24 ~ − 0.61), alendronate (SMD = − 0.95, 95% CrI: − 1.31 ~ − 0.58), abaloparatide (SMD = − 1.16, 95% CrI: − 1.37 ~ − 0.96) and PTH (SMD = − 0.78, 95% CrI: − 0.98 ~ − 0.58) led to a more pronounced improvement in FN‑BMD. Using minodronate as the reference, denosumab (SMD = − 1.01, 95% CrI: − 1.93 ~ − 0.09), calcitonin_minodronate (SMD = − 1.16, 95% CrI: − 1.89 ~ − 0.44), abaloparatide (SMD = − 0.90, 95% CrI: − 1.72 ~ − 0.09) demonstrated a superior increase in FN‑BMD. When compared with ibandronate, etidronate (SMD = 1.02, 95% CrI: 0.08 ~ 1.97) resulted in smaller increase in FN-BMD, while denosumab (SMD = − 0.25, 95% CrI: − 0.47 ~ − 0.03) yielded a higher FN‑BMD gain. Relative to etidronate, the increase in FN‑BMD was greater with denosumab (SMD = − 1.27, 95% CrI: − 2.20 ~ − 0.35), calcitonin_minodronate (SMD = − 1.42, 95% CrI: − 2.56 ~ − 0.29), alendronate (SMD = − 0.95, 95% CrI: − 1.82 ~ − 0.08) and abaloparatide (SMD = − 1.16, 95% CrI: − 1.98 ~ − 0.35). In contrast to calcitonin, alendronate (SMD = − 0.69, 95% CrI: − 1.00 ~ − 0.38), abaloparatide (SMD = − 0.90, 95% CrI: − 1.31 ~ − 0.49), PTH (SMD = − 0.52, 95% CrI: − 0.93 ~ − 0.11) resulted in a more substantial elevation of FN‑BMD. Compared to abaloparatide, PTH (SMD = 0.38, 95% CrI: 0.09 ~ 0.67) resulted in smaller increase in FN-BMD. The SMD values are shown in Fig. 5C.

Despite low heterogeneity (I2 = 6%), significant global inconsistency was observed (P = 0.006). A loop (placebo- calcitonin- alendronate) in the network showed significant loop inconsistency, as their 95% CrIs included zero. Additionally, the significant global inconsistency for FN-BMD was primarily driven by conflicting evidence within the alendronate subgroup (I2 = 92.4%), whereas the calcitonin subgroup showed perfectly consistent null effects (I2 = 0.0%). This suggests that inconsistencies may originate from substantial between-study heterogeneity in alendronate trials, possibly due to variations in dosage, treatment duration, or baseline FN-BMD levels. Figure S1.2 in Supplementary Material 6 shows the forest plot for global inconsistency, Figure S2.2 in Supplementary Material 7 shows the forest plots for FN-BMD and Figure S3.1 in Supplementary Material 8 shows subgroup analysis results.

Total hip bone mineral density (TH-BMD)

In total, 24 studies assessed the efficacy of 7 drugs on TH-BMD, of which 16 were included in the NMA. These included 12 studies on bisphosphonates (alendronate: 5, risedronate: 2, minodronate: 2, ibandronate: 2, etidronate: 1), 5 studies on denosumab(include GP2411), 3 studies on PTH analogs (teriparatide: 1, abaloparatide: 1; PTH:1), 4 studies on calcitonin, 1 study on SERMs (raloxifene: 1), and 10 studies involving placebo. Comparing the post-treatment TH-BMD, the highest number of studies involved placebo, followed by alendronate. The comparison between denosumab and placebo, between placebo and calcitonin, between between placebo and alendronate ,between alendronate and risedronate were relatively frequent. 16 direct comparison pairs were formed between the various drugs. One closed loops(denosumab-placebo-ibandronate,) formed among the interventions, indicating the presence of indirect evidence. A network of comparisons between interventions is shown in Fig. 6A.

Fig. 6

Network meta-analysis of TH-BMD. (A) Network plot of the network meta-analysis for TH-BMD. (B) Rankogram of overall intervention strategies from the network meta-analysis for TH-BMD. (C) League table of the network meta-analysis for TH-BMD.

According to the SUCRA analysis, among the 14 interventions, abaloparatide (SUCRA = 89.8%) emerged as the top-ranked intervention for TH-BMD increase, followed by denosumab (SUCRA = 85.1%) and GP2411 (SUCRA = 83.4%). A visual representation of the rankogram is provided in Fig. 6B.

Relative to teriparatide, a smaller increase in TH-BMD was observed with risedronate (SMD = 0.85, 95% CrI: 0.28 ~ 1.42), raloxifene (SMD = 0.81, 95% CrI: 0.17 ~ 1.45), placebo (SMD = 0.86, 95% CrI: 0.48 ~ 1.24) and calcitonin (SMD = 0.66, 95% CrI: 0.19 ~ 1.14). In comparison to ibandronate, the increase in TH-BMD was smaller for risedronate (SMD = − 0.74, 95% CrI: − 1.31 ~ − 0.18), raloxifene (SMD = − 0.70, 95% CrI: − 1.34 ~ 0.07) and placebo (SMD = − 0.76, 95% CrI: − 1.14 ~ − 0.38). Compared with denosumab, the TH-BMD increase was attenuated with risedronate (SMD = − 1.06, 95% CrI: − 1.55 ~ − 0.57), raloxifene (SMD = − 1.02, 95% CrI: − 1.59 ~ − 0.45), placebo (SMD = − 1.07, 95% CrI: − 1.33 ~ − 0.82), minodronate (SMD = − 0.87, 95% CrI: − 1.73 ~ − 0.01) and etidronate (SMD = − 0.99, 95% CrI: − 1.88 ~ − 0.10). Taking calcitonin as the comparator, ibandronate (SMD = 0.56, 95% CrI: 0.09 ~ 1.02) and denosumab (SMD = 0.87, 95% CrI: 0.49 ~ 1.25) showed a larger positive effect on TH-BMD. In a head-to-head comparison with alendronate, the probability of TH-BMD increase was lower with risedronate (SMD = − 0.38, 95% CrI: − 0.65 ~ − 0.10), placebo (SMD = − 0.39, 95% CrI: − 0.71 ~ − 0.07) and higher with denosumab (SMD = 0.69, 95% CrI: 0.28 ~ 1.09). Relative to abaloparatide, risedronate (SMD = − 1.15, 95% CrI: − 1.72 ~ − 0.58), raloxifene (SMD = − 1.11, 95% CrI: − 1.75 ~ − 0.47), placebo (SMD = − 1.16, 95% CrI: − 1.54 ~ − 0.78), minodronate (SMD = − 0.96, 95% CrI: − 1.87 ~ − 0.05), etidronate (SMD = − 1.08, 95% CrI: − 2.01 ~ − 0.14), calcitonin (SMD = − 0.96, 95% CrI: − 1.44 ~ − 0.48) and alendronate (SMD = − 0.77, 95% CrI: − 1.27 ~ − 0.28) yielded a lower magnitude of TH-BMD improvement. In contrast to PTH, risedronate (SMD = − 1.05, 95% CrI: − 1.63 ~ − 0.47), raloxifene (SMD = − 1.01, 95% CrI: − 1.66 ~ − 0.36), placebo (SMD = − 1.07, 95% CrI: − 1.47 ~ − 0.66), etidronate (SMD = − 0.98, 95% CrI: − 1.92 ~ − 0.04), calcitonin (SMD = − 0.86, 95% CrI: − 1.35 ~ − 0.37) and alendronate (SMD = − 0.68, 95% CrI: − 1.19 ~ − 0.16) resulted in a less substantial increase in TH-BMD. Taking GP2411 as the comparator, the effect on TH-BMD was smaller for risedronate (SMD = − 1.06, 95% CrI: − 1.71 ~ − 0.41), raloxifene (SMD = − 1.02, 95% CrI: − 1.73 ~ − 0.31), placebo (SMD = − 1.07, 95% CrI: − 1.57 ~ − 0.58), calcitonin (SMD = − 0.87, 95% CrI: − 1.44 ~ − 0.30) and alendronate (SMD = − 0.68, 95% CrI: − 1.27 ~ − 0.10). The SMD values are shown in Fig. 6C.

A non-significant global inconsistency was observed (P = 0.479), accompanied by small heterogeneity (I2 = 4%). Only one closed loop(denosumab-ibandronate-placebo) exhibited 95% CrI that included zero, indicating no significant loop inconsistency. As shown in Supplementary Material 6 (Figure S1.3) and Supplementary Material 7 (Figure S2.3), the forest plot for global inconsistency and the forest plots for TH-BMD are provided, respectively.

Trochanter bone mineral density (TRO-BMD)

12 studies have looked into the effects of seven different drugs on TRO-BMD, and 10 were ultimately analyzed. These included 9 studies on bisphosphonates (alendronate: 5, risedronate: 2, pamidronate: 1, etidronate:1), 2 studies on calcitonin, 2 study on SERMs (raloxifene and bazedoxifene), 1 study on PTH and 7 studies involving placebo. Comparing the post-treatment TRO-BMD, placebo were involved in a relatively high number of studies, followed by alendronate. Comparison between placebo and alendronate, between placebo and calcitonin, between alendronate and risedronate were the most frequent. 9 direct comparison pairs were formed between the various drugs. A closed loop formed among placebo, calcitonin, and alendronate, indicating the presence of indirect evidence. A network of comparisons between interventions is shown in Fig. 7A.

Fig. 7

Network meta-analysis of TRO-BMD. (A) Network plot of the network meta-analysis for TRO-BMD. (B) Rankogram of overall intervention strategies from the network meta-analysis for TRO-BMD. (C) League table of the network meta-analysis for TRO-BMD.

The SUCRA values revealed that, among the 9 interventions, alendronate (SUCRA = 94.3%) yielded the greatest TRO-BMD gain, with pamidronate (SUCRA = 84.9%) and risedronate (SUCRA = 78.9%) occupying the subsequent positions. Figure 7B illustrates the rankogram for all intervention strategies.

Compared with alendronate, the TRO-BMD gains were less pronounced with risedronate (SMD = − 0.17, 95%CrI: − 0.26 ~ − 0.08), raloxifene (SMD = − 0.40, 95%CrI: − 0.58 ~ − 0.22), placebo (SMD = − 1.08, 95%CrI: − 1.40 ~ − 0.75), etidronate (SMD = − 0.99, 95%CrI: − 1.83 ~ − 0.16), calcitonin (SMD = − 1.08, 95%CrI: − 1.35 ~ − 0.81), bazedoxifene (SMD = − 1.08, 95%CrI: − 1.79 ~ − 0.36). Relative to risedronate, a diminished increase in TRO-BMD was noted for raloxifene (SMD = 0.23, 95%CrI: 0.03 ~ 0.43), placebo (SMD = 0.91, 95%CrI: 0.57 ~ 1.25), calcitonin (SMD = 0.91, 95%CrI: 0.63 ~ 1.19), bazedoxifene (SMD = 0.91, 95%CrI: 0.19 ~ 1.63) and PTH (SMD = − 0.53, 95%CrI: − 0.64 ~ − 0.42). In head to head comparison with raloxifene, placebo (SMD = 0.68, 95%CrI: 0.31 ~ 1.05), calcitonin (SMD = 0.68, 95%CrI: 0.36 ~ 1.00) showed a smaller rise in TRO-BMD. Using placebo as the baseline, pamidronate (SMD = − 1.03, 95%CrI: − 1.67 ~ − 0.40) were linked to a greater enhancement of TRO-BMD. Taking pamidronate as the comparator, the increment in TRO-BMD was lower with calcitonin (SMD = 1.03, 95%CrI: 0.33 ~ 1.74), bazedoxifene (SMD = 1.03, 95%CrI: 0.13 ~ 1.93), PTH (SMD = − 0.53, 95%CrI: − 0.86 ~ − 0.20). Relative to PTH, a more marked increase in TRO-BMD was evident for alendronate (SMD = − 0.64, 95%CrI: − 0.99 ~ − 0.28). The SMD values are shown in Fig. 7C.

Although overall heterogeneity was low (I2 = 6%), the NMA revealed significant global inconsistency (P < 0.001). This was driven by a significant loop inconsistency in the alendronate-calcitonin-placebo loop, as evidenced by a 95% CrI that did not include zero. Subgroup analysis further localized the source of this inconsistency to the alendronate subgroup, which exhibited substantial within-group heterogeneity (I2 = 85.4%, P = 0.009), reflecting probably internal contradictions within the alendronate evidence base (Figure S3.2 in Supplementary Material 8). The forest plot for global inconsistency can be found in Supplementary Material 6 (Figure S1.4), and the corresponding forest plot for TRO-BMD is available in Supplementary Material 7 (Figure S2.4).

C-terminal telopeptide of type I collagen (CTX)

The number of studies that evaluated the impact of 7 medications on CTX was 20, with 12 included in the analysis. These included 9 studies on bisphosphonates (alendronate: 5, general bisphosphonate: 1, risedronate: 2, ibandronate: 1), 5 studies on denosumab(include 1 study on GP2411), 5 studies on PTH analogs (teriparatide: 2, abaloparatide: 1, PTH:2), 1 study on calcitonin, and 5 studies involving placebo. Comparing post-treatment CTX levels, the highest number of studies involved placebo, followed by alendronate. The comparisons between placebo and denosumab, and between alendronate and risedronate, were relatively frequent. 14 direct comparison pairs were formed between the various drugs. No closed loop formed in this analysis. A network of comparisons between interventions is shown in Fig. 8A.

Fig. 8

Network meta-analysis of CTX. (A) Network plot of the network meta-analysis for CTX. (B) Rankogram of overall intervention strategies from the network meta-analysis for CTX. (C) League table of the network meta-analysis for CTX.

When ranked by SUCRA values, among the 13 interventions, ibandronate (SUCRA = 9.6%) was the leading intervention for CTX reduction, trailed by alendronate (SUCRA = 11.5%), and PTH_alendronate (SUCRA = 27.2%), The rankogram summarizing these strategies is shown in Fig. 8B.

Relative to teriparatide, a greater reduction in CTX was observed with romosozumab (SMD = 1.17, 95%CrI: 0.50 ~ 1.84), risedronate (SMD = 1.46, 95%CrI: 0.40 ~ 2.53), placebo (SMD = 0.68, 95%CrI: 0.01 ~ 1.35), ibandronate (SMD = 2.15, 95%CrI: 1.08 ~ 3.21), denosumab (SMD = 1.36, 95%CrI: 0.52 ~ 2.21), alendronate (SMD = 1.94, 95%CrI: 0.98 ~ 2.90), PTH_alendronate (SMD = 1.63, 95%CrI: 0.42 ~ 2.83), GP2411 (SMD = 1.36, 95%CrI: 0.29 ~ 2.44). In comparison to alendronate, romosozumab (SMD = 0.77, 95%CrI: 0.08 ~ 1.45), risedronate (SMD = 0.47, 95%CrI: 0.01 ~ 0.94), and placebo (SMD = 1.26, 95%CrI: 0.09 ~ 2.43) were associated with a less pronounced decrease in CTX. In comparison to PTH, risedronate (SMD = − 1.07, 95%CrI: − 1.93 ~ − 0.21), ibandronate (SMD = − 1.75, 95%CrI: − 3.36 ~ − 0.15), alendronate (SMD = − 1.54, 95%CrI: − 2.27 ~ − 0.82), and PTH_alendronate (SMD = − 1.23, 95%CrI: − 1.95 ~ − 0.51) were associated with a more pronounced decrease in CTX. Using placebo as the reference, ibandronate (SMD = 1.47, 95%CrI: 0.63 ~ 2.30), denosumab (SMD = 0.68, 95%CrI: 0.16 ~ 1.20) demonstrated a larger reduction in CTX. When compared with ibandronate, calcitonin (SMD = − 1.24, 95%CrI: − 2.32 ~ − 0.16), bisphosphonates (SMD = − 1.33, 95%CrI: − 2.52 ~ − 0.15), abaloparatide (SMD = − 1.59, 95%CrI: − 2.66 ~ − 0.52) yielded a lower magnitude of CTX reduction. Relative to alendronate, the reduction in CTX was less substantial with abaloparatide (SMD = − 1.38, 95%CrI: − 2.55 ~ − 0.21). The SMD values are shown in Fig. 8C.

The network exhibited low heterogeneity (I2 = 0.06%) and no closed loop was formed, so loop inconsistency verification was not performed. Supplementary Material 6 (Figure S1.5) displays the forest plot for global inconsistency, whereas Supplementary Material 7 (Figure S2.5) displays the forest plots for CTX.

N-propeptide of type I collagen (P1NP)

Among the 18 studies that investigated the effects of eight drugs on P1NP, 11 were selected for analysis. These included 8 studies on bisphosphonates (alendronate: 5, risedronate: 2, ibandronate: 1), 4 studies on denosumab (include GP2411), 5 studies on PTH analogs (teriparatide: 2, abaloparatide: 1, PTH: 2), 1 study on calcitonin, 2 studies on romosozumab and 5 studies involving placebo. Comparing post-treatment P1NP levels, the highest number of studies involved placebo. The comparison between placebo and denosumab, between alendronate and risedronate was the most frequent. 13 direct comparison pairs were formed between the various drugs. No closed loop formed among these medications, indicating the absence of indirect evidence. A network of comparisons between interventions is shown in Fig. 9A.

Fig. 9

Network meta-analysis of P1NP. (A) Network plot of the network meta-analysis for P1NP. (B) Rankogram of overall intervention strategies from the network meta-analysis for P1NP. (C) League table of the network meta-analysis for CTX.

In terms of SUCRA probabilities, among the 12 interventions, teriparatide (SUCRA = 99.4%) achieved the highest likelihood of being the best for P1NP, followed by abaloparatide (SUCRA = 84.2%) and placebo (SUCRA = 72.6%). Figure 9B contains the rankogram for the overall interventions.

Using teriparatide as the baseline, the increment in P1NP was lower with romosozumab (SMD = 1.50, 95%CrI: 0.66 ~ 2.33), risedronate (SMD = 2.21, 95%CrI: 0.89 ~ 3.54), placebo (SMD = 1.21, 95%CrI: 0.37 ~ 2.04), ibandronate (SMD = 3.04, 95%CrI: 1.75 ~ 4.34), denosumab (SMD = 2.18, 95%CrI: 1.14 ~ 3.23), calcitonin (SMD = 1.45, 95%CrI: 0.26 ~ 2.64), alendronate (SMD = 2.71, 95%CrI: 1.52 ~ 3.90), PTH_alendronate (SMD = 2.38, 95%CrI: 0.90 ~ 3.87), PTH (SMD = 1.50, 95%CrI: 0.02 ~ 2.98), GP2411 (SMD = 2.09, 95%CrI: 0.76 ~ 3.43). When benchmarked against abaloparatide, ibandronate (SMD = − 2.25, 95%CrI: − 3.54 ~ − 0.96), denosumab (SMD = − 1.39, 95%CrI: − 2.44 ~ − 0.35), alendronate (SMD = − 1.92, 95%CrI: − 3.37 ~ − 0.46) were linked to a smaller improvement in P1NP. In relation to PTH, the P1NP increase was blunted with alendronate (SMD = − 1.21, 95%CrI: − 2.08 ~ − 0.33) and PTH_alendronate (SMD = − 0.88, 95%CrI: − 1.75 ~ 0.01). Compared with romosozumab, treatment with ibandronate (SMD = 1.55, 95%CrI: 0.01 ~ 3.09), alendronate (SMD = 1.21, 95%CrI: 1.36 ~ 2.09) led to a smaller enhancement in P1NP. Relative to placebo, ibandronate (SMD = 1.84, 95%CrI: 0.85 ~ 2.83), denosumab (SMD = 0.98, 95%CrI: 0.35 ~ 1.61), alendronate (SMD = 1.50, 95%CrI: 0.05 ~ 2.96) produced a less marked increase in P1NP. In comparison with ibandronate, the data indicated a greater P1NP increase for calcitonin (SMD = − 1.60, 95%CrI: − 2.89 ~ − 0.30). The SMD values are shown in Fig. 9C.

The NMA showed negligible heterogeneity (I2 = 7%) and loop inconsistency verification was not performed due to the absence of a closed loop. Refer to Supplementary Material 6 (Figure S1.6) for the forest plot of global inconsistency, and to Supplementary Material 7 (Figure S2.6) for the forest plot of P1NP.

Adverse events (AEs)

Serious adverse events (SAEs)

A total of 16 studies investigated serious adverse events following treatment with one of seven drug categories for PMOP. These included 11 studies on bisphosphonates (alendronate: 7, risedronate: 2, general bisphosphonate: 1, ibandronate: 1), 4 studies on denosumab, 3 studies on PTH analogs (teriparatide: 2, abaloparatide: 1), 2 studies on calcitonin, 4 studies on estrogen-related agents (raloxifene: 2, bazedoxifene: 2), 2 studies on romosozumab and 9 studies involving placebo. When comparing the incidence of SAEs across interventions, placebo was the most frequently studied, followed by alendronate. The comparison between placebo and alendronate had the highest number of studies. A total of 16 direct comparison pairs were formed among the drugs. Two closed loop was formed among placebo, bazedoxifene, raloxifene, alendronate, calcitonin, and among placebo, bazedoxifene, raloxifene, alendronate, romosozumab, teriparatide, abaloparatide, indicating the presence of indirect evidence. A network of comparisons between interventions is shown in Fig. 10A.

Fig. 10

Network meta-analysis of SAEs. (A) Network plot of the network meta-analysis for SAEs. (B) Rankogram of overall intervention strategies from the network meta-analysis for SAEs. (C) League table of the network meta-analysis for SAEs.

The SUCRA results indicated that calcitonin (SUCRA = 9.1%) was associated with the lowest rate of SAEs, ranking lowest among the 12 interventions, with ibandronate (SUCRA = 11.4%) and romosozumab (SUCRA = 19.1%) in the next two places. See Fig. 10B for the rankogram.

From the league table, several interventions demonstrated significantly lower risks of serious adverse events compared with bazedoxifene, including calcitonin (RR = 0.51, 95% CrI: 0.27 to 0.96), ibandronate (RR = 0.55, 95% CrI: 0.32 to 0.92), and romosozumab (RR = 0.68, 95% CrI: 0.49 to 0.95). Similarly, a significantly reduced SAE risk was observed for romosozumab (RR = 0.71, 95% CrI: 0.52 to 0.97), ibandronate (RR = 0.57, 95% CrI: 0.34 to 0.94) and calcitonin (RR = 1.90, 95% CrI: 1.01 to 3.56) relative to denosumab. Using ibandronate as the reference group, the highest incidence of SAEs occurred with raloxifene (RR = 1.71, 95% CrI: 1.01 to 2.88), placebo (RR = 1.72, 95% CrI: 1.03 to 2.85). The RR values are shown in Fig. 10C.

No significant global inconsistency (P = 0.579) or heterogeneity (I2 = 4%) was observed. The evidence network formed 4 closed loops: alendronate-placebo-romosozumab-teriparatide, alendronate-placebo-raloxifene, bazedoxifene-placebo-raloxifene, alendronate-calcitonin-placebo. All of these loops showed 95% CrI that included zero, indicating no significant loop inconsistency. The forest plot illustrating global inconsistency is given in Supplementary Material 6 (Figure S1.7), and the forest plot of SAE is given in Supplementary Material 7 (Figure S2.7).

Vertebral fracture

A total of 10 studies investigated vertebral fractures following treatment with one of six drug categories for PMOP. These included 4 studies on bisphosphonates (alendronate: 3, risedronate: 1), 2 studies on denosumab, 4 studies on PTH analogs (teriparatide: 2, abaloparatide: 1, PTH: 1), 2 studies on calcitonin, 1 study on romosozumab and 8 studies involving placebo. When comparing the incidence of vertebral fractures across interventions, placebo was the most frequently studied. The comparisons between alendronate and placebo, and between placebo and denosumab, and between placebo and calcitonin, had relatively high numbers of studies. A total of 9 direct comparison pairs were formed among the drugs. No closed loop was formed, indicating the absence of indirect evidence. A network of comparisons between interventions is shown in Fig. 11A.

Fig. 11

Network meta-analysis of Vertebral Fracture. (A) Network plot of the network meta-analysis for Vertebral Fracture. (B) Rankogram of overall intervention strategies from the network meta-analysis for Vertebral Fracture. (C) League table of the network meta-analysis for Vertebral Fracture.

According to the cumulative ranking probabilities, among the 9 interventions, abaloparatide (SUCRA = 6.1%) showed the greatest likelihood of being the best for preventing vertebral fractures, trailed by teriparatide (SUCRA = 16%) and romosozumab (SUCRA = 35.7%). Figure 11B illustrates this rankogram.

From the league table, teriparatide (RR = 0.20, 95% CrI: 0.08 to 0.48), romosozumab (RR = 0.31, 95% CrI: 0.19 to 0.53), denosumab (RR = 3.00, 95% CrI: 2.45 to 3.67), alendronate (RR = 2.04, 95% CrI: 1.31 to 3.17), abaloparatide (RR = 7.53, 95% CrI: 2.66 to 21.27) and PTH (RR = 3.06, 95% CrI: 2.10 to 4.45) were associated with a significantly lower risk of vertebral fracture compared with placebo. A significantly reduced risk was also observed for teriparatide (RR = 0.23, 95% CrI: 0.10 to 0.57), romosozumab (RR = 0.36, 95% CrI: 0.21 to 0.63), and denosumab (RR = 0.39, 95% CrI: 0.29 to 0.51) when compared with calcitonin. Additionally, teriparatide demonstrated a significantly lower risk relative to risedronate (RR = 0.45, 95% CrI: 0.29 to 0.69). A significantly reduced risk was also observed for alendronate (RR = 1.77, 95% CrI: 1.10 to 2.84), abaloparatide (RR = 6.51, 95% CrI: 2.27 to 18.69), and PTH (RR = 2.65, 95% CrI: 1.74 to 4.01) when compared with calcitonin. Using alendronate as the reference, the point estimates for vertebral fractures favored romosozumab (RR = 0.64, 95% CrI: 0.49 ~ 0.84) and abaloparatide (RR = 3.69, 95% CrI: 1.19 to 11.39). The RR values are shown in Fig. 11C.

The analysis revealed no closed loop with no heterogeneity (I2 = 1%). Supplementary Material 6 (Figure S1.8) and Supplementary Material 7 (Figure S2.8) present, respectively, the forest plot for global inconsistency and the forest plot for vertebral fracture.

Non-vertebral fracture

A total of 9 studies investigated non-vertebral fractures following treatment with one of 6 drug categories for PMOP. These included 3 studies on bisphosphonates (alendronate: 2, risedronate: 1), 2 studies on denosumab, 4 studies on PTH analogs (teriparatide: 2, abaloparatide: 1; PTH: 1), 2 studies on calcitonin, 1 study on romosozumab and 7 studies involving placebo. When comparing the incidence of non-vertebral fractures across interventions, placebo was the most frequently studied. The comparisons between calcitonin and placebo, placebo and denosumab had relatively high numbers of studies. A total of 9 direct comparison pairs were formed among the drugs. No closed loop was formed, indicating the absence of indirect evidence. A network of comparisons between interventions is shown in Fig. 12A.

Fig. 12

Network meta-analysis of Non-vertebral Fracture. (A) Network plot of the network meta-analysis for Non-vertebral Fracture. (B) Rankogram of overall intervention strategies from the network meta-analysis for Non-vertebral Fracture. (C) League table of the network meta-analysis for Non-vertebral Fracture.

According to the SUCRA values, abaloparatide (SUCRA = 12.8%) emerged as the most effective intervention in minimizing non-vertebral fractures among the nine options, followed by PTH (SUCRA = 17.8%) and teriparatide (SUCRA = 38.2%). See Fig. 12B for the rankogram of overall intervention strategies.

Relative to placebo, the data suggested that denosumab (RR = 1.21, 95% CrI: 1.05 ~ 1.40), abaloparatide (RR = 1.84, 95% CrI: 1.04 ~ 3.24) and PTH (RR = 1.61, 95% CrI: 1.14 ~ 2.28), reduced the number of non-vertebral fractures. The RR values are shown in Fig. 12C.

The network exhibited an absence of heterogeneity (I2 = 1%). Because no closed loop was formed, loop inconsistency verification was not conducted. Details of the global inconsistency test are shown in the forest plot of Supplementary Material 6 (Figure S1.9), while the forest plots for nonvertebral fracture is shown in Supplementary Material 7 (Figure S2.9).

Comparison-adjusted funnel plot

We used adjusted funnel plots to assess the risk of bias among the included studies. The funnel plots for all included studies were approximately symmetrical on both sides. In summary, the studies we included showed no significant risk of bias (Fig. 13).

Fig. 13

Adjusted funnel plots.

Sensitivity analysis

The results of the two sensitivity analyses are summarized in Tables S1.1 and S1.2 in Supplementary material 9. Overall, the primary findings for most outcomes remained robust across both sensitivity analyses. In the fixed-effects model sensitivity analysis, the SUCRA rankings and effect directions for LS-BMD, FN-BMD, TRO-BMD, SAEs, vertebral fractures, and non-vertebral fractures were generally consistent with those of the primary analysis. TH-BMD showed a minor fluctuation, with abaloparatide retaining the top rank while PTH moved into the top three (SUCRA: 88.2%). Notably, CTX and P1NP exhibited changes warranting cautious interpretation; under the fixed-effects assumption, the 95% CrIs for PTH_alendronate (SMD = 0.95, 95% CrI: 0.42 to 1.47) and abaloparatide (SMD = − 0.41, 95% CrI: − 0.62 to − 0.21) showed altered statistical significance. In the sensitivity analysis excluding studies with non-blinding of participants and personnel, the results for FN-BMD, TH-BMD, CTX, vertebral fractures, and non-vertebral fractures remained robust. Minor fluctuations were observed for LS-BMD (abaloparatide and calcitonin_minodronate showed marginal differences in ranking), TRO-BMD (alendronate remained first, though the effect estimate changed from − 1.08 to 0.64), and SAEs (calcitonin remained first, while abaloparatide entered the top three). P1NP again demonstrated a pattern requiring cautious interpretation, with a notable change in the 95% CrI for abaloparatide (SMD = − 0.41, 95% CrI: − 0.62 to − 0.21).

Discussion

A total of 41 studies were included in the present NMA, we found that no single drug was the best treatment for all outcomes in PMOP. In order to explore the efficacy of BMD treatment, we performed various subgroup analyses. Calcitonin_minodronate demonstrated the highest probability of being the most effective agents for increasing LS-BMD and FN-BMD. Abaloparatide showed the highest probabilities for increasing TH-BMD while alendronate was ranked best for increasing TRO-BMD. With respect to BTMs, ibandronate was ranked most effective for reducing CTX whereas teriparatide showed the highest probabilities for increasing P1NP. Regarding safety outcomes, calcitonin was associated with the lowest probability of overall SAEs. What’s more, abaloparatide was associated with the fewest vertebral and non-vertebral fractures. Table 2 depicts NMA results for all outcomes.

Table 2 Summary of network meta-analysis results for all outcomes.

Full size table

However, caution should be exercised when interpreting the SUCRA rankings and "best" interventions identified in this study. Although the SUCRA value reflects the probability that a given intervention will be ranked first among all treatments, it does not capture the magnitude of the intervention's effect or the certainty of the supporting evidence. For example, our analysis showed that calcitonin plus minodronate achieved the highest SUCRA (91%) for LS-BMD, with a substantial effect size (SMD =  − 1.86, 95% CrI: − 2.79 to − 0.94), indicating both a high ranking and a clinically meaningful improvement. In contrast, calcitonin ranked lowest for SAEs (SUCRA = 9.1%, RR = 1.84, 95% CrI: 0.99 to 3.45); however, its CrI narrowly crossed the null value of 1, suggesting that its apparent safety advantage should be interpreted with caution and that definitive conclusions cannot yet be drawn. For FN-BMD, calcitonin plus minodronate again ranked highest (SUCRA = 87.7%), with a stable effect estimate (SMD =  − 1.42, 95% CrI: − 2.24 to − 0.61). Subgroup analysis further revealed that the significant global inconsistency for this outcome (P = 0.006) stemmed primarily from conflicting evidence within the alendronate subgroup (I2 = 92.4%), whereas the calcitonin subgroup showed perfectly consistent null effects (I2 = 0%), suggesting that this inconsistency may reflect substantial between-study heterogeneity in alendronate trials, possibly due to variations in dosage, treatment duration, or baseline FN-BMD levels. Abaloparatide ranked highest for TH-BMD (SUCRA = 89.8%, SMD =  − 1.16, 95% CrI: − 1.54 to − 0.78), with low heterogeneity (I2 = 4%) and non-significant inconsistency (P = 0.479), indicating that the evidence for this site-specific effect is relatively robust and clinically informative. In contrast, alendronate ranked highest for TRO-BMD (SUCRA = 94.3%, SMD =  − 1.08, 95% CrI: − 1.40 to − 0.75). Subgroup analysis further localized the significant global inconsistency (P < 0.001) for this outcome to the alendronate subgroup, which exhibited substantial within-group heterogeneity (I2 = 85.4%, P = 0.009), reflecting internal contradictions within the alendronate evidence base. Regarding fracture outcomes, abaloparatide ranked highest for both vertebral (SUCRA = 6.1%, RR = 7.53, 95% CrI: 2.66 to 21.27) and non-vertebral fractures (SUCRA = 12.8%, RR = 1.84, 95% CrI: 1.04 to 3.24); however, the vertebral fracture estimate had a wide CrI, indicating considerable uncertainty, limiting the reliability of the evidence for this outcome. These examples collectively demonstrate that a single summary metric cannot fully capture the complex trade-offs among efficacy across different outcomes, safety, evidence certainty, and network consistency. Therefore, SUCRA rankings should be viewed as a summary metric for generating hypotheses about comparative effectiveness, rather than as definitive evidence of superiority. Clinical decision-making should be based on a comprehensive, multifaceted assessment of the following: first, the specific effect estimates (SMD or RR) and their 95% CrIs from the league table, which provide the actual magnitude of differences between interventions and serve as the core basis for determining clinical significance; second, the robustness of the evidence base, including network consistency (P-value for inconsistency) and heterogeneity (I2) for each outcome, with outcomes showing significant inconsistency or high heterogeneity warranting lower confidence in their rankings; and third, the clinical importance of the outcome itself, with fracture outcomes generally taking precedence over BMD and BTMs because they directly reflect patient benefit. Ultimately, translating these rankings into clinical practice requires integration with individual patient factors, including baseline fracture risk, the predominant sites of bone loss, comorbidities, and patient preferences, thereby achieving a balanced, individualized treatment choice that weighs efficacy, safety, and practicality.

Sensitivity analyses demonstrated that, while most primary results remained robust across different statistical models and study inclusion criteria, the findings for CTX and P1NP warrant cautious interpretation. Under the fixed-effects model, the 95% CrIs for both outcomes showed altered statistical significance, suggesting that their significance was sensitive to the assumption of between-study heterogeneity. In the sensitivity analysis excluding studies with non-blinding of participants and personnel, the results for CTX remained consistent with the primary analysis, whereas P1NP continued to show a pattern requiring caution, indicating that P1NP findings may also be influenced by study quality. Several factors may account for these observations. First, bone turnover markers exhibit greater biological variability than BMD, and their measurement is influenced by pre-analytical factors such as fasting status, circadian rhythms, and assay methods, all of which may contribute to between-study heterogeneity. Second, the relatively limited number of studies reporting CTX and P1NP may have reduced the statistical power to detect consistent treatment effects across different model assumptions. Third, variations in sample collection timing and assay protocols across trials may have introduced additional heterogeneity not fully captured by the model. Therefore, while the direction of effect consistently favored the top‑ranked interventions (ibandronate for CTX, teriparatide for P1NP), the magnitude and statistical significance of these effects should be considered exploratory rather than definitive. Future studies with standardized biomarker assessment protocols and larger sample sizes are needed to confirm these findings. From a clinical standpoint, the sensitivity of BTM results to model assumptions and study quality suggests that clinical decisions should prioritize the more robust BMD and fracture outcomes, which remained stable across all sensitivity analyses. BTM findings may serve as supplementary biological evidence supporting the mechanisms of action of the top‑ranked interventions, but they should not be used in isolation to guide treatment selection.

Our NMA found that different medications and combination therapies have site-specific effects on BMD improvement, with distinct advantages in enhancing BTMs and reducing AEs. Anabolic agents (PTH, abaloparatide, teriparatide) are highly effective in increasing BMD in the axial skeleton (lumbar spine and total hip). Anabolic agents (PTH and its analogs) primarily promote bone formation by stimulating osteoblast activity. The lumbar spine and total hip regions contain abundant cancellous bone, which has a high metabolic turnover rate and responds most rapidly to drugs that promote bone formation. Teriparatide functions primarily by directly stimulating osteoanabolism. Its mechanism involves promoting the differentiation of pre-osteoblasts into mature osteoblasts, enhancing bone-forming activity of existing osteoblasts, and minimizing osteoblast apoptosis57. Therefore, PTH analogues are significantly effective at increasing P1NP levels, which is consistent with our findings. Furthermore, abaloparatide significantly increases BMD at multiple sites and is associated with the lowest fracture risk. This aligns with prior meta-analyses, highlighting that anabolic bone therapy is more effective than bisphosphonates in preventing clinical and vertebral fractures58.

Denosumab and its analogues (GP2411) have been shown to significantly increase FN-BMD and TH-BMD. As a potent antiresorptive agent, denosumab effectively inhibits osteoclasts by blocking receptor activator of nuclear factor κ-B ligand (RANKL)10 . The hip is a weight-bearing, mixed-type bone containing a large amount of cortical bone. Potent anti-resorptive effects help maintain cortical bone thickness and mineralization, reduce cortical bone porosity, and thereby significantly increase hip bone density. However, we cannot help but wonder: as both are anti-resorptive drugs, what is the difference between denosumab and bisphosphonates? The anti-resorptive drugs denosumab and bisphosphonates increase BMD by inhibiting RANKL10 and inducing osteoclast apoptosis8, respectively. However, the effects of denosumab are reversible; once treatment is discontinued, bone turnover increases rapidly, followed by bone loss. Patients who discontinue denosumab should receive subsequent bisphosphonate treatment within the first year59.

On that note, let’s also discuss bisphosphonates. Bisphosphonates (such as alendronate, pamidronate, and risedronate) are most effective at increasing TRO-BMD. The trochanter is a part of the hip, primarily composed of cortical bone; it serves as a muscle attachment site and is subjected to complex forces. We believe that bisphosphonates have a high affinity for bone and can remain deposited in bone tissue over the long term to exert their effects. This long-term, stable anti-resorptive action may be particularly effective in maintaining the bone structure of a stress-bearing region such as the greater trochanter. Studies have found that among bisphosphonate users who have not experienced femoral fractures, the lateral cortex is thicker compared to patients who have not received bisphosphonate treatment. Quantitative CT scans reveal higher BMDSD in the lateral cortex of the subtrochanteric region prior to fracture in long-term bisphosphonate users60, which is consistent with the results of our NMA. Furthermore, bisphosphonates demonstrate significant efficacy in reducing CTX levels, which is consistent with their anti-resorptive effects.

The combination of calcitonin and minodronate is highly effective in increasing LS-BMD and FN-BMD. As an antiresorptive agent, minodronate inhibits osteoclasts, resulting in minimal bone loss. In addition to its mild antiresorptive effect, calcitonin possesses a unique analgesic action. Numerous animal studies have shown that calcitonin acts directly or indirectly on the hypothalamus and/or pituitary via monoaminergic neurotransmitters, thereby increasing plasma β-endorphin levels61. Other researchers have noted62 that calcitonin-induced β-endorphin release is not always observable; the analgesic effect of calcitonin is multifactorial, with β-endorphin playing a distinct role. Additionally, calcitonin has a protective effect on cartilage63. Recent studies have demonstrated that although the calcitonin receptor(CTR) is highly expressed in articular cartilage and subchondral bone and promotes osteophyte formation, endogenous CTR signaling does not affect cartilage degeneration or subchondral bone lesions in experimental post-traumatic osteoarthritis in mice64. Therefore, the combination of these two agents exerts a dual effect on bone metabolism, yielding better results than monotherapy. Additionally, while calcium supplements, as basic nutritional supplements, are expected to have placebo-like safety profiles, their efficacy in treating osteoporosis as a standalone therapy is limited and must be evaluated in context.

Several previous NMAs have compared pharmacological treatments for PMOP, yet our study differs from and extends this body of literature in several important aspects. First, whereas most prior NMAs focused on a single outcome—primarily LS-BMD—or were confined to a specific drug class (e.g., bisphosphonates or anabolic agents alone), our investigation simultaneously evaluated antiresorptive agents (bisphosphonates, denosumab, calcitonin), anabolic agents (teriparatide, abaloparatide, PTH), and their combination regimens across nine clinically relevant outcomes, including site-specific BMD, BTMs, safety, and fracture endpoints. This comprehensive framework enabled a more balanced appraisal of the trade-offs between efficacy and safety. Second, previous NMAs often pooled all bisphosphonates into a single class, potentially obscuring important differences among individual agents. By contrast, we analyzed each bisphosphonate separately where data permitted (e.g., alendronate, risedronate, ibandronate, pamidronate) and identified site-specific effects on BMD: alendronate ranked highest for TRO-BMD, whereas ibandronate was most effective in reducing CTX. Third, our analysis incorporated recently published RCTs and novel agents, including abaloparatide and romosozumab, thereby updating the evidence base and providing a more current comparative effectiveness landscape. Fourth, we explicitly explored the clinical interpretation of SUCRA rankings in the context of effect magnitude and evidence certainty, offering practical guidance that previous NMAs have not systematically provided. From a clinical standpoint, these findings carry several implications. The site-specific BMD effects observed across different drugs suggest that treatment selection should be tailored to the predominant sites of bone loss in individual patients. The robust performance of BMD and fracture outcomes across sensitivity analyses supports their use as primary endpoints in clinical decision-making, whereas BTMs results warrant more cautious interpretation. The favorable safety profiles of calcitonin and ibandronate indicate that these agents may be particularly suitable for patients at high risk of AEs or those with contraindications to other therapies. Furthermore, the superior efficacy of anabolic agents (abaloparatide and teriparatide) in reducing fracture risk supports their earlier use in patients with severe osteoporosis or prior fragility fractures, aligning with the current paradigm shift toward more aggressive initial therapy in high-risk populations.

This study has several limitations. First, we did not perform a formal analysis of heterogeneity in baseline characteristics, including age, sample size, or study duration. Although the primary analysis indicated low statistical heterogeneity for most outcomes (I2 < 50%), considerable clinical heterogeneity existed across the included trials. Patient populations varied substantially in mean age (ranging from 50 to 90 years), baseline BMD levels (from osteopenia to severe osteoporosis), history of prior fractures (most studies included patients with prevalent fractures, accounting for approximately 95% of all participants), and concomitant calcium/vitamin D supplementation (which differed in dosage and adherence). These variations may affect the generalizability of our findings to specific patient subgroups. Follow-up duration also varied widely, from 1 to 60 months; shorter trials may have underestimated fracture outcomes, whereas longer trials may have captured more AEs. Importantly, treatment effects on BMD may not fully reflect fracture risk reduction, as the relationship between BMD change and fracture reduction is nonlinear and varies across drug classes. Intervention protocols differed substantially, including drug dosages (e.g., alendronate 5 mg vs. 10 mg vs. 70 mg weekly), administration routes (oral, subcutaneous, intravenous, or intranasal), and dosing schedules (daily, weekly, monthly, or quarterly), all of which may influence both efficacy and safety. Moreover, outcome assessment methods varied across studies: BMD was measured using different brands of dual‑energy X‑ray absorptiometry (DXA) scanners with varying calibration protocols, and fracture endpoints were often based on different definitions (morphometric vs. clinical fractures) and verification methods. BTMs, in particular, were assessed using diverse assays with differing sensitivity and specificity, which may explain the sensitivity of CTX and P1NP results to model assumptions. Second, certain comparisons included a limited number of participants, which may have reduced the precision of the corresponding estimates. Third, nine RCTs were judged to be at high risk of bias due to their open‑label design. The lack of blinding may have influenced subjective outcomes, such as patient‑reported AEs, and introduced expectation bias in fracture assessment. The decision to retain these studies involved a trade‑off between network connectivity and potential bias—excluding them would have fragmented the evidence network and precluded quantitative synthesis. Therefore, while we prioritized network connectivity to provide clinically meaningful comparisons, this choice necessitates acknowledging potential bias and interpreting the results with due caution. Fourth, due to insufficient data, we were unable to analyze certain outcomes, including serum calcium levels, areal BMD (aBMD), and volumetric BMD (vBMD). Fifth, global and loop inconsistencies were identified within the evidence network. Statistical assessments revealed significant inconsistencies within certain closed loops, suggesting a lack of full coherence between direct and indirect evidence for some comparisons. Subgroup analyses further localized the source of inconsistency for FN‑BMD and TRO‑BMD to the alendronate subgroup, which exhibited substantial within‑group heterogeneity (I2 = 92.4% and 85.4%, respectively), reflecting internal contradictions within the alendronate evidence base. These findings indicate that the relative treatment effects and rankings for these specific comparisons should be interpreted cautiously. Future studies should focus on standardizing protocols, reporting patient‑level data, and conducting prospective head‑to‑head comparisons to reduce heterogeneity and strengthen the evidence base for clinical decision‑making.

Conclusion

In our network analysis of PMOP treatments, each medication has strengths but none is best for everything. Calcitonin-minodronate had the highest probability of being the most effective treatment for improving BMD at the LS and FN. Abaloparatide was the top performer for TH-BMD, and alendronate was most effective for TRO-BMD. For BTMs, ibandronate ranked first for reducing CTX, and teriparatide was most likely to increase P1NP. In safety analyses, calcitonin showed the lowest risk for overall SAEs. Finally, abaloparatide was linked to the fewest vertebral and non-vertebral fractures. However, these “best” rankings must be interpreted cautiously, considering the sometimes-limited effect size differences shown in league tables, and the overall risk of bias among included studies. Clinical decision-making should therefore integrate not only probability rankings, but also the magnitude of treatment effects, quality of evidence, drug safety, and individual patient factors. Further high-quality RCTs and direct head-to-head comparisons are needed to strengthen these findings.

来源:Nature 健康护理主题 · nature.com