Outcome measures

Our primary outcome measure is the incidence of vertebral and nonvertebral fractures. Our secondary outcome measures will include: 1) changes in femoral neck, lumbar spine, and total hip BMD from baseline, 2) incidence of adverse events, and 3) incidence of serious adverse events (SAEs). Fracture was chosen as a patient-important efficacy outcome for bone loss and osteoporosis, while BMD is chosen as a widely-recognized and commonly reported surrogate measure of fracture risk, osteoporosis severity, and bone health [13–15]. We will evaluate all outcomes based on data collected at the latest follow-up visit. Fractures and adverse events will be defined as per individual study criteria.

Eligibility criteria

We will include parallel-group or crossover RCTs that: 1) enrolled adult (age ≥18 years old) SOT recipients, 2) compared an anti-osteoporotic medication against placebo/no anti-osteoporotic treatment or against a different anti-osteoporotic treatment, and 3) reported any of our target outcomes. Studies assessing combinations of anti-osteoporotic medications are eligible. If a RCT used a crossover design, latest data from before the first crossover will be used. SOT is defined as transplantation of the kidney, liver, pancreas, heart, or lung [16].

We will include any anti-osteoporotic pharmacotherapies used for the purposes of preventing bone loss and fractures. This includes (but is not limited to): bisphosphonates (e.g., alendronate, risedronate, ibandronate, and zoledronic acid), calcitonin, calcium supplementation, vitamin D and D analogs, raloxifene, parathyroid hormones (i.e., teriparatide and abaloparatide), denosumab, and romosozumab.

Study identification

We will conduct a database search in Ovid Medline, Ovid Embase, Web of Science (Core Collection), EBSCOhost CINAHL, Cochrane CENTRAL, and the China National Knowledge Infrastructure (CNKI) from inception to February 2nd, 2024. The search strategy was developed and piloted on Ovid MEDLINE using relevant MeSH terms and title/abstract headings such as “transplant recipient,” “osteoporosis,” “bone resorption,” and “fractures” and a modified version of the Cochrane Highly Sensitive Search Strategy for identifying randomized trials in MEDLINE [11]. The search strategy was then translated to other databases using database-specific keywords. A sample search strategy is tabulated in S2 Table.

In addition to the database searches, we will also hand-search the reference list of previous systematic reviews, as well as clinical trial registries (i.e., ClinicalTrials.gov and the WHO International Clinical Trials Registry Platform) and preprint servers (i.e., medRxiv and Research Square) for relevant published data.

To account for potential changes in clinical practice and ensure that our findings are current, we will proactively monitor literature sources for new publications throughout the review process, and incorporate newly published data or conduct updated database searches, if needed.

Study selection

We will perform title and abstract screening independently and in-duplicate using Covidence (https://www.covidence.org/) [17]. Records deemed eligible will be subsequently retrieved and entered into an in-duplicate full-text screening process. Disagreements will be resolved by discussions mediated by a senior author to reach consensus.

Data extraction

We will perform data extraction independently and in-duplicate using standardized data extraction sheets developed a priori. Data from intention-to-treat (ITT) or modified ITT analyses will be preferentially extracted over data from per-protocol analyses. We will resolve discrepancies by recruiting a senior author to review the data.

Data items

Data items that will be extracted include:

1. Bibliometric data: author names, year of publication, trial registration number, digital object identifier, publication journal, funding sources, and conflict of interest.
2. Methodology descriptions: number of participating centers, study setting, blinding methods, enrollment period, randomization and allocation methods, BMD measurement methods, and fracture definition.
3. Treatment descriptions: description of anti-osteoporotic therapies (i.e., name, type, dosage, and duration), description of adjuvant therapies, and description of placebos (if applicable).
4. Baseline information: number of patients randomized, number of patients analyzed for efficacy and safety outcomes, number of patients lost to follow-up, mean age and age variance, and sex distributions.
5. Outcome data: Mean/median baseline BMD measurements with variance, mean/median final BMD measurements with variance, mean/median absolute change in BMD from baseline with variance, mean/median percentage change in BMD from baseline with variance, incidence of vertebral fracture, incidence of non-vertebral fracture, incidence of adverse events, incidence of SAEs, follow-up duration of the latest efficacy and safety outcome measurements and treatment adherence at the latest follow-up.

All extracted data will be available to readers of the final review publication upon request.

Risk of bias

We will assess risk of bias (RoB) independently and in-duplicate using the Revised Cochrane risk of Bias Tool for Assessing Randomised Trials (RoB2) [18]. The assessments will be aimed at examining the effect of assignment to interventions, given our focus on extracting intention-to-treat outcome data. Disagreements will be resolved by discussions mediated by a senior author to reach consensus.

Network meta-analysis

We will perform Bayesian NMAs using the gemtc and BUGSnet library [8] in R and Markov chain Monte Carlo engine JAGS [19]. Because we expect heterogeneity among studies due to differences in patient characteristics and study methodologies, we will use a random-effects model [8]. The NMAs will be stratified by SOT type. We will assume a vague prior for the between-study heterogeneity with uniform distribution. Patients receiving no active anti-osteoporotic interventions, including patients using placebo, will be used as a reference for comparison. Transitivity will be assessed qualitatively by comparing patients and study characteristics.

For changes in BMD, we will report the results of the analysis as mean differences with 95% credible intervals (CrIs) if all included studies utilized the same scale (e.g., if BMD changes are all reported as percentage changes). Otherwise, we will report these outcomes as standardized mean differences (SMDs) to include all available RCT data. We will calculate SMD by dividing the mean differences between treatment groups by the weighted pooled standard deviation (SD) using the DerSimonian-Laird method [20]. Fracture incidences will be reported as risk ratios with corresponding 95% CrIs. Model convergence will be assessed using the Brooks-Gelman-Rubin method as well as examination of trace and autocorrelation plots [21]. We will run all network models for a minimum of 100,000 iterations to ensure convergence. Network graphs will be provided to visualize the overall network structure of comparisons.

If there are outcomes for which we did not gather enough information to perform a NMA, we will provide a qualitative description of the available data and study outcomes or perform a pairwise meta-analysis if data permits.

Heterogeneity and inconsistency assessment

We will assess between-study heterogeneity within each pairwise comparison using I² statistics [22]. Inconsistency will be assessed using the node-splitting method [23]. If inconsistency is identified, we will conduct network meta-regression models and subgroup analysis to identify how potential effect modifiers may impact our results. Based on previous literature reviews, our assessments for sources of inconsistency will be focused on age, sex, and BMI [24]. In addition, we will conduct one-stage sensitivity analysis with E-value analogs (if applicable) to evaluate the strength of unmeasured confounding [25] if we identified RCTs with a high risk of bias.

Treatment ranking

We will use the surface under the cumulative ranking curve (SUCRA) scores to provide an estimate as to the ranking of treatments. SUCRA scores range from 0 to 1, with higher SUCRA scores indicating more efficacious treatment arms [26].

Missing data

We will attempt to contact the authors of the original studies to obtain missing or unpublished data. For studies that did not report change in BMD but reported BMD values at baseline and follow up visits, we will impute the mean change and associated variances using correlation coefficients, as recommended by the Cochrane Handbook. For studies that reported median and range or median and interquartile range as summary statistics for continuous measurements, we will first assess the normality of the data using methods proposed by Shi et al. [27]. If the data is not significantly skewed, we will impute mean and standard deviation using methods proposed by Shi et al. [28], Luo et al. [29], and Wan et al. [30].

Publication bias

We will use comparison-adjusted funnel plots to assess for small-study effects within networks as indications for publication bias [31]. As the treatment arms need to be sorted in a meaningful way to generate comparison-adjusted funnel plots, we will order the treatment arms to define each comparison as an intervention versus a control treatment, with the assumption that publication bias is likely to exaggerate the effectiveness of the intervention treatment [32]. Egger’s regression test will be used to examine the presence of asymmetry in the comparison-adjusted funnel plots [33].

Quality of evidence

We will use the Confidence in Network Meta-Analysis (CINeMA) web application (https://cinema.ispm.unibe.ch/) [34] to evaluate confidence in the findings from our NMA. The CINeMA framework is similar to the Grading of Recommendations, Assessment, Development, and Evaluations (GRADE) framework [35,36] and evaluates network quality based on six criteria: within-study bias, across-study bias, indirectness, imprecision, heterogeneity and incoherence [37]. We will summarize the results and quality of our NMA using a summary of findings table.

Ethics and dissemination

No ethics approval is needed as only published data will be used and analyzed. The completed review will be submitted for publication in a peer-reviewed journal.

Limitations

There are several potential limitations to consider in this proposed study. Firstly, we expect to include a clinically heterogeneous patient population, even after stratifying by SOT types. We aim to assess and mitigate the impact of clinical confounders through the use of node-splitting, meta regressions, and sensitivity analyses. Secondly, there may be certain differences in clinical practices adopted by RCTs originating from different countries that may impact our findings and contribute to higher heterogeneity. However, we expect to account for this by seeking expertise from our multinational and multilingual research team. This effect may also be mitigated by the existence of standardized international clinical practice guidelines for post-SOT care, which reduces the diversity in clinical practices among different geographical regions [39]. Finally, our analysis may be limited by sparse literatures surrounding the use of prior distributions in this field. We will be using vague priors to allow the data to inform the posterior distribution more heavily.