Materials and methods

Protocol and registration

This meta-analysis was conducted in strict adherence to the updated Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 guidelines [8]. Flow diagrams mapping the study selection process were programmatically generated utilizing the PRISMA2020 R package to ensure reproducible reporting [9]. The study protocol was prospectively registered with the International Prospective Register of Systematic Reviews (PROSPERO; registration number CRD420251081490) [10]. The protocol defined the a priori research questions, eligibility criteria, pharmacological data extraction frameworks, and statistical methodologies.

Literature search and inclusion criteria

A comprehensive systematic search was executed to identify preclinical in vivo studies evaluating mesenchymal stem cell-derived extracellular vesicles (MSC-EVs) in animal models of OA. PubMed, Embase, and Web of Science were searched from January 2017 to March 2025. The search strategy combined controlled vocabulary (e.g. MeSH terms) and free-text keywords covering “extracellular vesicles”, “exosomes”, “mesenchymal stem cells” and “osteoarthritis”. Eligible studies included controlled in vivo OA models utilizing MSC-EV interventions that reported at least one quantitative structural or histological endpoint (e.g. Osteoarthritis Research Society International (OARSI) scores). In vitro only studies, non-MSC EV sources and articles lacking extractable quantitative data were excluded.

Data extraction

Data extraction was independently performed using a standardized template, with discrepancies resolved by consensus. To address the recognized heterogeneity in EV isolation and ensure rigorous pharmacological reproducibility, a comprehensive study-by-study methodological matrix was constructed (Supplementary material, Table S1). For each included study, this matrix systematically captures: (1) reagent names and critical solution compositions, (2) instrumentation details (e.g. transmission electron microscope (TEM) accelerating voltage, nanoparticle tracking analysis (NTA) models) and (3) specific procedural conditions including centrifugation speeds/relative centrifugal force (RCF), incubation times, temperatures, and injection routes. Where specific parameters were omitted by the original authors, they were explicitly recorded as "not reported" (NR). To align with translational pharmacokinetic (PK) evaluation standards, EV dosing parameters were rigorously extracted. Dose per injection (particles) was used as the primary metric, while protein concentration (μg mL-1) was treated as a secondary proxy when particle counts were unavailable. Variables, including injection volume, number of injections, dosing intervals, and follow-up duration, were recorded to evaluate cumulative exposure and intra-articular therapeutic retention.

Risk of bias and quality assessment

The methodological quality of the included preclinical studies was evaluated using the Systematic Review Center for Laboratory Animal Experimentation (SYRCLE) risk of bias tool [11]. SYRCLE risk of bias domains span selection, performance, detection, attrition and reporting biases (Supplementary Material). Studies were categorized as having a low, high or unclear risk of bias for each domain. Adherence to the Minimal Information for Studies of Extracellular Vesicles (MISEV) 2023 guidelines was also systematically audited for each study [7].

Statistical analysis, meta-regression and publication bias quantitative synthesis was conducted using a random-effects model (REML estimator) to account for the anticipated inter-study heterogeneity arising from varying EV sources, animal models and dosing regimens. OARSI histological score served as the primary endpoint. Because the OARSI score represents a uniform continuous scale across the literature, pooled effect sizes were calculated as mean differences (MDs) with 95 % confidence intervals (CIs). Statistical heterogeneity was evaluated using Cochran’s Q test, the between-study variance (_τ_2), and the _I_2 statistic, with _I_2 > 50 % indicating substantial heterogeneity.

Subgroup analyses were conducted a priori based on (i) experimental animal species, (ii) EV cellular source, and (iii) OA induction model. Mixed-effects meta-regression was utilized to investigate continuous covariates, specifically evaluating whether the administered dosage per injection or the total number of injections moderated the therapeutic effect size.

Potential publication bias was assessed visually using funnel plots and formally quantified using Egger’s regression asymmetry test. To ensure the robustness of our findings against potential small-study effects, a non-parametric Trim-and-Fill analysis was performed to estimate adjusted effect sizes accounting for theoretically missing studies. Furthermore, the fragility of the overall effect was evaluated using Fail-Safe N calculations via the Rosenthal, Orwin, and Rosenberg approaches. A two-sided p-value < 0.05 was considered statistically significant for all analyses.

Compliance with ethical guidelines

This study is a meta-analysis of previously published preclinical studies and did not involve any new animal or human experimentation. Therefore, no additional ethical approval was required.