To guarantee the rigor and traceability of the study design, this study was performed according to the PRISMA guidelines of systematic reviews and meta-analyses (JVDS et al., 2023), with adaptations appropriate for preclinical animal studies. It was registered in INPLASY before the start of the study (registration number: INPLASY2025110085).

To ensure the comprehensiveness and reliability of the meta-analysis results, this study systematically searched both English and Chinese databases to gather all relevant preclinical evidence. The English sources included PubMed, Web of Science, and Embase, while the Chinese sources covered CNKI, Wanfang, and VIP databases. The search was conducted for all publications available up to November 2025. The search framework integrated both thesaurus-based terms and unrestricted text words. Diabetic kidney disease–related terminology—namely “Nephropathies, Diabetic”, “Nephropathy, Diabetic”, “Diabetic Kidney Disease”, “Diabetic Kidney Diseases”, “Kidney Disease, Diabetic”, “Kidney Diseases, Diabetic”, “Diabetic Nephropathy”, “Diabetic Glomerulosclerosis”, “Glomerulosclerosis, Diabetic”, “Intracapillary Glomerulosclerosis”, “Kimmelstiel-Wilson Disease”, “Kimmelstiel Wilson Disease”, “Nodular Glomerulosclerosis”, “Glomerulosclerosis, Nodular”, “Kimmelstiel-Wilson Syndrome”, “Kimmelstiel Wilson Syndrome”, and “Syndrome, Kimmelstiel-Wilson”—was matched with compound-specific identifiers, including “Rutin”, “Rutoside”, “Quercetin-3-Rutinoside”, “Quercetin 3 Rutinoside”, “3-Rhamnosyl-Glucosyl Quercetin”, and “Quercetin, 3-Rhamnosyl-Glucosyl”. Full search details are available in Figure 2 and the Supplementary Material.

This study followed the PICO framework to establish inclusion criteria. Studies meeting the following conditions were considered for inclusion: (Zhang et al., 2025): animal models of DN; (Sun et al., 2022); interventions using rutin monomer, with at least one dosage group; (Rossing et al., 2022a); a DN model control group, receiving either a placebo or no treatment; (Rossing et al., 2022b); reporting key renal outcome measures, including serum creatinine (Scr), blood urea nitrogen (BUN), and 24-h urinary protein (24-h UTP). If studies also provided data on blood glucose, lipid levels, oxidative stress, or fibrosis biomarkers, these were extracted as secondary outcomes. Exclusion criteria included: (Zhang et al., 2025): non-animal studies, such as clinical trials, reviews, case reports, or in vitro studies; (Sun et al., 2022); studies that met inclusion criteria but were not accessible in full text; (Rossing et al., 2022a); studies lacking an independent control group, or those with combined treatments making it impossible to isolate the effects of rutin; (Rossing et al., 2022b); duplicate publications; (Tabolacci et al., 2023); studies without the necessary data for quantitative analysis, such as sample size, standard deviation, or standard error.

Literature search, study selection, data extraction and quality assessment were conducted by two reviewers (Zongtao Li and Yashi Wang) based on the predetermined inclusion and exclusion criteria. Any discrepancies were resolved through discussion or by consulting the corresponding author.

The step of data extraction involved the reviewers separately gathering the following key pieces of information: (Zhang et al., 2025): publication date and first author, (Sun et al., 2022), animal characteristics, such as species, sex, age, and weight, (Rossing et al., 2022a), methods of establishing the diabetes model, including induction drugs, route of administration, doses, modeling duration, and diagnostic criteria, (Rossing et al., 2022b), intervention factors in both experimental and DN model groups, such as administration method, dose, and duration, and (Tabolacci et al., 2023) primary and secondary outcome measures. For studies with co-interventions or multi-component designs, only comparisons between the rutin monotherapy arm and the DN model control arm (without additional active treatment) were extracted to isolate the effect of rutin; combined-treatment arms were not synthesized. When multiple control arms were available, the control arm with an identical background regimen (except rutin) was preferentially selected. For multi-arm animal studies with multiple rutin doses but a single shared control group, we followed the guidance on multi-arm trials from the Cochrane Handbook (sixth edition) (Chndler, 2024). The sample size of the shared control group was split as evenly as possible among the comparisons, ensuring the total sample size matched the original control group size to avoid double-counting the same control animals, which would artificially inflate the study’s weight. During this process, the mean and standard deviation of the control group remained unchanged, and only the sample size was adjusted to ensure reasonable weight distribution. In studies that reported data at multiple time points, the most recent measurement obtained before euthanasia was used.

In cases where study results were provided only in the form of graphs, we first attempted to contact the authors to obtain the original data. If no response was received, we used WebPlotDigitizer 4.8 software to extract data from the graphs. To ensure the accuracy of the data, the reviewers digitized the data separately and cross-checked the results. Any discrepancies were resolved through consultation or discussion with the respective author. Results presented as standard error of the mean (SEM) were converted to standard deviation (SD) using the formula SD = SEM × √n.

The risk of bias in the included studies was independently assessed using the SYRCLE risk of bias assessment tool for animal intervention studies (Carlijn et al., 2014). This tool is specifically designed for methodological evaluation of animal studies and consists of 10 items aimed at identifying potential sources of systematic bias during the research process. The assessment covers the following areas: (Zhang et al., 2025): random sequence generation; (Sun et al., 2022); comparability of baseline characteristics; (Rossing et al., 2022a); allocation concealment; (Rossing et al., 2022b); random housing; (Tabolacci et al., 2023); blinding during the study (performance bias); (Jang et al., 2025); random outcome assessment; (David et al., 2023); blinding of outcome measurements (detection bias); (Serri et al., 2024); completeness of outcome data; (Choi et al., 2021); selective reporting; and (Karunarathne et al., 2023) other potential sources of bias. Disagreements during the assessment were resolved through discussion or consultation with the corresponding author. Additionally, a stratified analysis was performed by categorizing studies based on their risk-of-bias scores from the SYRCLE tool. Pooled effect sizes were calculated separately for each risk-of-bias group to explore whether studies with higher risk-of-bias scores reported larger effects compared to those with lower scores.

To investigate the potential impact of different study characteristics on effect sizes, this study pre-defined four subgroups: (Zhang et al., 2025): animal species; (Sun et al., 2022); modeling methods for DN; (Rossing et al., 2022a); duration of rutin intervention; and (Rossing et al., 2022b) rutin dosage. Subgroup analyses were primarily exploratory and aimed at identifying possible sources of heterogeneity when significant variation was observed. All subgroup analyses were performed using the Meta-Analysis module in Stata 17 via the graphical interface.

All statistical analyses were conducted using Stata 17.0. The data included in this study were continuous outcomes, and the overall effect size was represented by the standardized mean difference (SMD), with 95% confidence intervals (CIs) used to indicate the uncertainty of the estimates. Heterogeneity was evaluated using the I² statistic. An I² value greater than 50% indicates significant heterogeneity, and in this case, a random-effects model (DerSimonian–Laird method) was used. When I² was low, a fixed-effects model was applied. All analyses were performed through the Meta-Analysis menu in Stata 17, without the need for command inputs. Statistical significance was considered at P < 0.05. Given the generally unclear or high risk of selection bias, including the lack of reported allocation concealment in the included studies, pooled effect sizes were interpreted with caution. The purpose of quantitative synthesis was to summarize the overall direction and consistency of effects across studies rather than to provide precise or unbiased estimates of treatment efficacy.

To assess the risk of publication bias for the primary outcomes, a funnel plot was created, and Egger’s regression test was performed. These analyses were conducted using the Publication Bias function in the Meta-Analysis module of Stata 17. When Egger’s test suggested potential bias, the trim-and-fill method was further applied to estimate the missing studies and obtain an adjusted pooled effect.

To assess the robustness of the meta-analysis results, sensitivity analysis was performed using the “leave-one-out analysis” function in the Meta-Analysis module of Stata 17. This method involves systematically excluding each study and recalculating the pooled effect to determine if any individual study significantly impacts the overall effect size. If the pooled effect remains largely unchanged after each exclusion, the results are considered to be robust.

To explore the effects of varying dosing schedules and treatment durations on key changes in DN, this paper created three-dimensional dose-time response plots for Scr, BUN, and 24-h UTP based on available literature. This visualization analysis was used to illustrate the relationship between rutin dosage, treatment duration, and changes in efficacy, providing a clearer method to determine the potential differences in the effects of different dosing strategies on improving kidney function.

A total of 314 articles were retrieved from six databases in this study, including PubMed (33 articles), Embase (124 articles), Web of Science (121 articles), CNKI (19 articles), Wanfang (11 articles), and VIP (6 articles). After merging the search results and removing 78 duplicate records, 236 articles were included in the initial screening. Based on title and abstract screening, 134 articles were excluded for not meeting the inclusion criteria. The main reasons for exclusion were: review articles (94), network pharmacology studies (Choi et al., 2021), clinical studies (Choi et al., 2021), conference abstracts (Zhang et al., 2025), in vitro studies (Karunarathne et al., 2023), and mass spectrometry/compound analysis studies (Page et al., 2021). Subsequently, 102 articles underwent full-text assessment, and 89 articles were further excluded. The reasons for exclusion included: non-DN models (Tanase et al., 2022), studies not using rutin as an intervention (Wu et al., 2021), full text unavailable (Sun et al., 2022), and studies not reporting the predefined primary outcomes (Tabolacci et al., 2023). Ultimately, 13 studies met the inclusion criteria and were included in the meta-analysis (see Figure 2), comprising 9 English-language and 4 Chinese-language publications. Because included studies could be indexed in more than one database, the database coverage of each included study is summarized in Supplementary Table S1.

This study included 13 experimental studies published between 2006 and 2025, covering 318 DN animals. Among them, two studies used db/db mice (36/318, 11.32%), seven studies used Sprague-Dawley rats (210/318, 66.04%), three studies used Wistar rats (36/318, 11.32%), and one study used C57BL/6 mice (36/318, 11.32%). Regarding animal sex, 11 studies used male animals, while two studies used a mix of both sexes.

Three studies reported animal age, while 10 did not provide this information. Ten studies recorded the initial body weight of the animals, while three studies did not. Regarding the modeling method, eight studies used streptozotocin (STZ) to induce diabetes, two used spontaneous diabetic models, and three used alloxan for induction.

The blood glucose thresholds for successful modeling varied across studies, with four studies using a fasting glucose >16.6 mmol/L, one using >13.88 mmol/L, three studies using >200 mg/dL, and two studies using >250 mg/dL. One study did not specify the threshold. Intervention durations ranged from 4 to 12 weeks, with rutin doses ranging from 10 to 200 mg/kg. All 13 studies administered rutin via oral gavage.

Regarding outcome measures, all studies measured Scr, 11 studies measured BUN, seven studies reported 24-h UTP, and 11 studies measured blood glucose (Glu). Additionally, two studies evaluated triglycerides (TG) and total cholesterol (TC). Some studies assessed oxidative stress markers, including reactive oxygen species (ROS), superoxide dismutase (SOD), malondialdehyde (MDA), glutathione (GSH), and catalase (CAT). Moreover, some studies reported fibrosis-related signaling molecules, such as transforming growth factor-β (TGF-β). A detailed summary is provided in Table 1.

We evaluated the risk of bias of all included studies, as shown in Figure 3 and Supplementary Figure S1. The risk-of-bias scores of the included studies ranged from 4 to 6, indicating that several key items for bias control were incompletely reported. Although all 13 studies reported random allocation, none of them described the specific method used to generate the random sequence. Two studies did not provide information on baseline comparability. No study reported allocation concealment.

For performance bias, 12 studies were rated as high risk, with only one study assessed as low risk, indicating that blinding of personnel involved in the intervention was largely absent. In contrast, blinding of outcome assessment was generally adequate, suggesting that outcome evaluation was mostly conducted independently of treatment information. All studies reported outcomes consistent with their stated protocols, and no additional sources of bias were identified. In the supplementary section, we present the results of the stratified analysis, where studies were categorized into high-risk, medium-risk, and low-risk of bias based on the SYRCLE tool. The pooled effect sizes for each group are presented in the following figures (Supplementary Figures S2–S4; Supplementary Table S2). Overall, our analysis did not reveal significant differences in effect sizes between high-risk and low-risk groups (P > 0.05). However, a significant difference was observed between medium-risk and low-risk groups (P < 0.05), suggesting that the risk of bias may have a minor influence on effect sizes in these groups.

Leave-one-out sensitivity analyses for Scr, BUN, and 24-h UTP showed that removing any single study did not materially change the pooled effect sizes. For Scr, the pooled Hedges’s g remained approximately between −2.19 and −2.00; for BUN, between −2.27 and −2.02; and for 24-h UTP, between −5.78 and −5.12. All estimates remained statistically significant with P < 0.001. These results indicate that no individual study unduly influenced the overall findings, confirming the robustness and stability of the meta-analytic estimates (Supplementary Figures S5–S7).

Subgroup analyses were conducted for 24-h UTP, BUN, and Scr to explore potential sources of heterogeneity. Overall, rutin consistently improved renal outcomes across all subgroup categories. For 24-h UTP, stratification by intervention duration, dosage, animal species, and modeling method revealed comparable effect sizes, and no subgroup differences reached statistical significance (all P > 0.05), indicating that the reduction in proteinuria was stable across study characteristics. For BUN, several subgroup comparisons—particularly those based on duration and dosage—showed significant between-group differences (P < 0.05), suggesting that the extent of BUN reduction may be partly influenced by exposure time or dosing level. For Scr, a few subgroup tests yielded borderline or modestly significant differences, yet all subgroups maintained the same direction of effect, and none contradicted the overall finding. Although heterogeneity within many subgroups remained high (often I² > 50%), the consistently beneficial direction across all subgroup categories supports the robustness of rutin’s renoprotective effects (Supplementary Figures S8–S19).

Potential publication bias was evaluated using funnel plots (Figure 9), Egger’s regression tests (Figure 10), and the Duval and Tweedie trim-and-fill method (Figure 11) for the three primary outcomes (Scr, BUN, and 24-h UTP). The funnel plots exhibited varying degrees of asymmetry, suggesting the presence of small-study effects. Egger’s tests further supported this observation, yielding statistically significant intercepts across all outcomes, indicating directional rather than random scatter of study estimates. To assess the potential impact of missing studies, the trim-and-fill procedure was applied as an exploratory sensitivity analysis. Although a small number of studies were imputed, the adjusted pooled effect sizes remained largely unchanged relative to the original estimates. However, trim-and-fill relies on assumptions that may not fully account for consistent small-study effects; therefore, these results should be interpreted cautiously and do not rule out publication or other reporting-related biases. Overall, under the trim-and-fill assumption, the direction and magnitude of the pooled effects were broadly similar, but our conclusions remain conservative.

To describe the dose–time–response trends of rutin, 3D plots were generated for Scr (Figure 12a), BUN (Figure 12b), and 24-h UTP (Figure 12c). Across studies, effective data points were mainly concentrated within 25–200 mg/kg and 4–8 weeks. For Scr, most observable improvements clustered around 30–100 mg/kg and 6–8 weeks, indicating that moderate doses with mid-term treatment were more frequently associated with better outcomes. For BUN, responsive points showed a similar pattern, predominantly distributed between 25 and 100 mg/kg, with effective durations mainly around 4–8 weeks. For 24-h UTP, the effective region appeared slightly broader in dose, extending approximately from 20 to 150 mg/kg, but treatment durations remained mostly within 4–8 weeks. Overall, rutin demonstrated a relatively consistent therapeutic window, with most beneficial effects occurring at moderate doses (25–100 mg/kg) and short-to-mid treatment durations (four to eight weeks) rather than across the entire dose–time spectrum. We also estimated the human equivalent dose for the most frequently “effective” preclinical range (25–100 mg/kg) using standard body-surface-area scaling. This corresponds approximately to 4.1–16.2 mg/kg (rat-based) or 2.0–8.1 mg/kg (mouse-based), provided for context only rather than as a clinical dosing recommendation.

The preclinical evidence on the use of rutin in animal models of DN has been summarized in this systematic review and meta-analysis. Thirteen studies involving 318 animals were included. Overall, rutin consistently improved renal function across different species and modeling strategies. It reduced Scr, BUN, and 24-h UTP. Beyond these renal indicators, rutin also relieved hyperglycemia, lipid abnormalities, oxidative stress, inflammation, and tissue fibrosis—suggesting that it acts on multiple pathological pathways in DN and may serve as a useful therapeutic adjunct.

Although the direction of effect was generally uniform, considerable heterogeneity existed among studies. This variation could be due to differences in animal species, model induction techniques, dose, treatment period, and measurement methods. Subgroup analyses showed that rutin exerted protective effects across nearly all categories, without reversing trends. The variation observed in BUN and Scr may relate to inconsistent doses and intervention length across studies. Publication bias was suggested by funnel plots and Egger’s test. After applying the trim-and-fill method, the pooled results remained largely unchanged, supporting the robustness of the findings. Our three-dimensional dose–time visualization indicated that most effective data were clustered around 25–100 mg/kg for 4–8 weeks, implying a relatively concentrated therapeutic range for rutin.

Integrating our meta-analysis with existing mechanistic evidence suggests that rutin protects the kidney through a network of actions rather than a single target. Its benefits arise from coordinated effects on oxidative stress, inflammation, fibrosis, and metabolic disturbance, which together slow the pathological progression of DN (Figure 13). To avoid overstating mechanistic certainty, we distinguish pathways directly supported by pooled quantitative outcomes from those discussed as narrative, hypothesis-generating interpretations based on individual experiments. In this meta-analysis, pooled data support mechanistic domains related to metabolic regulation, oxidative stress/antioxidant defense, and fibrosis-related signaling. In contrast, inflammatory pathways were not quantitatively synthesized because inflammatory outcomes were insufficiently and inconsistently reported across the included studies; therefore, inflammation-related signaling is discussed only to provide biological context. In addition, since blood glucose was also reduced in pooled analyses, the observed renal benefits may reflect both kidney-specific actions and indirect improvements related to metabolic control.

This study systematically evaluated the preclinical evidence of rutin in the treatment of DN based on existing animal experiments, but there are still certain limitations. Firstly, the included studies showed significant differences in animal species, strains, modeling methods (STZ, alloxan, db/db), dose range (10–200 mg/kg), intervention duration (4–12 weeks), and experimental conditions. Such methodological and biological heterogeneity may affect the stability and generalizability of the effect size. Moreover, diagnostic criteria and glycemic thresholds for model establishment varied across studies, and baseline disease severity was often insufficiently reported, which may have reduced the comparability of outcomes and introduced residual confounding. Additionally, sex was insufficiently considered across studies. As most experiments used male animals, and sex-stratified outcomes were not reported, potential sex-specific responses to rutin could not be assessed. Secondly, only 13 studies and 318 animals were ultimately included, with a relatively small overall sample size. This may reduce the statistical power of the meta-analysis and increase the risk of publication bias, which may increase the likelihood of an overestimation of the therapeutic effect.

In addition, data from some studies were extracted from graphs using WebPlotDigitizer. Even after cross-validation by two researchers, there may still be minor errors. In terms of methodological quality, most of the literature fails to adequately report key information such as randomization, blinding, and allocation concealment, resulting in a high or unclear risk of bias. In particular, the lack of reported allocation concealment may have contributed to an overestimation of effect sizes and limits the internal validity of individual studies. Although our stratified analysis revealed no significant difference between high-risk and low-risk studies, the observed effect size differences between medium-risk and low-risk studies suggest that bias risk may have influenced some of the reported effects. Therefore, the pooled effect sizes in this meta-analysis should not be interpreted as precise or unbiased estimates of treatment efficacy, but rather as a quantitative summary of the overall direction and consistency of effects across studies. Accordingly, our conclusions should be interpreted cautiously, especially for outcomes with substantial heterogeneity. Nevertheless, the consistency of findings observed across multiple independent experiments and different animal models suggests that the observed associations are unlikely to be driven by a single study or methodological artifact. Finally, there are essential differences between animal models and human DN in terms of disease course characteristics, metabolic status and pharmacokinetics, which makes the convertibility of the results in clinical practice uncertain. Another translational limitation is that no included animal studies performed head-to-head or combination comparisons between rutin and established standard-of-care therapies for DN (e.g., ACEi/ARB or SGLT2 inhibitors); therefore, the relative efficacy and potential add-on benefit of rutin remain uncertain. Future studies should incorporate active-comparator and add-on designs with clinically relevant background therapy to strengthen translational interpretation.

More broadly, to further strengthen the preclinical evidence base, future research should adopt more standardized and rigorous experimental designs, expand sample sizes, and improve reporting transparency to strengthen the robustness and reproducibility of preclinical evidence. Greater attention to baseline disease severity, consideration of sex as a biological variable, and harmonization of outcome measures across DN models would further enhance translational relevance. In addition, systematic evaluation of dose–time relationships, together with basic pharmacokinetic and safety profiling, may help reduce uncertainty prior to clinical translation. Meanwhile, deeper mechanistic investigations using multi-omics approaches could further support the preclinical and clinical development of rutin.