This review is designed and implemented according to the Systematic Review and Meta-Analysis (PRISMA) Preferred reporting project (Chitra et al., 2015), which can be reviewed against the published protocol (INPLASY202360084). (See Annex 1 for search information).

The criteria for inclusion in meta-analyses follow the Participant Type, Intervention, Comparison, Outcome and Type of Study (PICOS) framework. Type of participant (P): Included all pure PF animal models without species, sex, and modeling method restrictions, and excluded non-PF animal models and other disease-related PF animal experiments. Type of intervention (I): Included any type of CUR intervention, regardless of dose, platform of administration, route of administration, time of administration, or duration, and CUR was not the only intervention. While CUR analogues were excluded. Comparison (C): The model group was modeled only and received the same amount of non-functional substances (normal saline or distilled water) or received no intervention. No control group was excluded from the study. Result type (O): No predetermined result indicator was excluded. Type of study (S): All controlled studies of the administration of CUR in PF animals, reviews, conference abstracts, in vitro, human, or clinical studies were not considered.

We had the primary outcome: hydroxyproline (HYP) content. The secondary outcomes were TGF-β concentration, myeloperoxidase (MPO) activity, tumor necrosis factor (TNF)-α concentration, malondialdehyde (MDA), NO, GSH.

To identify animal reports on the use of CUR for PF, we searched the following electronic literature databases from 1 January 2000 to 19 April 2023, Pubmed, Embase, Web of Science, and the Cochrane Library. Searches of PubMed database mainly used a combination of MeSH and free text terms to identify diseases, interventional drugs, and animals, and the search algorithm adjusted the search terms according to different database requirements. Specific 4 database search strategies are described in the Supplementary Materials.

Two reviewers independently conducted initial screening based on titles and abstracts, then the full text of potentially eligible articles was screened and finalized, with any disagreements resolved through discussion and negotiation, before the other two reviewers scrutinized the data.

Two reviewers independently extracted data manually from the included literature, and the Excel table included the following items: 1) Publication details: first author’s last name and year of publication. 2) Details of the animal (sample size, species, age, sex, weight, and PF animal model). 3) Treatment information: source, concentration, route of administration, delivery system, duration of intervention, and dose of CUR. 4) Observation indicators: All outcome indicators were continuous data, so the mean values and standard deviations of the control group and the intervention group were extracted. If there are several doses or intervention time available in an experiment, the highest dose or the longest intervention duration should be selected for analysis in order to better reflect the characteristics of CUR action. If there are multiple sample sizes for the same outcome indicator, the minimum sample size will be selected for analysis in order to reduce the risk of bias. When the raw data is only available in the graph, the GetData Graph Digitization software (version 2.26) will be used to estimate the data.

Quality assessment was carried out independently by two reviewers and any differences were resolved through discussion, negotiation, and discussion with the third reviewer. The Center for Systematic Review of Laboratory Animal Studies (SYRCLE) Risk of Bias (RoB) Tool, a 10-item scale was used to assess the quality of included animal studies, indicated by “yes”, “No” and “unclear” when the assessment results were “low,” “high” and “unclear.”

Statistical analysis data from all included studies were summarized using Revman 5.3 and Stata 16.0. All results were continuous variables, so we used the standard mean difference (SMD) of 95% confidence intervals (CIs) to represent the intervention effect. The heterogeneity between the study and the subgroup was evaluated using I² statistics. When I² ≤ 50%, the heterogeneity of the included study was small, and the fixed-effect model was used for analysis. And when I²> 50%, a random effects model was used for meta-analysis, while subgroup analysis and sensitivity analysis were used to investigate sources of heterogeneity. If the results were at least 10 animal trials, publication bias was assessed using funnel plots, Egger tests, and Begg tests.

Figure 2 is a flow chart of search, inclusion and exclusion through a preliminary search of the following four electronic databases: PubMed, Web of Science, Embase and Cochrane Library, we found 550 studies and 439 studies remained after removing duplications. After reading the titles, abstracts and full text, 401 studies were excluded, and a total of 38 studies met the inclusion criteria. Eleven studies (Xu et al., 2007; Chen et al., 2008; Zhang et al., 2011; Avasarala et al., 2013; Chen et al., 2018; Gouda and Bhandary, 2018; Shaikh et al., 2020b; Durairaj et al., 2020; Shaikh and Prabhakar Bhandary, 2020; Shaikh et al., 2021b; Fathimath Muneesa et al., 2022) which did not include primary or secondary outcome measures were excluded. In the end, there were 27 studies were included (Punithavathi et al., 2000; Venkatesan, 2000; Punithavathi et al., 2003; Punithavathi et al., 2006; Zhou et al., 2006; Zhang et al., 2007; Zhao et al., 2008; Jiang et al., 2009; Lee et al., 2010; Smith et al., 2010; Hamdy et al., 2012; Cho et al., 2013; Tyagi et al., 2014; Banerjee et al., 2015; Li et al., 2015; Kar et al., 2016; Tyagi et al., 2016; Chen et al., 2017; Hu Y. et al., 2018; Chen et al., 2019; Gouda M. M. et al., 2020; Barsan et al., 2021; Durairaj et al., 2021; Hemmati et al., 2021; Hosseini et al., 2021; Miao et al., 2021; Kumari and Singh, 2022), two of them (Smith et al., 2010; Miao et al., 2021) were included in this meta-analysis, each of which included two independent experiments, so this study included a total of 27 studies and 29 independent experiments.

We present the characteristics of the included studies in Annex 1. A total of 396 experimental animals (201 in the intervention group and 195 in the control group) were included in this meta-analysis. All publications were published between 2000 and 2022, and most of the research was conducted in China (n = 9) and India (n = 10). Sample sizes ranged from 6 to 35. Wistar rats, Sprague—Dawley rats, BALB/c mice and C57BL/6 mice were preferred. Rat weight fluctuated between 160 and 350 g and mice weight fluctuated between 20 and 30 g. Of the studies detailing the sex of the animals, female (n = 3), male (n = 18), and female half and half (n = 1), six did not report sex and only seven reported the age of the animals in detail.

All studies controlled the dose of CUR between 30 and 1000 mg/kg, but the dose varied widely. Researchers preferred gastric tube administration (n = 9) and intraperitoneal injection (n = 5). CUR carriers or mediators were reported in 13 studies, and the source of CUR was reported in 20 studies, and 12 of them were provided by Sigma-Aldrich, United States. In terms of time point of drug intervention, each independent experiment selected prevention (n = 2), treatment (n = 19), and combination of prevention and treatment (n = 8) for intervention, respectively. In terms of modeling methods, bleomycin (BLM) (n = 13), paraquat (PQ) (n = 7), and silica (n = 4) had the largest selection of independent experiments.

According to SYRCLE’s Risk of Bias tool, the risk of bias and quality of all included studies were assessed. The evaluation results are as follows: 2 studies (Zhang et al., 2007; Chen et al., 2017) were identified as “low risk” because they used a random number table, while the remaining 25 studies were identified as “uncertain” in terms of random sequence generation and allocation of different groups. Among them, 12 studies (Zhou et al., 2006; Zhao et al., 2008; Jiang et al., 2009; Cho et al., 2013; Tyagi et al., 2014; Li et al., 2015; Kar et al., 2016; Tyagi et al., 2016; Chen et al., 2019; Hemmati et al., 2021; Hosseini et al., 2021; Kumari and Singh, 2022) only described “randomization” without explaining the randomization method, and the rest of the studies lacked explanation on this aspect. Not all studies reported whether the distribution of relevant baseline levels was balanced between the experimental and control groups, so the “baseline characteristics” were “uncertain;” 10 studies (Venkatesan, 2000; Punithavathi et al., 2006; Cho et al., 2013; Tyagi et al., 2014; Tyagi et al., 2016; Chen et al., 2019; Durairaj et al., 2021; Hosseini et al., 2021; Miao et al., 2021; Kumari and Singh, 2022) were rated as low risk because they detailed the consistency and randomness of the conditions and environment in which laboratory animals were kept; None of the included studies mentioned blinding animals during the implementation of the intervention; 7 studies (Zhou et al., 2006; Zhao et al., 2008; Smith et al., 2010; Tyagi et al., 2014; Chen et al., 2017; Chen et al., 2019; Kumari and Singh, 2022) only described the “randomized” selection of experimental animals for result evaluation, without explaining the randomization method, and the other studies did not explain this aspect. In the measurement of results, only 2 studies (Punithavathi et al., 2006; Cho et al., 2013) reported the blinding of surveyors and were identified as “low risk,” while the other studies were identified as “uncertain.” 3 studies (Li et al., 2015; Chen et al., 2017; Chen et al., 2019) were rated as high risk for “loss of follow-up bias,” because they had animal death phenomenon but did not explain whether the missing data affected the authenticity of the results, and the other studies were rated as “low risk.” All studies were rated as “low risk” when assessing “selective reporting bias” and other biases (Figure 3).

Forest maps were created based on 19 independent experiments from 17 studies (Punithavathi et al., 2000; Punithavathi et al., 2003; Zhou et al., 2006; Zhang et al., 2007; Zhao et al., 2008; Lee et al., 2010; Smith et al., 2010; Hamdy et al., 2012; Banerjee et al., 2015; Tyagi et al., 2016; Chen et al., 2017; Hu Y. et al., 2018; Durairaj et al., 2021; Hemmati et al., 2021; Hosseini et al., 2021; Miao et al., 2021; Kumari and Singh, 2022). Only one of them originated from bronchoalveolar lavage fluid (BALF) (Durairaj et al., 2021) and all others were from lung tissue showed that: In comparison with the control group, CUR significantly reduced HYP content (SMD = −4.96; 95% CI = −6.05 to −3.87; heterogeneity χ2 = 77.37, df = 18, p = 0.000, I² = 76.7%, Figure 4A). To explore the heterogeneity among the studies, the results of meta regression showed: The effect size of HYP content was correlated with the choice of intervention time (t = −3.52, p = 0.004). Animal model of type (t = −0.32, p = 0.751), animal sex (t = −0.04, p = 0.966), dosing mode (t = −1.36, p = 0.197), treatment time (t = 2.13, p = 0.055), and animal models in model selection (t = −1.2, p = 0.253) were irrelevant. According to the pre-determined subgroup analysis: In animal experiments, Wistar rats (p = 0.000), female (p = 0.000), gastric intubation (p = 0.000), Pre-treatment combined treatment (p = 0.003), and more beneficial effects were observed in less than 7 days of treatment (p = 0.020) and treatment of SiO2 models (p = 0.004).

For outcome measures other than GSH, we performed sensitivity analysis by omitting data from one independent experiment at a time and calculated aggregate data from the remaining studies to demonstrate that the results of all the remaining outcome measures were robust. (See Annex part 2).

Due to the high heterogeneity among studies, we evaluated relevant indicators in terms of animal species, animal sex, dosing mode, intervention time selection, treatment time, and PF model selection. The results showed that animal species and route of administration may be the heterogeneous sources of HYP content. (See Annex part3).

In order to further explore publication bias and heterogeneity, egger and begg tests were carried out for HYP content. Egger: p = 0.000, t = −8.54; begg: p = 0.001, z = 4.16, the results showed that there was publication bias in HYP content. (See Annex part 4).

To our knowledge, no previous meta-analysis has quantitatively assessed the efficacy of CUR in the treatment of PF. In the meta-analysis, which included 27 publications and 29 studies involving 396 animals and was of low methodological quality overall, CUR significantly improved the degree of fibrosis, levels of inflammation, and oxidative imbalances in lung tissue in animal models of PF. This evidence suggests that CUR acts as an antifibrotic agent in animal models of PF through several mechanisms. These findings can provide scientific basis for clinical trials of CUR in the treatment of PF.

We further performed subgroup analysis and meta regression for the main outcome HYP content. The results of subgroup analysis and meta-regression indicated that animal species and route of administration may be the source of heterogeneity of HYP content. However, due to the small sample size, these results should be interpreted with caution and more studies are needed to provide precise evidence. Animal sex, intervention time selection, course of treatment, and PF model selection may not be sources of inter-study heterogeneity. Therefore, we speculate that heterogeneity may arise from other differences in the studies, including the dose of CUR, the source of the sample, the modeling method, etc. In addition, we conducted sensitivity analysis on all the results, and the results did not change significantly, indicating that the results were stable.

This meta-analysis did not limit the type and sex of experimental animals. The included animals were mice and rats. Through subgroup analysis, we found that animal species was one of the sources of heterogeneity, and the effect on HYP content was the greatest in Wistar rats. Considering that different models have different mechanisms and sensitivities to PF, further research is needed.

This meta-analysis did not limit the types of modeling. BLM was found in clinical application to induce PF in patients, but this model did not replicate the two characteristics of slow and irreversible progression of human IPF (Li et al., 2022). The advantage of silica model is that silica particles in the lungs are difficult to remove, which can form a persistent stimulus (Degryse and Lawson, 2011). Through subgroup analysis, we found that the CUR model has the greatest effect on HYP content. The inclusion of animal models including radiation, amiodarone, cyclophosphamide, etc., can compensate for the limitations of the PF model to a certain extent, and make a more comprehensive response to the therapeutic effect of CUR in treating PF.

This meta-analysis did not restrict the route of administration for CUR, and meta-regression and subgroup analysis indicated that the route of administration might be the source of heterogeneity in HYP content, and CUR achieved better effect size in HYP through gastric intubation intervention. In view of the considerable differences in the absorption rate and bioavailability of CUR by different routes of administration, as well as the fact that we do not know the conversion relationship between different routes of administration, dose-related subgroup analysis and meta regression were not performed in this meta-analysis. We carefully read all the included studies to clarify the effect of dose versus effect size.

In the PF model, the first 7 days were dominated by alveolitis, and the inflammation decreased and gradually produced extensive fibrosis in the later stage. Subgroup analysis showed that treatment of less than 7 days had a better effect on HYP content, and the effect in this case may reflect more anti-inflammatory effects by blocking the early response. For the choice of intervention time, pretreatment combined therapy achieved better efficacy in HYP content.

Based on the above findings, we suggest that in the treatment of PF, more attention should be paid to the choice of administration mode, dose, administration time and intervention time of CUR to determine the best intervention mode and course of treatment.

In our study, we found that more attention was paid to the diagnostic significance and mechanism of genetic genes for PF in modern times. Studies have shown that the prevalence of interstitial abnormalities (ILA) in those related to patients with IPF was 15%. At the same time, a perspective from the PF foundation genetic testing work group indicates that attention should be paid to the clinical significance of genetic diagnosis of PF and proposes that gene sequencing and TL measurement for PF patients are meaningful. This seems to be one of the explanations for heterogeneity, but there is still a lack of relevant scientific research, which also provides ideas and directions for further research.

No side effects were observed in any of the included studies, and further study of the safety of CUR in the treatment of PF is the focus and direction, which will also help to observe and understand the limitations of CUR therapy. Based on included studies, we endeavored to find studies on curcumin in the hope of finding reports of related toxicity or adverse reaction studies. We would like to refine and add to the reports in this area, but there is a lack of various pharmacologic studies suggesting that it is safe to use, and no reports of toxicity/adverse reactions yet. In various animal models and clinical studies, CUR has been shown to be very safe, even at doses as high as 8 or 12 g/day. Based on the above findings, the following two points need to be noted in future studies: First, the association between the dose and intervention time of CUR and adverse reactions needs to be further studied. Second, some researchers do not report adverse effects, which can mislead people to believe that there are no adverse effects. We call on researchers to standardize the reporting of animal experiments, including adverse events.

First, the measurement methods used in different laboratories were different, and the measurement units were different. Although we used standardized mean differences to reduce the statistical effect size, we still did not completely eliminate bias, which was an important factor in study quality. Second, although we used meta-regression and pre-specified subgroup analyses, due to the limited sample size and insufficient statistical power of each subgroup, no source of significant heterogeneity could be found, which could result from different animal ages, different routes of administration, and different treatment initiation times. In addition, methodological defects might be the key factor leading to the high heterogeneity among studies. The evaluation data were not original data, but from electronic data calipers, which might cause the error of the results to some extent. Third, none of the studies reported the calculation of sample size, and the number of included studies was relatively small, which might have a large positive bias effect, and the existence of publication bias confirmed that the real effect of CUR could be overestimated. Finally, the lack of data from large animals in our statistical analysis, as they shared more pathophysiological features with humans, might also limit the interpretation and extension of our results. So large animal studies are needed to further confirm the favorable effects of CUR on PF.