PubMed, Embase, The Cochrane Library, Web of Science, CNKI, Wanfang, VIP, and CBM were the eight databases we scanned through. All databases are available from their inception to 21 April 2022. The main search terms are: “Chinese herbal medicine”, “painful diabetic neuropathy”, “Chinese medicine”, and “diabetic neuropathic pain”.

Two researchers (ZP-S and WY-L) independently imported all the retrieved literature into the software EndnoteX9.0 for management and screening. For controversial literature, two researchers negotiate with a third researcher (YT-X).

The inclusion criteria were:1) participants: PDN patients (regardless of race, gender, or age); 2) study format: RCTs; 3) Interventions: the intervention group received CHM treatment (whether CHM as a single therapy or adjuvant therapy) and the control group used western medicine (WM) or placebo; and 4) outcomes: the main outcome indicators include SCV, MCV, pain score, and clinical efficiency The Secondary outcome indicators include TCM syndrome score and adverse events. The exclusion criteria were:1) treatment using acupuncture, massage, or other Chinese medicine; 2) Intervention time is not appropriate; and 3) patients without clear diagnostic criteria or accompanied by other diseases.

Herein, the definition of clinical efficiency in each trial is not the same, and the clinical efficiency of the included trials is based on the following criteria. a: Effectiveness: peripheral nerve function or clinical symptoms improved. b: Ineffectiveness: peripheral nerve function or clinical symptoms were not significantly improved or not improved (Xiaoyu, 2002).

Data was separately extracted and cross-checked by the two researchers (ZG-L and JH-Z). The extracted data primarily include: 1) the fundamental information of the selected research; 2) key elements of bias risk assessment; 3) outcome data: if the data type were measurement data, the mean and standard deviation were extracted, if the data were count data, the number of events and the total number were extracted.

Using the Risk of Bias instrument developed by the Cochrane Collaboration (Higgins et al., 2011), two researchers (MS and JY-Z) independently assessed the caliber of RCTs and cross-checked their results. The following seven components made up the evaluation content: the creation of random sequences, the concealment of allocations, the blinding of individuals and researchers, the integrity of outcome data, report bias, and other biases.

The meta-analysis of the research data was carried out using Stata16.0 software. The effect analysis statistic for categorical data was OR, while the effect analysis statistic for continuous data was either MD or SMD. For each effect, the 95% CI was calculated. The χ ² test (test level = 0.10) was utilized to examine the heterogeneity of the outcomes of these studies, and the I² test was used to quantify the heterogeneity. If p≥0.10 and I²<50%, the fixed effect model was used for analysis. If p<0.10 and I² ≥ 50%, it indicates that a huge heterogeneity appeared among the studies, and then subgroup analyses have been conducted so as to make out the origin of heterogeneity. The analysis was carried out using a random effect model when methodological heterogeneity and clinical heterogeneity are absent. α = 0.05 was used as the meta-analysis test level. Through sensitivity analysis, the stability and reliability of the analysis’s results are examined. Publication bias was evaluated using the funnel plot, Begg’s test, and Egger’s test.

A sum of 492 papers was found during the preliminary screening, however, 184 papers were eliminated due to repetition, 247 papers were eliminated by reading their titles and abstracts, and 40 papers were eliminated by reading the complete text. Finally, the quantitative analysis covered 21 articles (Hongwei et al., 2008; Honggang, 2011; Laibiao et al., 2012; Tsai et al., 2013; Fanrong and Mingkong, 2014; Laibiao et al., 2015; Ruixia et al., 2015; Haibo, 2016; Xiaorui, 2016; Di et al., 2017; Cizhen, 2018; Honggang et al., 2018; Liting et al., 2018; Qiaren and Xiaohong, 2018; Tao et al., 2018; Ye et al., 2018; Dianrong et al., 2020; Haiyan and Yange, 2020; Yanli et al., 2020; Qingqing, 2021; Shuquan et al., 2021). Figure 1 depicts the literature screening procedure.

There are 1737 patients altogether in the sample size of the 21 studies, which include 871 patients in the intervention group and 866 patients in the control group. The intervention group and the control group had equivalent pre-treatment data (such as age, sex ratio, outcome indicators, etc.). All 21 studies were from China (Table 1, Table 2).

Among the 21 studies included, 7 studies (Tsai et al., 2013; Fanrong and Mingkong, 2014; Ruixia et al., 2015; Di et al., 2017; Cizhen, 2018; Tao et al., 2018; Shuquan et al., 2021) were grouped by a random number table, 1 study (Yanli et al., 2020)used a draw for grouping, 1 study (Haibo, 2016) adopted the odd-even number method for grouping, and the remaining studies did not explain the specific methods. The allocation concealment method was not discussed in any of the 21 studies. One study (Tsai et al., 2013) used double-blind, and the remaining research failed to state whether or not blinding was utilized. All studies’ outcome data were complete, and there were no other biases or selective reporting results (Figure 2).

Among the 21 papers included, adverse occurrences were described in 5 pieces of literature (Tsai et al., 2013; Laibiao et al., 2015; Ruixia et al., 2015; Xiaorui, 2016; Qingqing, 2021), covering liver function, renal function, blood routine, urine routine, and digestive system. Four of them (Laibiao et al., 2015; Ruixia et al., 2015; Xiaorui, 2016; Qingqing, 2021) stated that the intervention group’s incidence of negative events was lower compared to the western medicine intervention group. One paper (Tsai et al., 2013) reported that the intervention group had more negative occurrences than the placebo group did. The commonest adverse reactions were stomach discomfort, nausea, dry mouth, etc. However, all adverse reactions were not treated specially, and the symptoms gradually relieved or disappeared. Although the above results suggest that CHM in the treatment of PDN is safe because the sample size is small, more large sample clinical studies are needed to prove the conclusion.

Individual studies were excluded one by one for sensitivity analysis. The findings indicate that after removing the studies, there was no substantial change in the outcomes for any outcome indicators, revealing that the results were stable.

The funnel plot showed that there was asymmetry in the pain score (Figure 10A), and the symmetry of the clinical effective rate (Figure 10B) and the symmetry of the median nerve SCV (Figure 10 C) were acceptable. Using Begg’s test and Egger’s test, it showed that there was a substantial publication bias in the pain ratings (p = 0.002; p < 0.05), but not in the clinical effective rate (p = 0.216; p = 0.357) or SCV of the median nerve (p = 0.721; p = 0.157).

In this study, a total of 492 publications were found, and 21 articles were used for the meta-analysis. The primary findings of this meta-analysis demonstrated that CHM, whether used as an adjuvant therapy or a stand-alone treatment, enhanced nerve conduction velocity and clinical efficacy during the therapy of PDN and reduced pain scores and TCM syndrome scores. To identify the reasons for heterogeneity, we carried out subgroup analyses based on intervention time, intervention methods, sample size, and so on. The findings of the subgroup analysis demonstrated that one of the reasons of heterogeneity in the SCV of the median nerve, SCV of the common peroneal nerve, and pain score was the intervention time. The sample size is one of the heterogeneous sources of SCV of median nerve, SCV of common peroneal nerve, MCV of common peroneal nerve, and MCV of median nerve. The pain rating scale is also one of the heterogeneous sources of pain scores. The study’s results are steady and dependable, according to sensitivity analysis. According to the publication bias test, there is a risk of bias in this study. In addition, we found that adverse events of CHM treatment are less than conventional western medicine treatment, indicating that Chinese herbal treatment of PDN is safe. Therefore, we provide supporting evidence that CHM is effective and safe in treating PDN.

A total of 72 Chinese medicines were involved in all formulations. Ranked according to the frequency of Chinese herbal medicines, the top 15 flavors of Chinese medicine frequency distribution are shown in Table 4, of which Astragalus mongholicus Bunge [Fabaceae; Astragali radix], Angelica sinensis (Oliv.) Diels [Apiaceae; Angelicae sinensis radix], and Cnidium monnieri (L.) Cusson [Apiaceae; Chuanxiong rhizoma] frequency rank among the top three, whose frequency is more than 10 times. This conclusion agrees with the findings of Zhang Fuzhi et al. (Fuzhi et al., 2020).

Chemical constituents are the key material basis for revealing the efficacy of CHM (Chuan, 2017). Astragaloside IV, which is the main chemical component of Astragalus, can inhibit Schwann cell apoptosis, regulate nerve growth factor gene expression, increase NA + -K + -ATPase, inhibit erythrocyte aldose reductase activity, thereby inhibiting nerve injury, enhancing the speed of motor nerve conduction and minimizing the buildup of nerve and erythrocyte advanced glycation end products (Yuanyuan et al., 2021). Xie et al. have shown that quercetin, a common flavonoid compound in astragalus, can regulate the expression of intestinal flora and reactive oxygen species in diabetic peripheral neuropathy rats to improve peripheral nerve injury (Xie et al., 2020). Isoeugenol, one of the main components of Angelica Sinensis volatile oil, plays an important role in reducing oxidative stress markers and glutathione levels in the sciatic nerve area (Prasad and Muralidhara, 2013). Additionally capable of removing oxygen free radicals and minimizing peroxide damage are ferulic acid and angelica polysaccharide (Gong et al., 2016). One of the key ingredients of Chuanxiong Rhizoma is ligustrazine, which has the ability to up-regulate the expression of heme oxygenase 1/carbon monoxide and superoxide dismutase and down-regulate the expression of tumor necrosis factor-, nitric oxide synthase/nitric oxide, and malondialdehyde. It is proved that ligustrazine can enhance the body’s antioxidant and anti-inflammatory effects and play a protective role in diabetic pain neuropathy (Fangqin et al., 2021).

Wnt protein is an important signalling molecule. The wnt signalling pathway mediates a variety of biological processes in the body, such as embryogenesis, organogenesis and tumour formation (Sharma et al., 2022). Among them, the wnt signalling pathway is involved in the development of T2DM and its complications by directly influencing the differentiation and proliferation of pancreatic β-cells as well as the secretion and action of insulin (Nie et al., 2021). This provides a direction to explore the mechanism of action of CHM in the treatment of diabetes and its complications. It (Zou et al., 2016; Wang et al., 2020; Zhang et al., 2020; Wu et al., 2022) was found that astragaloside, angelica polysaccharide and chuanxiongzine could all activate or inhibit the wnt signalling pathway in tissues or cells to regulate glucose and lipid metabolism, tissue or organ repair in T2DM patients.

In addition, the traditional Chinese medicine involved in this study is not only effective in the treatment of PDN, but also has a good effect on type 2 diabetes and other complications (such as diabetic heart disease, diabetic retinopathy). Studies have found that ligustrazine can delay the development of DR by inhibiting oxidative stress and retinal ganglion cell apoptosis, down-regulating AGEs content and other ways (Wei, 2021). Astragalus polysaccharide alleviated mitochondrial damage and apoptosis induced by metabolic memory by regulating the miR-182/Bcl-2 axis (Nie et al., 2019). Astragaloside IV improves endothelial dysfunction in thoracic aortas from diabetic rats by reducing oxidative stress and calpain-1 (Gao et al., 2021).

Although there are more and more clinical studies on the treatment of PDN with CHM, there is no systematic evaluation and meta-analysis in this direction. This study is the first study to systematically evaluate CHM for PDN, filling the evidence-based gap in CHM for PDN. The advantages of this meta-analysis mainly include a clear research topic, its selects high-quality RCTs that meet the inclusion criteria and conducted a statistical analysis of this study in strict accordance with the systematic review method. At the same time, we are more cautious about the explanation of the results. We conducted a subgroup analysis to find out the cause of heterogeneity, and we also conducted a sensitivity analysis and publication bias test. This study found that CHM can be recommended for the treatment of PDN, which offers fresh perspectives and ideas for researching PDN.

However, there are a few shortcomings in this work that deserve discussion. First of all, the majority of the included researchers did not use the allocation concealment and blind method, which could cause bias in both selection and implementation. Second, there was significant clinical heterogeneity in the 21 studies with differences in composition, dose, and dosage form of CHM, as well as differences in interventions (type of WM) and duration of intervention in the control group. This would result in a high level of heterogeneity within subgroups when subgroup analyses are performed. However, because of the limited number of studies, we cannot perform relevant subgroup analysis, which in turn affects the accuracy of the results. Third, the duration of the study’s intervention ranged from 4 to 12 weeks, and we could not assess the long-term safety of CHM treatment. Finally, since the randomized controlled trials included in this study are all from China, our study may not be extended to the world. Therefore, a large sample of multi-center studies is needed in the future.