We conducted a systematic literature search in the following clinical research databases: PubMed, Embase, Cochrane Library, Web of Science (WOS), China National Knowledge Infrastructure (CNKI), Wanfang Database, China Biomedical Database (CBM), and the China Science Journal Database (VIP), from the establishment of each database through June, 2021. We developed a search formula for the treatment of URSA by KTHs based on the PICOS strategy (Detailed strategies are provided in Supplementary Data Sheet S1), and then collated the screened literature using ENDNOTE software.

We also established a bibliometric analysis strategy for the CNKI and WOS databases (Supplementary Data Sheet S2). Between (“2011-01-01,” “2021-06-01”) and (literature classification was limited to traditional Chinese medical science, Chinese herb and integrated traditional Chinese and Western medicine); Search scope: general database. After downloading the refworks format, the synonyms were converted into Chinese and English, and the synonyms were clustered uniformly, including KTHs. The drug was defined as “KTH”; the selection used a modified g-index in each slice: g 2 ≤ k ∑ i ≤ g c i k ∈ Z + .

To include more or fewer nodes, increase or decrease the scale factor k = 25. The keywords co-occurrence analysis, cluster analysis, mutation term analysis, as well as time line graph analysis were performed using CiteSpace (Version 5.8.R1).

The systematic review and meta-analysis were conducted in accordance with the PRISMA guidelines (Moher et al., 2009; McInnes et al., 2017). The detailed research process is illustrated in the Supplementary Materials. All randomized controlled trials (RCTs) were included to study the efficacy of KTHs alone or combined with modern drugs, in the treatment of URSA. Studies reported in languages besides Chinese and English were excluded. Nonrandomized controlled trials or animal experiments were excluded.

Patients included were in their first 3 months of pregnancy, had a pregnancy confirmed by serum human chorionic gonadotropin (hCG) or ultrasound, and had a history of URSA diagnosis, which was defined as two or more spontaneous abortions. The following four causes were excluded: infection, abnormal parental karyotype, endocrine imbalance, and anatomical abnormalities, regardless of the maternal or gestational age, race or nationality, educational level, or economic status. The inclusion criteria also ensured that no treatment was received before pregnancy or before entering the trial. Trials with recurrent spontaneous abortion and nonpregnant URSA participants with a definite etiology were also excluded.

The treatment of interest was KTHs, regardless of dose, administration methods, administration time, or whether KTHs were used in combination with Western medicine. The KTH group was compared with a treatment using only Western medicine. A randomized controlled trial was excluded if the Western medicine changed within the control group. Studies using only bed rest and/or psychological support were also excluded.

The primary outcome was the clinical response rate and pregnancy outcome reported in the trials. The secondary outcomes included reported hormone levels, serum immunological parameters, and incidence of adverse events during treatment.

The search strategy was the same as the bibliometric analysis. The reference list of all identified articles was also manually searched to find possible related studies to supplement the relevant literature.

Two reviewers (Hang Zhou and Yi Yang) extracted the general information from eligible studies through a predesigned standardized data extraction table: first author, year of publication, TCM syndrome difference, sample size, age, abortion frequency, definition of abortion and live birth, intervention time, treatment intervention and control group, treatment time, and results. Any inconsistencies were resolved by a third reviewer (Yongqing Yao). The methodological quality of each individual study was independently evaluated by two researchers (Yi Yi and Hang Zhou) referring to the Cochrane Handbook (Cumpston et al., 2019; Propadalo et al., 2019) for systematic review of interventions. We used the following criteria for evaluation: random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome evaluation, incomplete outcome data, selective reporting, and other biases. Each study was classified as either low-risk, high-risk, or unclear. If there were differences in opinion, the third researcher was referred to (Yongqing Yao).

A meta-analysis was conducted using the Review Manager (Revman) (computer program; version 5.3, Copenhagen: Nordic Cochrane Center, Cochrane Collaboration, 2014). The relative risk (RR) of 95% confidence interval (CI) was used for binary variables, while weighted mean variance (WMD) and 95% confidence intervals were used for continuous variables. Cochrane’s p-value and I2 were used to test the heterogeneity of the study.

A total of 1,012 articles were obtained through the literature searches, and the number of published articles increased annually (Figure 2A). Additionally, Citespace was used for keyword colinear analysis (Figure 2B), and the polar coordinate histogram was calculated according to the count value (Figure 2C). Analysis from the publications over the past 10 years showed that the research hotspots for traditional Chinese medicine in the treatment of RSA mainly focused on the following key words: recurrent spontaneous abortion, TCM treatment, tonifying kidneys, and promoting blood circulation, KTH, integrated traditional Chinese and Western medicine, clinical research, experience of famous doctor, prethrombotic state, progesterone, kidney deficiency and blood stasis, the damage of pre culture, and deficiency of spleen and kidney (Yang et al., 2013; Obstetrics; Subgroup et al., 2019; Cao et al., 2021; Li et al., 2021). The formula was used to show the keyword saliency map (Figure 2D), with a total of seven keywords. Finally, a clustering time line chart was constructed. The results showed that the main related research areas were divided into one of four categories: TCM treatment, USRA, KTH, and famous doctor experience (Figure 2E).

After excluding duplicate studies, 562 studies were examined according to their abstracts and titles, resulting in 41 articles in the final evaluation. Finally, nine studies (Liu et al., 2015; Xu et al., 2018; Feng, 2019; Feng et al., 2019; Han et al., 2019; Feng et al., 2020; He, 2020; Niu and Hao, 2020; Xu et al., 2020) were included in the present systematic review.

Table 1 summarizes the basic information pertaining to the randomized controlled trials included. All of these trials were published in China. These studies included 1,054 URSA patients, of which 629 participants were assigned to the treatment group, and 425 participants were assigned to the control group. There was no significant difference in the baseline parameters of all trials.

The patients in the treatment group were treated with KTHs or combined with Western medicine, while the patients in the control group were treated with Western medicine alone. Of those, six randomized controlled trials used natural progesterone, including injection, capsule, or desgesterone (Liu et al., 2015; Feng, 2019; Feng et al., 2019; Han et al., 2019; He, 2020; Niu and Hao, 2020). One study used natural progesterone in combination with other treatments (Mekinian et al., 2016), and one used active immunotherapy (Xu et al., 2018). The treatment duration ranged from 4 to 20 weeks of gestation.

Overall, the methodological quality of the included trials was poor. All nine studies were designed with two arms and were declared randomized controlled trials. None of the trials reported any details relating to the blinding of patients and researchers. No trials showed the number and reasons of dropouts. The evaluation results are shown in Figure 3.

Although the funnel plot of early pregnancy loss rate was asymmetrically distributed, Egger’s test analysis showed only marginally significant publication bias (p = 0.09). The sensitivity analysis of early pregnancy loss rates showed that the effect evaluation remained unchanged, indicating no strong publication bias of the combined results.

Three studies (Feng et al., 2019; Niu and Hao, 2020; Xu et al., 2020) included changes in indicators related to coagulation and anticoagulation, such as PT, APTT, TT FIB, t-PA, and PAI-I in URSA patients before and after treatment with KTHs. Unfortunately, due to the large variations in detection methods, modes of intervention, and modes of comparison among the studies, a quantitative analysis of this data was not possible. The wide variation in results across the studies precludes formation of a consistent conclusion. Finally, we developed a grade assessment for the meta-analysis results based on the Cochrane recommendation (Supplementary Data Sheet S3), but the quality of clinical evidence and grade of recommendation for these indicators are of concern.

A drug network was constructed for the nine articles screened (Figure 6A). Statistics were also performed on the drug doses and frequencies used (Figure 6B). The drugs with the top four tastants of frequency, which were set at ob ≥ 30% and DL ≥ 0.18 in the tcmsp database, were selected to screen the active ingredients of the screened drugs acanthopanax, Sambucus parasitica, and Cuscuta cuspida sequestration. Additionally, the ingredients of the four herbs in KTHs were also searched in the Batman database (Figure 6D), resulting in a total of 32 potential active ingredients after removal of any duplicates from both databases. Furthermore, 546 drug targets were screened using the Swiss target prediction database with strict criteria of the species as “human” and a probability greater than 0.6. A total of 708 disease targets for URSA were obtained after dereplication. Finally, 546 drug targets and 708 disease targets were established, and a Venn diagram was drawn using the venny2.1 online software mapping tool platform, and 113 drug disease common targets were obtained after intersection of the two sets (Figure 6E).

The 113 intersection targets were brought into the string database to construct the PPI network conditional on the species “human” and a confidence score ≥0.4. The results were imported into Cytoscape. Figure 7 shows the intersection target PPI network. The larger the node, the larger the degree value was. Next, the PPI network of URSA was further analyzed by mcode module with score cutoff = 0.2 and K core = 2. The clustering analysis was performed conditional on maximum depth = 100 and degree cutoff = 2. A total of 16 DEGs were obtained, and the top five DEGs with the highest score were taken (Figure 8B). In total, five DEGs (Figures 8C–G) and four core genes were obtained, and these five foundation clusters were scored. After screening, the four core genes obtained were mitogen-activated protein kinase (MAP1), matrix metalloprotinase1 (MMP1) (Zhang et al., 2021), ATP-binding cassette subfamily G member 2 (ABCG2), and recombinant caspase 1 (CASP1).

Using different drug complexes in KTH as active compounds, the core proteins analyzed by mcode were used as targets for the molecular docking validation (Figure 9A), and the best binding compounds were selected to construct the docking map. The binding energy of Sylvestroside III small molecules with MAPK1 protein was −7.6 kcal/mol, which proved to have a good binding interaction (Figure 9B). Sylvestroside III small molecules interact with the MAPK1 protein, mainly through the formation of hydrogen bonds as well as hydrophobic forces. Six hydrogen bonds formed with ARG-148, ARG-172, THR-63, GLN-62, HIS-61, and PHE-183, respectively, with hydrogen bond lengths of 3.2, 3.4, 3.0, 3.1, 3.0, and 3.3 Å. Sylvestroside III has hydrophobic interactions with LEU-170, ARG-67, GLN-66, GLU-186, GLY-182, THR-181, and HIS-180.

Japonine small molecules, on the other hand, also bind well to the MMP1 protein, with a binding energy of −6.3 kcal/mol (Figure 9C). Avicularin small molecules interact with the MMP1 protein, mainly through the formation of hydrogen bonds as well as hydrophobic forces, resulting in seven hydrogen bonds with PRO-238 and TYR-240 with a hydrogen bond of 3.3 and 3.3 Å, respectively. Japonine has hydrophobic interactions with ASN-180, LEU-181, VAL-215, ARG-214, HIS-218, GLU-219, and SER-239. Molecular docking results confirmed the results of the network analysis relating to the effect of KTHs when treating URSA.

After running mcoed core clustering through Metascape, GO analysis selected biological processes for analysis. The BP results showed that the intersection gene set was enriched to 2,045 biological process pathways (Figure 10A), mainly including: positive regulation of kinase activity, cellular responses to organic cyclic compound, responses to inorganic substance, cellular responses to nitrogen compound, phosphatidylinositol 3-kinase signaling, responses to oxygen levels, protein kinase B signaling, responses to molecules of bacterial origin, inflammatory responses, epithelial cell proliferation, and others. Core genes were selected by Metascape, with parameters min overlap = 3, p-value cutoff = 0.01, and min enrichment = 1.5. Twenty KEGG pathways were screened out, and the results of the top 20 formed a bar graph of KEGG functional enrichment (Figure 10B). Padjust represented the significance of enrichment; the more the red the color is, the more significant. Finally, based on the core target screening and molecular docking, together with the results of BP and KEGG enrichment analyses, the 32 potentially active ingredients from the lifespan pill were the inputs with the 113 drug disease cotargets into Cytoscape software to remove isolated ingredients with no intersection with targets. A network diagram of “drug ingredient target disease” interactions was drawn (Figure 10C). Using the score of 30 potentially active ingredients, and the average of the degrees as the screening criterion, we selected six compounds including Sylvestroside III, Japonine, sophranol, Gentisin, Lysine, and Matrine as core ingredients that were either large at the network nodes or showed high binding to protein targets (The screening process and criteria are detailed in Supplementary Data Sheet S4). The most important active compounds that were selected are listed in Table 2.