Chinese herbal compound for multidrug-resistant or extensively drug-resistant bacterial pneumonia: a meta-analysis and trial sequential analysis with association rule mining to identify core herb combinations

Purpose: Antibiotic-resistant bacterial pneumonia poses a significant therapeutic challenge. In China, Chinese herbal compound (CHC) is commonly used to treat bacterial pneumonia. We aimed to evaluate the efficacy and safety of CHC and identify core herb combinations for the treatment of multidrug-resistant or extensively drug-resistant bacterial pneumonia. Methods: Stata 16 and TSA 0.9.5.10 beta software were used for meta-analysis and trial sequential analysis (TSA), respectively. Exploring the sources of heterogeneity through meta-regression and subgroup analysis. Results: Thirty-eight studies involving 2890 patients were included in the analyses. Meta-analysis indicated that CHC combined with antibiotics improved the response rate (RR = 1.24; 95% CI: 1.19–1.28; p < 0.0001) and microbiological eradication (RR = 1.41; 95% CI: 1.27–1.57; p < 0.0001), lowered the white blood cell count (MD = −2.09; 95% CI: −2.65 to −1.53; p < 0.0001), procalcitonin levels (MD = −0.49; 95% CI: −0.59 to −0.40; p < 0.0001), C-reactive protein levels (MD = −11.80; 95% CI: −15.22 to −8.39; p < 0.0001), Clinical Pulmonary Infection Scores (CPIS) (MD = −1.97; 95% CI: −2.68 to −1.26; p < 0.0001), and Acute Physiology and Chronic Health Evaluation (APACHE)-II score (MD = −4.08; 95% CI: −5.16 to −3.00; p < 0.0001), shortened the length of hospitalization (MD = −4.79; 95% CI: −6.18 to −3.40; p < 0.0001), and reduced the number of adverse events. TSA indicated that the response rate and microbiological eradication results were robust. Moreover, Scutellaria baicalensis Georgi, Fritillaria thunbergii Miq, Lonicera japonica Thunb, and Glycyrrhiza uralensis Fisch were identified as core CHC prescription herbs. Conclusion: Compared with antibiotic treatment, CHC + antibiotic treatment was superior in improving response rate, microbiological eradication, inflammatory response, CPIS, and APACHE-II score and shortening the length of hospitalization. Association rule analysis identified four core herbs as promising candidates for treating antibiotic-resistant bacterial pneumonia. However, large-scale clinical studies are still required. Systematic Review Registration: https://www.crd.york.ac.uk/prospero/, identifier CRD42023410587.


Introduction
Lower respiratory tract infections are the fourth leading cause of mortality worldwide, claiming 2.6 million lives in 2019 (World Health Organization, 2020).Although several organisms are implicated, bacteria remains the primary cause of pneumonia.The introduction of antibiotics considerably improved bacterial pneumonia treatment efficiency.However, antibiotic overuse and misuse in hospitals and the agricultural industry have contributed to the emergence of drug resistance (Chawla et al., 2022).In recent years (Zaman et al., 2017;Duval Raphaël et al., 2019;Salam et al., 2023), multidrug-resistant (MDR) bacteria, defined as bacteria with resistance to at least three different classes of antimicrobial agents, have become common, promoting selection pressure toward extensively drug-resistant(XDR) bacteria.One study reported that 28%-75% of patients with hospital-acquired pneumonia (HAP) carried MDR bacteria (Catia et al., 2019).These bacteria, most often Pseudomonas aeruginosa (P.aeruginosa)and methicillinresistant Staphylococcus aureus (MRSA), are the main causes of early-onset and late-onset ventilator-associated pneumonia (VAP) (Torres et al., 2017).In addition to nosocomial pneumonia, recent studies have reported an increased risk of MDR and XDR bacteria spreading to the community (Van Duin and Paterson, 2020;NAIR and Niederman, 2021).In one study, the "PES" pathogens, including P. aeruginosa, extended-spectrum β-lactamase-positive Enterobacteriaceae, and MRSA, were responsible for 6% of hospitalized community-acquired pneumonia (CAP) cases and resulted in high mortality (Prina et al., 2015).Macaux et al. (Lou et al., 2018) reported that 35% of XDR bacteria carriers had not been hospitalized in the past 12 months.As the emergence of MDR and XDR bacteria reduces antibiotic efficacy, lower respiratory tract infections can not be controlled, leading to longer hospitalization durations, higher costs, and increased mortality (Venter, 2019;Yahav et al., 2021;Raycheva et al., 2022).The shortage of new antibiotics developed against MDR and XDR bacteria aggravates this problem (Terreni et al., 2021).
Currently, researchers have shifted their focus to developing new antibiotics from botanicals.Complex natural products with numerous molecular targets can reduce the incidence of resistance and thus exhibit considerable antibacterial activity against MDR bacteria (Filipa et al., 2020;Hafsa et al., 2022;Thangaiyan et al., 2022).Traditional Chinese medicine(TCM) is characterized by the use of botanicals, a rich source of bioactive phytocompounds.Many bioactive phytocompounds originating from TCM herbs, such as phenols, terpenoids, alkaloids, flavonoids, isothiocyanates, and indoles, hinder major drugresistant factors, such as efflux pumps, enzyme activity, membrane permeability, and other virulence mechanisms, including quorum sensing (QS) and biofilm development (Gupta and Birdi, 2017;Su et al., 2020;Li et al., 2022).Chinese herbal compound (CHC) represents a prevalent therapeutic approach in TCM.Several randomized control trials (RCTs) have demonstrated the efficacy of CHC alone or in combination with antibiotics in treating MDR and XDR bacterial pneumonia, improving the eradication rate of antibiotic-resistant bacteria, reducing the inflammatory response and adverse effects, and shortening the duration of hospitalization.However, the therapeutic efficacy and safety of CHC or CHC combination therapy remains indeterminate due to the lack of comprehensive evaluation of these RCTs.Therefore, this study aimed to produce a novel meta-analysis and trial sequential analysis (TSA) to reliably estimate the clinical effects and safety of CHC for treating MDR and XDR bacterial pneumonia.Association rule mining using the Apriori algorithm was performed to identify core herbal combinations that could serve as potential therapeutic agents.

Study selection
EndNote reference management software (version 20; Clarivate Analytics) was used for study selection.Two reviewers (DD.and YY.) separately screened the titles, abstracts, and full text for eligibility using the same criteria.Disagreements were resolved by a third reviewer (DY.).

Eligibility criteria 2.2.1.1 Type of studies
All RCTs, with or without blinding, were included in this study.No specified date or language restrictions were set.

Type of participants
Patients with any type of pneumonia (VAP, HAP, or CAP) caused by MDR or XDR bacteria were included without restrictions for sex, age, or race.Positive cultures from endotracheal aspirates or bronchoscopic sampling techniques were necessary for the diagnosis of antibiotic-resistant bacterial pneumonia.MDR bacteria were defined as bacteria unsusceptible to at least three different antimicrobial categories, and XDR bacteria were defined as bacteria unsusceptible to at least one agent in addition to two or fewer antimicrobial classes (Magiorakos et al., 2012).The pathogenetic diagnosis was based on at least two etiological examinations of cultures from lower respiratory tract secretions, including sputum, endotracheal aspirate, or bronchoalveolar lavage fluid.Furthermore, abnormal radiological findings indicating the presence of new or progressed pulmonary infiltrate(s) and clinical signs of infection, including the onset of fever (≥38 °C), increased sputum production or change in sputum color to a more purulent state, peripheral leukocytosis, and decreased oxygenation or the requirement of oxygen supplementation therapy, were required (Kalil et al.;Olson and Davis, 2020).

Type of intervention
RCTs comparing CHC in the treatment group with antibiotic treatment in the control group, or comparing combined CHC + antibiotic treatment with antibiotic-only treatment, with or without a placebo, were included.All forms of oral administration or nasal feeding, such as through decoction, granules, or capsules, were included.No restrictions were placed on the dose or duration of CHC therapy.

Outcome measures
The primary outcome was set as response rate, calculated as (total number of patients-number of patients without efficacy)/total number of patients.Inefficacy was defined as unaltered or worsening signs and symptoms, such as pulmonary shadows visualized on X-ray or CT, pulmonary rales, fever, cough, and chest pain.
Secondary outcomes included: 1) microbiological response, defined as bacterial eradication determined by etiological examination of clinical samples collected from the site of infection showing no bacterial growth at the end of treatment; 2) inflammatory indicators, including absolute white blood cell (WBC) count, procalcitonin (PCT) levels, and C-reactive protein (CRP) levels; 3) scoring systems that assessed the severity of pneumonia, prognosis, and drug efficacy, including the Clinical Pulmonary Infection Scores (CPIS) and Acute Physiology and Chronic Health Evaluation (APACHE)-II score; 4) safety, evaluated by the incidence of adverse events (AEs) and adverse drug reactions (ADRs).

Exclusion criteria
Exclusion criteria included: 1) cohort studies, case reports, reviews, and conference abstracts; 2) studies that lacked primary outcomes, control groups, or full-text articles; 3) participants diagnosed without an etiological examination or a drug sensitivity test, or those diagnosed with an etiological examination indicating chlamydia, mycoplasma, or fungal infection; 4) interventions involving non-drug TCM therapy, such as acupuncture or massage.

Data extraction and quality assessment
A standardized data extraction form including the title, first author's name, publication year, age, sample size, sex, interventions in the treatment and control groups, treatment duration, infection type, pathogen, setting, outcome measures, and reported AEs or ADRs was used.Five methodological quality domains were assessed using Rob 2.0 and included: 1) bias arising from the randomization process; 2) bias due to deviations from intended interventions; 3) bias due to missing outcome data; 4) bias in outcome measurements; 5) bias in the selection of the reported result.Each domain was categorized according to three levels, namely, "high risk of bias," "low risk of bias," or "some concerns".Two reviewers (Y.G. and S.Z.) independently evaluated the methodological quality of each study, and disagreements were resolved by a third reviewer (JQ.).

Statistical analysis
Meta-analysis was performed using Stata 16.0.Binary outcomes were expressed in terms of relative risk (RR) and 95% confidence interval (CI).For continuous variables, the mean difference (MD) and 95% CI were used.When I 2 was <50%, the fixed-model Mantel-Haenszel method was used to evaluate the pooled effect.When I 2 was >50%, a random-effects model was used.A sensitivity analysis was performed to evaluate the stability of the research findings.To explore potential sources of heterogeneity, metaregression and subgroup analyses were performed.Metaregression was used when the number of included studies was greater than 10.Begg's and Egger's tests were used to assess publication bias, and trim-and-fill analyses were performed to confirm the results.To avoid type I and type II errors from the meta-analysis and to obtain more robust results (Pogue and Yusuf, 1998;Shah and Smith, 2019), TSA was conducted using TSA 0.9.5.10 beta software.The quality of evidence for all outcomes was evaluated using the GRADE system.
To investigate likely association rules of core herb combinations, we examined CHC constituent herb data.Preliminary information for data extraction was obtained by analyzing the frequency of each herb prior to association rule analysis.Apriori association rule analysis and plot generation were performed using the "arules" R-package (version 2023.03.0 + 386) to fit data and "arulesViz" to generate the plots (Merdun and Ozturk, 2005;Hahsler, 2017).To detect meaningful connections between variables in large databases, frequent hub item sets and association rules were mined using the Apriori algorithm.This method can identify relationships between projects and can be used in various medical studies designed to predict variable characteristics (Hsieh et al., 2020).
In the Apriori algorithm, the primary metrics used to measure associations include support, confidence, and lift.Support, defined as the proportion of transactions that contain the itemset in the dataset, measures the importance of the itemset and is expressed as P (A∩B).Confidence, which measures the probability of observing herb A in a transaction containing herb B, can be expressed as P(A∩B)/P(A) = P(B|A) but does not explain whether there is a meaningful correlation between the two transactions.Lift, which can be expressed as P (A with B)/P (A) P (B) = P(B|A)/P(B), can identify meaningful correlations between two transactions.Herb A and Herb B are unrelated when lift is close to 1, as they are probabilistically close to independent, whereas lift >1 indicates a strong relationship between Herb A and Herb B. To identify association rules, minimum support and confidence values were set at 20% and 80%, respectively, and core herb combinations demonstrating the most evident associations were identified.

Setting
Of the 38 studies, 22 were conducted in intensive care units (ICUs); 6 were conducted in non-ICU settings, including neurology, rehabilitation medicine, respiratory, and integrated traditional Chinese and Western medicine wards, and 10 RCTs did not provide information on the setting.

Treatment
Regarding treatment groups, only one study adopted CHC monotherapy, whereas the others used combined interventions.CHC therapies were prepared as decoctions or granules.Twenty-three RCTs included patients who received Chinese herbal compound prescriptions orally, while the remaining studies involved patients receiving these prescriptions via nasogastric feeding.All of these patients were from ICUs.Eighteen studies documented antibiotic treatment regimens based on sputum-culture drug-sensitivity test results or recommended guidelines, while the remaining 20 studies provided detailed documentation of the types and dosages of antibiotics administered in treatment.The antibiotics involved in these studies included carbapenems, β-lactam antibiotics, glycopeptides, aminoglycosides, oxazolidinones, and tetracyclines.The most frequently utilized antibiotic was a combination of imipenem and cilastatin sodium.

Risk of bias assessment
The Rob 2.0 tool was used to assess the quality of the included studies.The results of the risk of bias assessment are as follows: 1) Randomization process: All included studies declared randomization, but only 25 studies described the details of the randomization methods (by software or random number table).The allocation concealment approach was illustrated by only two studies.None of the studies exhibited baseline differences between the intervention and control groups, indicating poor randomization.Consequently, two studies were assessed as low-risk, and the rest had some concerns about the risk of bias.2) Deviations from the intended interventions: As only one study adopted a single-blind method, and none of the studies used intention-to-treat (ITT) or modified-ITT analysis to evaluate differences between the intervention and control groups, the risk of bias associated with deviation from the intended interventions was rated as high for most studies.3) Missing outcome data: Two studies reported missing outcome data due to dropouts.The rest of the included RCTs had no incomplete data and were marked as low-risk.4) Measurement of the outcome: All studies were rated as having a low risk of bias, mostly because participants' symptoms and laboratory results were used to assess response rates, which were rarely influenced by their knowledge.5) Selection of the reported result: As none of the included studies had a registered protocol, adherence to pre-planned statistical analysis methods could not be assessed.Consequently, all included studies were assessed as having unclear risks.The overall risk of bias was high.The summary of methodological quality is displayed in Figure 2.

Response rate
Thirty-eight studies involving 2890 patients compared CHC + antibiotic treatment with antibiotic-only treatment and reported the response rate for treating pneumonia caused by MDR or XDR bacteria.The heterogeneity test results revealed very low heterogeneity among studies (X 2 = 38.33;p = 0.41; I 2 = 3.47%), and therefore a fixed-effects model was used for the statistical analysis of the response rate.The meta-analysis revealed a significant response rate for CHC + antibiotic treatment (response rate = 1.24; 95% CI: 1.19-1.28;p < 0.0001).Subgroup analysis according to the average age, treatment duration, sample size, setting, infection type, and pathogen (Table 2, Supplementary Material S4) revealed statistically significant differences between CHC + antibiotic treatment and antibiotic-only treatment in the average age, treatment duration, sample size, and  Frontiers in Pharmacology frontiersin.orgsetting subgroups.In contrast, subgroup analysis of various infection types revealed no statistical difference for patients with CAP.Sensitivity analyses, in which single study was systematically excluded to determine whether one study had a significant impact on the results, indicated similar aggregated statistics and robust results (Figure 4, Supplementary Material S6).
3.5 Secondary outcomes

Microbiological response
Twenty-one studies involving 1542 patients reported the efficacy of CHC + antibiotic versus antibiotic-only treatment for microbiological eradication.The heterogeneity test revealed low heterogeneity among studies (X 2 = 23.97;p = 0.24; I 2 = 16.55%);Therefore, the statistical analysis of the microbiological response was performed using a fixed-effects model.The results indicated that CHC + antibiotic treatment significantly eradicated bacteria compared with antibiotic-only treatment (RR = 1.41; 95% CI: 1.27-1.57;p < 0.0001).Subgroup analysis according to total sample size, setting, and infection type revealed no statistical improvement for the large sample size, non-ICU setting, and CAP subgroups.In contrast, among the other subgroups, CHC + antibiotic treatment exhibited statistically superior bactericidal effects when compared with antibioticonly treatment (Table 3, Supplementary Material S4).MRSA eradication in the CHC + antibiotic group was lower than that in the antibiotic-only group, although the difference was not statistically significant.However, as the number of studies within the MRSA subgroup was limited, further investigation is required.Sensitivity analysis indicated that the pooled effect was robust (Figure 4, Supplementary Material S6).

White blood cell count
Eighteen studies involving 1284 patients reported the WBC count for CHC + antibiotic treatment and antibiotic-only treatment groups.Heterogeneity tests revealed high heterogeneity among the studies (X 2 = 261.52;p < 0.0001; I 2 = 93.49%),and statistical analysis of this outcome was conducted using a randomeffects model.The results indicated superior efficacy of CHC + antibiotics in reducing the WBC count compared with that of the control group (MD: −2.09; 95% CI: −2.65 to −1.53; p < 0.0001).Meta-regression indicated no significant difference in WBC count based on average age (p = 0.055, adjusted R 2 = 26.77%),publication year (p = 0.603, adjusted R 2 = −7.77%),sample size (p = 0.105, adjusted R 2 = 10.52%), or treatment duration (p = 0.933, adjusted R 2 = −10.40%)(Supplementary Material S5).Although study characteristics, such as setting, infection type, and pathogen, may have also contributed to heterogeneity, meta-regression analyses were not possible due to incomplete documentation of these baseline characteristics.Subgroup analysis revealed undiminished heterogeneity within each subgroup, indicating that the heterogeneity could not be explained by setting, infection type, or pathogen.Furthermore, subgroup analysis of various settings revealed no statistical difference between CHC + antibiotics and antibiotics only in the treatment of non-ICU patients (Supplementary Material S4).Sensitivity analysis indicated that the pooled effect was robust (Figure 4, Supplementary Material S6).

Procalcitonin levels
Twenty-three studies involving 1693 patients reported PCT levels for CHC + antibiotic and antibiotic-only groups.The heterogeneity test revealed high heterogeneity among the studies (X 2 = 491.60;p < 0.0001; I 2 = 95.52%), and statistical analysis of this outcome was conducted using a random-effects model.The results revealed a significantly lower PCT level in the CHC + antibiotic group than in the antibiotic-only group (MD = −0.49;95% CI: −0.59 to −0.40; p < 0.0001).Metaregression revealed no significant difference in PCT level based on average age (p = 0.92, adjusted R 2 = −7.22%),publication year (p = 0.485, adjusted R 2 = −4.17%),sample size (p = 0.080, adjusted R 2 = 12.78%), or treatment duration (p = 0.831, adjusted R 2 = −6.49%)(Supplementary Material S5).Subgroup analysis of various pathogens, settings, and infection types indicated that these factors were not the source of heterogeneity.Subgroup analysis of various infection types revealed that, although no statistical difference was seen between the CHC + antibiotic and antibiotic-only groups for patients with VAP, CHC + antibiotics significantly reduced the PCT level in patients with HAP, CAP, or a mixed infection type compared with antibiotics only (Supplementary Material S4).Sensitivity analysis indicated that the pooled effect was robust (Figure 4, Supplementary Material S6).

C-reactive protein levels
Eighteen studies involving 1282 patients reported the CRP levels for CHC + antibiotic and antibiotic-only groups.High heterogeneity among studies was detected (X 2 = 1733.56;p < 0.0001; I 2 = 99.02%), and statistical analysis of this outcome was therefore conducted using a random-effects model.The results revealed significantly lower CRP levels in the CHC + antibiotic group than those in the antibiotic-only group (MD = −11.80;95% CI: −15.22 to −8.39; p < 0.0001).Metaregression revealed no significant difference in CRP levels based on average age (p = 0.958, adjusted R 2 = −8.57%),publication year (p = 0.062, adjusted R 2 = 22.64%), sample size (p = 0.092, adjusted R 2 = 14.66%), or treatment duration (p = 0.355, adjusted R 2 = −0.24%)(Supplementary Material S5).Subgroup analysis indicated that the pathogen, setting, and infection type were not the source of heterogeneity.Subgroup analysis of various pathogens revealed that there was no statistically significant difference between CHC + antibiotic treatment and antibiotic treatment only in patients with pneumonia caused by MDR P. aeruginosa, whereas CHC + antibiotics had a superior outcome for treating pneumonia caused by mixed MDR/XDR bacteria and MDR/XDR A. baumannii than antibiotics alone (Supplementary Material S4).Sensitivity analysis indicated that the pooled effect was robust (Figure 4, Supplementary Material S6).

Clinical pulmonary infection scores
Eleven studies involving 898 patients reported CPISs for CHC + antibiotic and antibiotic-only groups.High heterogeneity (X 2 = 332.03;p < 0.0001; I 2 = 94.66%) was detected among the studies, and statistical analysis of this outcome was conducted using a randomeffects model.The results indicated that CHC + antibiotics could Frontiers in Pharmacology frontiersin.org11 significantly lower the CPIS more than antibiotics alone (MD = −1.97;95% CI: −2.68 to −1.26; p < 0.0001).Metaregression revealed no significant association between CPIS and average age (p = 0.608, adjusted R 2 = −10.20%),publication year (p = 0.504, adjusted R 2 = −7.70%),sample size (p = 0.240, adjusted R 2 = 8.05%), or treatment duration (p = 0.399, adjusted R 2 = −3.90%)(Supplementary Material S5).The source of heterogeneity could not be explained by the pathogen, setting, or infection type.Subgroup analysis based on the pathogen type revealed a superior outcome for CHC + antibiotics in the treatment of pneumonia caused by MDR P. aeruginosa but not for pneumonia caused by mixed MDR/XDR bacteria (Supplementary Material S4).Sensitivity analysis indicated that the pooled effect was robust (Figure 4, Supplementary Material S6).

APACHE-II score
Seven studies involving 424 patients reported APACHE-II scores for CHC + antibiotic and antibiotic-only treatment.High heterogeneity among the studies (X 2 = 16.62;p = 0.011; I 2 = 63.91%) was detected, and statistical analysis of this outcome was conducted using a randomeffects model.The results indicated that the CHC + antibiotic group had lower APACHE-II scores than did the antibiotic-only group (MD = −4.08;95% CI: −5.16 to −3.00; p < 0.0001).Subgroup analysis showed statistical differences in the average age, treatment duration, pathogen, and infection type subgroups.However, as the number of studies within each subgroup was limited, the impact of these characteristics on the APACHE-II score could not be assessed (Supplementary Material S4).Sensitivity analysis indicated that the pooled effect was robust (Figure 4, Supplementary Material S6).

Length of hospitalization
Six studies involving 394 patients reported the length of hospitalization for CHC + antibiotic and antibiotic-only groups.High heterogeneity among the studies (X 2 = 27.26;p < 0.0001; I 2 = 81.66%)was detected, and statistical analysis of this outcome was conducted using a random-effects model.The results revealed a shorter length of hospitalization for the CHC + antibiotic group than for the antibiotic-only group (MD = −4.79;95% CI: −6.18 to −3.40; p < 0.0001).Although subgroup analyses by age, duration of treatment, pathogen, setting, and type of infection were conducted, the limited number of studies within each subgroup prevented an assessment of the impact of these characteristics on the length of hospitalization (Supplementary Material S4).Sensitivity analysis indicated that the pooled effect was robust (Figure 4, Supplementary Material S6).

Safety
Eleven studies involving 859 patients reported AEs, with fewer AEs reported for the CHC + antibiotic group than for the antibiotic-only group.All 11 studies reported gastrointestinal tract symptoms, such as nausea, vomiting, and diarrhea.In one study, four cases of renal dysfunction were reported in the CHC + antibiotic group.Six studies claimed that, during the study, there were no AEs in the CHC + antibiotic group.AEs were not mentioned in the remaining 21 studies.No heterogeneity was detected among the studies (X 2 = 8.79; p = 0.552; I 2 = 0%), and statistical analysis of this outcome was conducted using a fixed-effects model.The results indicated fewer AEs in the CHC + antibiotic group than in the antibiotic-only group; However, the difference was not statistically significant (RR = 0.70; 95% CI: Frontiers in Pharmacology frontiersin.org0.49-1.01;p = 0.058) (Figure 3).Sensitivity analysis indicated that the pooled effect was robust (Figure 4, Supplementary Material S6).

Publication bias
We assessed the impact of publication bias on response rate using Begg's and Egger's tests.The p-value for both Begg's and Egger's tests were <0.0001, suggesting publication bias.However, trim-and-fill analysis performed using the fixed-effects model did not reveal any publication bias.For the microbiological response, the p-values for Begg's and Egger's tests were 0.266 and 0.004, respectively, which were not indicative of publication bias.This was supported by trim-and-fill analysis, which also did not reveal any publication bias (Figure 5, Supplementary Material S7).According to Begg's test, there was no evidence of publication bias for "WBC count," "PCT level," "CRP level," "CIPS score," "APACHE-II score," "length of hospitalization," and "safety."However, the results of Egger's test indicated a potential risk of publication bias for "WBC count," "CRP level," and "CIPS score."

Quality of evidence according to outcome measures
The GRADE method was used to assess the quality of the evidence.When comparing CHC + antibiotics with antibiotics-only, the overall quality of evidence for each outcome was evaluated as moderate to very poor.The high risk of bias, inconsistency across studies, and imprecision of findings contributed to the low quality of evidence.
Supplementary Material S8 provides a summary of the overall quality of evidence for each outcome.

Trial sequential analysis
The false positive and false negative findings in systematic reviews and meta-analyses can be controlled by TSA, which has therefore become an attractive statistical method.Therefore, we used TSA 0.9.5.10 beta software to ensure that the results of our meta-analyses were reliable and conclusive.The required information size in this study was estimated by α = 0.05 (two-sided) and β = 0.20 (power of 80%), based on the O'Brien-Fleming alpha-expenditure function.The RR reduction and event rates in both experimental and control groups were calculated from the mean of the event proportions.

Response rate
The TSA included 38 RCTs and was performed based on a controlled-event proportion of 71.21% and an RR reduction of −23%.The TSA identified the required information size (RIS) as 192.The cumulative Z-curve reached the RIS and crossed the RISadjusted boundary in support of CHC + antibiotic treatment, thus suggesting conclusive and robust results (Figure 6).

Microbiological response
The TSA included 21 RCTs and was performed based on a controlled-event proportion of 34.40% and an RR reduction of −41%.The RIS was 621.However, the number of patients did not reach the RIS, and the cumulative Z-curve crossed the conventional boundary line and RIS-adjusted boundary, indicating that CHC +

FIGURE 3
The forest plot of rate of adverse events.Frontiers in Pharmacology frontiersin.org14 Zhao et al. 10.3389/fphar.2023.1282538antibiotic treatment was effective in eradicating MDR or XDR bacteria in patients with pneumonia (Figure 6).

Apriori algorithm-based association rule analysis
The association rule analysis included 110 herbs from 38 CHC prescriptions.CHC prescriptions are summarized in Table 4, and the distribution frequency of the herbs is summarized in Table 5. Eighteen association rules were detected when the minimum support and confidence values were set at 0.2 and 0.8, respectively.The lift was >1, which was indicative of effective association rules, with a higher degree of lift indicative of a stronger association.The scatter plot reflected the degree of lift, indicated by color depth, of each association rule, ranging from 0.950 to 1.425 (Table 6).The general distribution of the determined association rules was evaluated using a grouping matrix (Figure 7).

FIGURE 6
Trial sequential analysis (TSA) for response rate and microbiological response.(A) TSA of response rate with a control event proportion of 71.21%, type 1 error of 5%, power of 80%, and relative risk reduction of −23%.(B) TSA of microbiological response with a control event proportion of 34.40%, type 1 error of 5%, power of 80%, and relative risk reduction of −41%.Traditional Chinese medicine(TCM) has been used to treat infectious diseases for thousands of years.The classic works of TCM, such as the "Inner Canon of Huangdi," "Treatise on Febrile Disease," and "Febrile Disease Differentiation," record the rich theory behind treating infectious diseases and the effective methods of treatment.Recently, antibioticresistance has emerged as a global threat.The incidence of MDR and XDR bacteria increases annually, limiting the efficacy of antibiotic agents for the treatment of severe refractory infections (Ayukekbong et al., 2017).Accumulating evidence from clinical studies indicates that patients with pneumonia caused by MDR or XDR bacteria may achieve better clinical outcomes using CHC.

Summary of the main finding
In this study, 38 RCTs were analyzed, of which 37 compared CHC + antibiotics with the same antibiotics alone, and one study compared CHC alone with antibiotics.We found that the combination of CHC and antibiotics significantly increased the response rate.Sensitivity analysis demonstrated that the results were robust, and there was no evidence of publication bias as assessed by trim-and-fill analysis.TSA further explained that the clinical evidence of the efficacy rate of CHC combination therapy for pneumonia caused by MDR or XDR bacteria was determinate; thus, the analytic result is worth asserting.CHC + antibiotic combination treatment was more effective than antibiotics alone in improving the microbiological eradication rate.Subgroup analysis revealed no statistically significant difference in the CAP subgroups.TSA demonstrated that the results were conclusive and robust.
For inflammatory indicators, the combination treatment led to significant improvements in WBC, CRP, and PCT.We identified high heterogeneity in WBC, PCT, and CRP levels; however, the source of heterogeneity remains unclear.We speculated that large individual differences in WBC, PCT, and CRP levels may account for the high heterogeneity, leading to high variation in baseline levels, as well as the degree of improvement.In addition, inadequate study protocols, such as the lack of blinding and allocation concealment, may have contributed to heterogeneity.
In terms of score systems, we also found the CHC combination therapy group had lower CPIS and APACHE-II scores after treatment than the antibiotics-alone group.The CPIS and APACHE-II score were recommended to predict the clinical outcomes of HAP/VAP by recent guidelines (Torres et al., 2017;Martin-Loeches et al., 2018); Lower scores are associated with better outcomes (Zhou et al., 2015).According to our findings, combining CHC and antibiotics could eradicate MDR/XDR bacteria, reduce inflammatory indicator levels, and lower the CPIS and APACHE-II scores, resulting in better clinical outcomes for patients with pneumonia caused by MDR/XDR bacteria.Our results also indicated that patients receiving CHC combination therapy had a shorter length of hospitalization than those receiving antibiotic therapy alone.

Implications of mechanism research
Evidence from this study suggests that the concomitant use of CHC may be more effective than antibiotics alone in the treatment of pneumonia caused by MDR or XDR bacteria.During the antibiotic resistance crisis, many researchers shifted their attention to TCM and investigated the inhibitory effect of TCM herbs or bioactive phytocompounds on antibiotic resistance (Liu et al., 2020;Mohamed et al., 2020;Lan et al., 2021).As the main source of TCM herbs, natural products (NP) are characterized by more complex phytochemicals and spatial configurations than synthetic drugs (Koehn and Carter, 2005).Different extracts from the same herbs vary wildly, which may lead to distinct antimicrobial activity against various antibioticresistant bacteria.In one study, the high antimicrobial activity of L. ethanol extracts from Hypericum perforatum Linn against Lactobacillus plantarum (L.plantarum), Enterococcus faecalis (E.faecalis), Streptococcus mutans (S. mutans), and Streptococcus sobrinus (S. sobrinus)strains was observed.While S. sobrinus and L. plantarum were highly inhibited by their water extracts, a moderate antibacterial effect was observed in S. mutans and E. faecalis (Süntar et al., 2015).Antibioticresistance mechanisms of TCM herbs may involve the efflux pump system, enzyme activity, inhibition of biofilm (Tsai et al.,  2018), drug target changes, permeability of the bacterial cell membrane (Zhao et al., 2019), and other drug-resistant mechanisms (Anand et al., 2020).Baicalin, a bioactive compound derived from S. baicalensis Georgi, could hinder the formation of P. aeruginosa biofilms by suppressing QSrelated genes, including lasI, lasR, rhlI, rhlR, pqsR, and pqsA (Luo et al., 2017).Moreover, in vivo experiments demonstrated the bioactive anti-inflammatory effects of this compound in an acute pneumonia rat model induced by MDR P. aeruginosa (Lei et al., 2023).Other studie demonstrated that lonicerin inhibited biofilm formation and subsequently ameliorated P. aeruginosa infections in A549 cells by directly inhibiting AlgE activity without affecting bacterial viability or AlgE expression (Xu et al., 2019).Licorice extract was found to stimulate excessive intracellular reactive oxygen species production, inducing oxidative stress and generating a significant number of free radicals.This disrupted the cell wall membrane structure and functionality of Staphylococcus aureus (S. aureus).Additionally,  et al., 2023).In one study, under the intervention of 1/4 MIC berberine and 1/8 MIC imipenem (Su and Wang, 2018), the MexXY-OprM efflux pump in P. aeruginosa was blocked due to the downregulation of MexZ, MexX, MexY, and the outer membrane protein OprM.Chelerythrine, isolated from the root of Toddalia asiatica (Linn) Lam, has been reported to damage the bacterial cell membrane and destroy channels across cell membranes, allowing protein leakage from the cell and suppressing protein biosynthesis.Therefore, it has a strong inhibitory effect on S. aureus and MRSA and on the betalactamases produced by these bacteria (He et al., 2018).Additionally, the synergistic effect of TCM herbs can enhance the sensitivity of antibiotic-resistant bacteria to antibiotics, leading to their use as sensitizers.For example, there is no significant difference in the antibacterial effect of pterostilbene and gentamycin alone.However, when combined, the two compounds completely inhibited bacterial growth and exhibited synergistic antibacterial effects (Lee et al., 2017).
Research reported that the antimicrobial effect of mupirocin against strains of S. aureus and MRSA was enhanced when combined with piperine isolated from black pepper through the inhibition of ethidium bromide efflux (Mirza et al., 2011).

Limitations
Our study had some limitations.First, the randomization process of most studies was unclear, and blinding and allocation concealment was often lacking, which introduced a high risk of bias and lowered the methodological quality.Furthermore, none of these studies were registered; therefore, a study protocol was not obtained.Second, the sample sizes of all included studies were small, and therefore, the findings of these studies may not be generalizable.Third, some studies lacked comprehensive reports regarding research characteristics, such as pathogens, setting, and infection type, which limited the analyses of heterogeneity and exploration of the dominant population.Furthermore, the included studies lacked long-term outcome evaluation indicators for CHC + antibiotic treatment in drug-resistant bacterial pneumonia, such as potential relapse rates.Fourth, although we utilized meta-regression and subgroup analysis to investigate the sources of heterogeneity in this article, we were unable to identify them.Therefore, in future research, we aim to establish comprehensive protocols that yield more stable and reliable results.Fifth, all participants included in this study were exclusively sourced from China limiting the generalizability of the findings to other populations.We look forward to the emergence of future large-sample RCTs conducted in various regions to enhance the external validity and reliability of research findings.

Conclusion
Patients with pneumonia caused by MDR and XDR bacteria could benefit from CHC + antibiotic combination therapy, which improved response and microbiological eradication rates, reduced the inflammatory response, reduced the CPIS and APACHE-II score, and shortened the length of hospitalization.Scutellaria baicalensis Georgi [Labiatae], F. thunbergii Miq [Liliaceae], G. uralensis Fisch [Fabaceae], and L. japonica Thunb [Caprifoliaceae] were identified as core herbs for the treatment of antibiotic-resistant bacterial pneumonia.

FIGURE 1
FIGURE 1Flow chart of study selection and identification.

FIGURE 2
FIGURE 2 Risk of bias assessment for included studies: (A) Risk of bias graph.(B) Risk of bias summary.

FIGURE 5
FIGURE 5Trim-and-fill analysis for (A) Response rate, (B) Microbiological response.

FIGURE 8
FIGURE 8Network graph of the association rules in the meta-analysis of Chinese herbal compound prescribed for pneumonia caused by MDR or XDR bacteria.
Thirty-eight RCTs conducted in China between 2006 and 2022, involving 2890 patients, with 1431 patients in the intervention group and 1459 patients in the comparison group, were included.The median age of the patients was 59.55 years.Most of the studies included middle-aged or elderly adults, and only one study recruited toddlers.Table1summarizes the characteristics of the 38 included studies.

TABLE 1
Characteristics of the included studies.

TABLE 1 (
Continued) Characteristics of the included studies.

TABLE 1 (
Continued) Characteristics of the included studies.

TABLE 1 (
Continued) Characteristics of the included studies.

TABLE 1 (
Continued) Characteristics of the included studies.

TABLE 2
Subgroup analysis of response rate according to average age, treatment duration, sample size, setting, infection type, and pathogen.

TABLE 3
Subgroup analysis of microbiological response according to average age, treatment duration, sample size, setting, infection type, and pathogen.

TABLE 4 (
Continued) CHC ingredients used in the included studies.

TABLE 4 (
Continued) CHC ingredients used in the included studies.

TABLE 4 (
Continued) CHC ingredients used in the included studies.

TABLE 4 (
Continued) CHC ingredients used in the included studies.

TABLE 4 (
Continued) CHC ingredients used in the included studies.

TABLE 5
Top 10 herbs frequently prescribed for pneumonia caused by MDR or XDR bacteria.

TABLE 6
Apriori algorithm-based association rules in the meta-analysis of CHC prescribed for pneumonia caused by MDR or XDR bacteria.
(Singh et al., 2015; aureus with licorice extract resulted in a notable reduction in the enzymatic activity of two critical cellular metabolic enzymes: succinate dehydrogenase and succinate dehydrogenase ATPase, thereby interfering with the organism's energy metabolism system(Singh et al., 2015; Zhang