Comparative Efficacy of Chinese Herbal Injections for the Treatment of Herpangina: A Bayesian Network Meta-Analysis of Randomized Controlled Trials

Background Considering the limitations of broad-spectrum antiviral drugs for the treatment of herpangina and the extensive exploration of Chinese herbal injections (CHIs), systematic evaluation of the efficacy of different CHIs in the treatment of herpangina is a key imperative. In this study, we performed a network meta-analysis to investigate the efficacy of CHIs, including Reduning injection (RDN), Shuanghuanglian injection (SHL), Tanreqing injection (TRQ), Xiyanping injection (XYP), and Yanhuning injection (YHN), in the treatment of herpangina. Methods A systematic literature review including studies published before December 17, 2018, was conducted in several databases. The quality of the included studies was assessed using the Cochrane risk of bias tool. Data were analyzed using STATA 13.0 and WinBUGS 1.4.3 software. Surface under the cumulative ranking curve (SUCRA) probability values were applied to rank the examined treatments. Clustering analysis was performed to compare the effects of CHIs between two different outcomes. Results A total of 72 eligible randomized controlled trials involving 8,592 patients and five CHIs were included. All patients were under the age of 15 years, and most were under 7 years. The results of the network meta-analysis showed that RDN, XYP, and YHN had significantly better treatment performance than ribavirin. SHL (OR: 0.18; 95% CI: 0.09–0.34) and TRQ (OR: 0.18; 95% CI: 0.10–0.31) were obviously superior to ribavirin with respect to total clinical effectiveness. The results of SUCRA and cluster analysis indicated that RDN is the best intervention with respect to total clinical effectiveness, antipyretic time, and blebs disappearing time. Fifty-four studies described adverse drug reactions/adverse drug events (ADRs/ADEs), and 32 studies reported ADRs/ADEs in detail. Conclusions CHIs were found to be superior to ribavirin in terms of treatment performance and may be beneficial for patients with herpangina. RDN had the potential to be the best CHI with respect to all outcome measures. More evidence is needed to assess the safety aspects of CHIs.


INTRODUCTION
Herpangina is a common pediatric disease that is mainly caused by Coxsackie A virus; respiratory and fecal-oral routes are the main routes of transmission. Coxsackie A virus is a small RNA virus that is present in the intestines. The virus exhibits rapid transmission, especially in summer and early autumn. Children in the age group of 1-7 years are particularly vulnerable to infection (Jiang et al., 2015;Huang, 2016). Children infected with herpangina can manifest sore throat, excessive salivation, fever, oral herpes, anorexia, and other symptoms. Enteroviruses are also known to cause serious diseases such as myocardial damage or myocarditis (Wu, 2018;Guo and Li, 2019). Currently, there is no specific treatment for herpangina. Antiviral drugs, symptomatic supportive care, and prevention of complications are the mainstays of treatment (Guo and Li, 2019). Ribavirin is a broad-spectrum antiviral drug that is commonly used for the treatment of herpangina. However, the mechanism of action of ribavirin is highly dependent on viral adenosine kinase; this results in a high probability of the development of drug resistance, which in turn affects the therapeutic effect (Li and Zhan, 2017;Liu et al., 2019). Several recent studies have documented the efficacy of Chinese herbal injections (CHIs) in the treatment of herpangina (Zhu et al., 2014;Xu et al., 2016;. However, several varieties of CHIs have been used to treat herpangina, and further research is required to identify the best type of CHI for this purpose. Therefore, in this study, we used the network meta-analysis (NMA) method to systematically evaluate the efficacy of different CHIs in the treatment of herpangina. The objective was to identify an optimal intervention measure and provide a basis for clinical drug use.

METHODS
This study is reported in strict accordance with the standard format of the Preferred Reporting Items for Systematic Reviews and Meta-Analysis Specification: PRISMA Extension Statement specification (Hutton et al., 2015;Ge et al., 2017).

Search Strategy
PubMed, the Cochrane Library, Embase, the Chinese Biological Medicine Literature Service System (SinoMed), the China National Knowledge Infrastructure (CNKI) database, the Chinese Scientific Journal Database (VIP), and the Wanfang Database were searched for randomized controlled trials (RCTs) of CHIs for the treatment of herpangina. Studies published as of December 17, 2018 were eligible for inclusion. In addition, the reference lists of the included studies were manually searched to identify relevant literature. There were three parts of the search strategy, including herpangina, Chinese herbal injection, and random controlled trial. A total of 132 types of CHIs incorporating national standards of the Chinese Food and Drug Administration and 36 kinds of Chinese medicinederived chemical injections were included in the prescreening.

Types of Studies
RCTs of CHIs for the treatment of herpangina were eligible if they were referred to as "random," with or without blinding.

Types of Participants
All patients included were clinically diagnosed with herpangina according to clear diagnostic criteria, with no limitations of sex, race, or age.

Types of Interventions
The interventions included were comparisons between CHIs and ribavirin or between different types of CHIs. Ribavirin and CHIs were administered intravenously; in addition, according to the patient's condition, certain symptomatic supportive treatments were adopted (e.g., cooling, rehydration, maintenance of water and electrolyte balance, and antibiotic therapy for concurrent bacterial infection). No limitations were imposed with respect to the dosage or treatment course. No other Chinese medicine or remedies were used, such as decoction, proprietary Chinese medicine, acupuncture, or massage.

Types of Outcomes
Outcome indicators included total clinical effectiveness, antipyretic time, blebs disappearing time, and adverse reactions (ADRs)/adverse events (ADEs). Total clinical effectiveness = (total number of patients-;number of patients in whom treatment was ineffective)/total number of patients×100%. The evaluation criteria for efficacy were based on the posttreatment recovery of clinical symptoms and signs; ineffective treatment implies deterioration or no change in symptoms and signs after the treatment course.

Data Extraction and Quality Assessment
All retrieved studies were managed using NoteExpress software. After excluding duplicates, two researchers independently screened the retrieved studies based on the inclusion and exclusion criteria and extracted the data from the included RCTs. The titles and abstracts of retrieved studies were screened to exclude animal studies, literature reviews, and other unrelated articles. Subsequently, studies that met the inclusion criteria were identified, and their full texts were reviewed. A specially designed form (created using Microsoft Excel 2016 software) was used to extract data pertaining to the following information from the included studies: (1) name of first author and the year of publication; (2) basic characteristics of patients: the numbers of patients in the treatment group and the control group, sex distribution, average age or age range, interventions, and treatment details; (3) outcome measures;     and (4) study types and main factors affecting the risk of bias. Any disagreement between two researchers during the screening of studies and extraction of data was resolved by consensus or by consulting a third researcher. Two authors independently assessed the risk of bias in the included studies in accordance with the risk of bias assessment tool recommended in the Cochrane Handbook 5.1 (Higgins and Green, 2010). The following elements were assessed: (1) selection bias associated with random sequence generation; (2) selection bias associated with allocation concealment; (3) performance bias: blinding of participants and personnel; (4) detection bias: blinding of outcome assessment; (5) attrition bias: integrity of outcome data; (6) reporting bias: selective reporting; and (7) bias from other sources. Each element was categorized as "low risk," "high risk," or "unclear." "Low risk" implies that the implementation method is correct or does not affect the result; "high risk" implies that the implementation method is incorrect and affects the measurement of the result; "unclear" means that the information is insufficient, and the risk of bias cannot be judged. Consensus was attained by discussion or involving a third researcher.

Data Analysis
WinBUGS 1.4.3 software was used to perform NMA, and the Markov chain Monte Carlo method with random-effects model was performed for Bayesian inference. In the WinBUGS software, the number of iterations was set as 200,000, with the first 10,000 iterations used for burn-in to eliminate the impact of the initial value. On NMA, the odds ratio (OR) and 95% confidence intervals (95% CI) were calculated for the binary outcomes; the mean difference (MD) and 95% CI were calculated for continuous outcomes. When the 95% CI for the OR value did not contain 1 and the 95% CI for MD value did not contain 0, the difference between groups was deemed to be statistically significant. Stata 13.0 software was used to map the network of different interventions for each outcome measure, showing the results of the direct and indirect comparison of CHIs. When using the results of WinBUGS software with Stata software, the surface under the cumulative ranking probability (SUCRA) of different CHIs in each outcome index was obtained. The larger the SUCRA and the higher the ranking, the greater the probability that the CHI is the best intervention. A comparison-adjusted funnel plot was used to assess potential publication bias. If points on both sides of the midline in the funnel diagram were symmetric, which meant the correction guideline was at right angles to the midline, it was considered indicative of no significant publication bias. The cluster analysis method was used to comprehensively analyze and compare interventions for two different outcome indicators; then, the optimal injection variety for the two outcome indicators was obtained. The farther from the origin in the cluster map, the better the effect is in these two outcome indicators. If there was a closed loop, the inconsistency test was used to evaluate the consistency of each closed loop, and the inconsistency factors (IFs) and 95% CI were calculated. When the 95% CI contained 0, the consistency was good; otherwise, the closed loop was considered to exhibit significant inconsistency.

Search Results
Out of the 1,123 retrieved articles, 72 RCTs (shown in Table 1) were selected and included in the NMA. Further details of the literature screening process are presented in Figure 1.

Inclusion Studies and Characteristics
The 72 Figure 2. All treatment courses lasted < 7 days. The details of the included studies are shown in Table 1.

Methodological Quality
Of the 72 included studies, 12 RCTs used a random number table for group allocation, while two RCTs used a random sampling method. The selection bias associated with random sequence generation of the above studies was evaluated as "low risk." All studies reported complete data, and their attrition bias was evaluated as "low risk." One RCT did not indicate whether the  baseline characteristics of the two groups were comparable at the time of grouping, which may have impacted the results, and other corresponding biases were evaluated as "high risk." The risk of bias entries for the remaining studies was rated as "unclear" due to insufficient information. The results of the risk of bias evaluation are shown in Figure 3.

Total Clinical Effectiveness
Sixty-nine RCTs reported the total clinical effectiveness, involving five CHIs and six interventions. The network graph is shown in Figure 2. The OR value of the NMA is shown in Table 2. Compared with ribavirin treatment, RDN, SHL, TRQ, XYP, and YHN were found to have greater total clinical effectiveness in patients with herpangina; the between-group differences were statistically significant. There were no significant differences between the remaining intervention groups.
The SUCRA ordering and probability value results (Figure 4, Table 3) indicate that RDN is the most likely to improve total clinical effectiveness in herpangina patients compared with ribavirin, followed by SHL and TRQ.

Antipyretic Time
Forty-five RCTs reported antipyretic time, involving five kinds of CHIs and six interventions. The network diagram is shown in Figure 2. The results of NMA ( Table 2) showed that RDN, XYP, and YHN can shorten the antipyretic time compared with ribavirin; between-group differences in this respect were statistically significant. The difference between the remaining interventions was not statistically significant. The SUCRA ordering and probability value results (Figure 4,

Blebs Disappearing Time
Thirty-eight RCTs reported the blebs disappearing time; these involved four interventions (RDN, XYP, YHN, and ribavirin). The network diagram is shown in Figure 2. On NMA (Table 2), RDN, XYP, and YHN were found to be associated with a shorter blebs disappearing time compared with ribavirin; the betweengroup difference in this respect was statistically significant. No significant between-group differences were observed for other interventions. The SUCRA ordering and probability value results ( Figure 4, Table 3) indicated that RDN has the best treatment effect, followed by XYP and YHN.

Cluster Analysis
The cluster analysis method allowed for a comprehensive comparison of the effects of different interventions on total clinical effectiveness, antipyretic time, and blebs disappearing time. The results showed ( Figure 5) that RDN was the best intervention in terms of total clinical effectiveness and antipyretic time, total clinical effectiveness and blebs disappearing time; these findings suggest that the efficacy of RDN in the treatment of herpangina is worthy of attention.  The warmer the color, the greater the SUCRA, and the greater the probability of becoming the best intervention. Figure 6 shows the comparison-correction funnel plot for total clinical effectiveness to assess potential publication bias. The points on both sides of the centerline of the funnel plot are not completely symmetrical, and there is a large angle between the correction guideline and the centerline. This suggests that our results may have been affected by publication bias to some extent.

Consistency Test
To evaluate the consistency of each closed loop, the IF and its 95% CI were calculated using Stata software. When the 95% CI contained 0, it was considered to be consistent; otherwise, there was a significant inconsistency in the closed loop. For example, an inconsistency plot of total clinical effectiveness is shown in Figure 7. The inconsistency test results showed the inclusion of five rings, and only the 95% CI of 1 ring did not contain 0; this indicates that there was a small inconsistency in the included studies and that the results were relatively reliable.

Adverse Drug Reactions/Adverse Drug Events
Of the 72 included studies, 18 (25.00%) did not monitor ADRs/ ADEs during treatment. Out of the 54 (75.00%) studies that described ADRs/ADEs, 22 studies recorded no ADRs/ADEs, while 32 studies reported the occurrence and the number of affected patients in detail. The total number of patients who experienced ADRs/ADEs was 6,647, which accounted for 77.36% of the total patients. No ADRs/ADEs on TRQ were   Table 4.

DISCUSSION
In this study, we evaluated the use of five types of commonly used CHIs (RDN, SHL, TRQ, XYP, YHN) and ribavirin for the treatment of herpangina. The efficacy of the CHIs was systematically evaluated based on the results of 72 included studies and three outcomes. The results of NMA indicated that the efficacy of RDN, XYP, and YHN was better than that of ribavirin with respect to all outcome measures. With respect to total clinical effectiveness, the efficacy of SHL and TRQ was better than that of ribavirin, and the between-group difference was statistically significant. From the results of SUCRA ordering, among the three outcome indicators, RDN ranked as the best intervention, while all CHIs showed better efficacy than ribavirin. On cluster analysis, RDN was found to be the best intervention with respect to all three outcome measures. Our results highlight the efficacy of RDN in the treatment of herpangina. However, the effect of publication bias on our results cannot be ruled out; therefore, treatment decision-making in individual cases should be guided by specific situations and the experience of clinicians.  In terms of safety, 75% of the included studies monitored ADRs/ ADEs. Compared with the medication monitoring of other common respiratory diseases, the RCTs included in this study were better with regard to monitoring the safety of drug use. Among the patients monitored, no significant ADRs occurred in patients treated with TRQ; therefore, its safety needs to be further confirmed by observational studies. In the reported ADRs/ADEs, except for one case of dyspnea and mild chest pain in the ribavirin group, no serious cases occurred in the other groups. The most frequently reported ADRs/ADEs of CHIs were gastrointestinal reactions, followed by rash and leukopenia. Leukopenia occurred primarily in the ribavirin group. The incidence of ADRs was most common in the YHN group, followed by the ribavirin group; the XYP group had the lowest incidence of ADRs/ADEs. Therefore, due care should be taken to avoid ADRs, especially when using YHN and ribavirin. This is the first study that used the NMA method to evaluate the efficacy and safety of CHIs in the treatment of herpangina and ranked the results of clinical total effectiveness and the disappearing time of two main clinical symptoms. The objective was to provide evidence and recommendations for the clinical selection of drugs. However, some limitations of this study should be considered when interpreting our results: (1) The methodological quality of the included studies was not very high. Only 14 of the 72 RCTs described the correct generation of random sequences. None of the studies mentioned allocation concealment and blinding, and one study did not describe whether the two groups had comparable baseline characteristics. (2) All the included studies were published in Chinese journals; therefore, the findings may not be entirely generalizable to other settings. (3) Most of the included RCTs compared CHIs versus ribavirin, and there was a lack of a more direct comparison of two or more CHIs.
(4) This meta-analysis has not been registered online.
Based on the above limitations, we make the following recommendations: (1) For future clinical RCTs, the registration of the protocol should be carried out in advance, and the study should strictly adhere to the protocol to ensure transparency of the implementation process and avoid selective reporting.
(2) Future studies should use robust methods for random sequence generation (such as the use of a random number table), implement allocation concealment (e.g., with the use of opaque envelopes), and implement strict blinding to ensure the reliability of the results.
(3) More studies should be conducted to evaluate the efficacy of CHIs.

CONCLUSION
In conclusion, the use of CHIs was associated with improved treatment performance and could be beneficial for patients with herpangina compared to ribavirin. RDN showed the best efficacy with respect to all three outcome measures. However, more direct comparison studies of two or more CHIs are needed to further confirm the results. Future studies should include meticulous monitoring of the safety of CHIs.

AUTHOR CONTRIBUTIONS
JW and XD done conception and design of the network metaanalysis. XD, HW and KW performed the network metaanalysis. XD, WZ and XL assessed the quality of the network meta-analysis. XD, HW and KW analyzed study data. XD and HW wrote the paper. All authors read and approved the final version of the manuscript.