This study was conducted following the criteria of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA-NMA)[6]. Accordingly, a systematic search of databases, organization of documents for review, selection of studies following the criteria defined by the authors, information extraction, analysis, and finally, the presentation of the final report was performed.
Data resources and search strategies
To find related articles, a systematic search of articles in Persian and English databases, including SID, Magiran, the Cochrane Central Register of Controlled Trials (CENTRAL), PubMed, EMBASE, CINAHL, and AMED was carried out without a time limit until March 10, 2022. Also, hand searches were performed in other resources in the reference list of the identified manuscript and Google scholar. To find the appropriate keywords, the preliminary published studies, Medical Subject Headings (MESH Terms) in the PubMed database, as well as the careful examination of the questions of this study, were selected according to PICO criteria [7]. PICO criteria included:
Participants: All women with menopause without age restrictions were examined.
Intervention: Since this study aimed to select the best and most effective herbal medicine to improve hot flashing, the articles selected herbal medicines.
Control: Evaluating the effects of each treatment group compared to the placebo.
Outcome: outcome measures included hot flush frequency and severity.
Eligible primary studies were identified in English electronic databases using the search strings containing four English keywords, and their Persian equivalents were used in Persian databases. These keywords included Menopause, vasomotor symptoms, Herbal medicine, and hot flashes. The Boolean search method is used to combine keywords. Sample search strategies are shown in Table1.
Table1 search strategy keywords
population
|
outcome
|
Intervention and control
|
study
|
("menopause"[TIAB] NOT Medline[SB])
|
vasomotor disorder[All Fields]
|
herbal remed
|
randomized controlled trial
|
"menopause"[MeSH Terms]
|
vasomotor disorders[All Fields]
|
herb$ medic
|
controlled clinical trial
|
menopausal[Text Word]
|
"climacteric"[MeSH Terms]
|
placebo
|
randomized
|
menopause[Text Word]
|
climacteric[Text Word]
|
control
|
clinical trials as topic
|
"postmenopause"[MeSH Terms]
|
("hot flashes"[MeSH Terms]
|
|
randomly
|
postmenopause[Text Word]
|
hot flashes[Text Word
|
|
trial
|
"premature ovarian failure"[Text Word]
|
("flushing"[MeSH Terms] OR flushing[Text Word])
|
|
|
"ovarian failure, premature"[MeSH Terms]
|
"night sweats"[All Fields]
|
|
|
ovarian failure[Text Word]
|
flushing"[MeSH Terms] OR flushing[Text Word]
|
|
|
Inclusion and Exclusion Criteria
Inclusion criteria: 1.RCT studies, 2. studies that examined the effect of Herbal remedies on hot flashes
Exclusion criteria: 1. Case-control studies 2. Case reports 3. Letter to editor 5.Cross-over studies 4. Studies whose full text is not available 5. Unrelated studies 6. Studies with insufficient data 7. Duplication studies 8. A systematic review and meta-analysis studies.
Data Extraction
For each chosen study, socio-demographic and clinical variables were extracted. The risk of bias was assessed through a checklist resulting from the integration of quality assessment tools for quantitative studies[8], and two authors (Shahsavari S and Keshavarzi F) independently evaluated the risk of bias by the Cochrane Collaboration Tool [9]. Since the weight of the conclusion from the meta-analysis depends mainly on the validity of the findings of individual studies, it is essential to evaluate the quality of the study (attachment table1). The tool used in the present study assesses the potential areas for the following bias that:
1. Bias in selecting to evaluate recruitment and randomization methods.
2. Blinding Outcome Assessment examining the knowledge of outcome assessors about participants' intervention or exposure situation
3. Incomplete data as a result of random cancellation and deletion assessment. This item assesses whether the risk of accidental removal is related to treatment.
4. Assessing significant differences between groups before intervening in confounding variables
5. Data collection methods evaluate the validity and reliability of instruments
Search results were uploaded to Excel2013, and duplicates were removed. Two independent researchers evaluated the suitability of the studies. The first screening related to titles and abstracts, and the final screening related to the full text. The first and second authors independently screened and rated each study, and disagreements were resolved through consensus discussion. The final score determined whether the study was at low, moderate, or high risk of bias. The longest follow-up period was considered for an article with several follow-up periods. The mean frequency and severity of hot flashes in each study before and after the intervention was extracted for both experimental and control groups.
Measures of Treatment Effect
All responses (frequency and severity of hot flash) were continuous. We calculated mean and STD deviance for responses in each group and presented 95% confidence intervals (CI) for the mean difference between treatment groups.
Assessment of Network Assumed
To obtain valid NMA results and easier interpretation, it is assumed that network transitivity (potential modifiers of treatment effects are distributed similarly across trials), network consistency (estimates of indirect effects are consistent with direct effects), and homogeneity (interpretation of treatment effects should be homogeneous throughout the trial) should be established.
Statistical Analysis
A Bayesian network meta-analysis was carried out using “BUGSnet” (Bayesian inference Using Gibbs Sampling to conduct a Network meta-analysis) package in R software. BUGSnet requires the user to have installed Just Another Gibbs Sampler (JAGS) on their computer[10]. In this package, for meta-analysis network analysis such as paired meta-analysis analysis, the first step is to define and process the data in such a way that the names of studies (ID) and treatment arms in them are defined as variable names and match each other using the function data.prep(). Net.plot() and net.tab() functions can be used to draw meta-analysis networks in graphical and tabular format[11]. Once identified, one can use the data.plot() function within BUGSnet to assess the heterogeneity of these modifiers within an evidence network. To examine inconsistency at the global level, the fit of the inconsistency model can be compared against a model in which consistency is assumed using the nma.fit() function and comparing the DICs. Local inconsistency can be explored on the leverage plots produced by nma.fit() and using the nma.compare() function, which produces a plot comparing the posterior mean deviance for the inconsistency and the consistency model. We chose to implement the inconsistency of the model in BUGSnet because it easily handles different network structures and multi arm trials, which is not the case with other methods such as the Bucher method [12]. Because the outcomes were continuous and multi-arm trials were included in the analysis, a normal probability model for multi-arm trials was used, and random effects network meta-analyses were performed. We conducted an NMA on the dataset by fitting a generalized linear model with an identity link function and normal likelihood function to account for the continuous outcome specified through the nma.model(). To be consistent with the NICE-DSU technical support document, we specified a burn-in of 50,000 iterations followed by 100,000 iterations with 10,000 adaptations in the nma.run() function. We can also graphically display the probability of the ranking of each treatment within a surface under the cumulative ranking curve (SUCRA) plot.