2.3. Data collection and analysis

Two reviewers (SK and SC) trained in the process and purpose of selection will independently review the title, abstract, and manuscript of all the studies to check for their eligibility to be included in the analyses. All studies searched by both electronic and hand methods will be uploaded to Endnote X8 (Clarivate Analytics, Philadelphia). The reason for excluding the study will be recorded and shown in PRISMA flow chart (Fig. ​(Fig.1).1). All disagreements will be resolved by a consensus and discussion between the 2 reviewers. If the agreement is not successful between the 2 reviewers, the arbiter (JK) will intervene and resolve the problem.

Flow chart of the search process.

Two review writers (KK and NH) will independently extract the data and fill the standard data extraction form, which will include the study information such as the first author, publication year, research design, interventions, treatment period, and outcomes. All differences between the 2 reviewers will be resolved by a consensus and discussion; however, if necessary, the arbiter (JK) will intervene to reach an agreement.

Two reviewers (KK and NH) will assess the risk of bias (RoB) based on the Cochrane collaboration's tool, which includes random sequence generation (selection bias), allocation concealment (selection bias), blinding of participants and personnel (performance bias), blinding of outcome assessment (detection bias), incomplete outcome data (attrition bias), selective reporting (reporting bias), and other bias. Assessment result will be shown through 1 of the 3 categories: low, unclear, and high. All disagreements will be resolved by a consensus and discussion between the 2 reviewers, and if necessary, the arbiter (JK) will intervene.

Dichotomous data will be evaluated through the relative risk with 95% confidence intervals (CIs), and continuous data will be evaluated by the mean difference with 95% CIs.

To avoid carry-over effects, we will use only the 1st-phase data in case of the randomized cross-over trials. If the trials have multiple intervention groups, a pair-wise comparison will be made.

In case of missing data or difficulty in finding enough data, we will try to contact the original investigators via e-mail to obtain sufficient data. Despite these attempts, if we are unable to obtain the data, it will be sought from the original source or trial reports. Statistical analysis will be performed based on the intent-to-treat principle.

Random effects model will be used for meta-analysis. We will use I² statistics and χ² test to assess the heterogeneity. I² ≥ 50% and P value <.10 will indicate substantial heterogeneity.

If the analysis includes more than 10 studies, a funnel plot will be generated to evaluate the publication bias or small-study effects.

We will use the review manager program (V5.3.5 Copenhagen: The Nordic Cochrane Centre, The Cochrane Collaboration, 2014) to perform the statistical analyses. All studies will be synthesized according to the type of intervention and/or control as follows: BST vs no treatment, BST vs placebo control, BST vs conventional Western medicine, and BST–Western medicine combined therapy vs conventional Western medicine alone. The herbs-added BST will be included in the BST group as described in the “Types of intervention” section.

In case of availability of enough subgroup studies to investigate the cause of heterogeneity, subgroup analysis will be performed. Its criteria will include pattern identification in Traditional Chinese Medicine, physical form of BST, number and type of added herbs, and treatment duration. If the quality of the study is judged to be low after the subgroup analysis, these studies would be removed to confirm the robustness of the results.

We will use “the consolidated standards of reporting trials extension for herbal interventions” to evaluate the methodological and reporting quality of the studies, and the sensitivity analysis will be performed to evaluate the robustness of the results obtained from the meta-analysis.

We will use “The Grading of Recommendations Assessment, Development and Evaluation” to examine the quality of evidence.