2.3. Selection of data extraction
Eligible studies were screened in their entirety and developed a data
extraction form, and the information including authors and year of
publication, publication type, sample size, gender, type of treatment
was recorded in an Excel spreadsheet. We pre-tested it on five studies
and subsequently adapted the final version. If the inclusion criteria
were met, the full-text of each study was coded by the first authors
(J.Y. and H.J) using this template, difficulties in deciding the
inclusion status of studies were discussed between two authors and
resolved by consensus.
If
there was any evidence for the use of the same sample in different
publications, authors were contacted for
clarification25. If it was confirmed that two studies
were based on the same data, we chose the study that reported the most
comprehensive results. Study authors were also contacted to request
additional information if a study was eligible based on the inclusion
criteria, but not all required data could be retrieved from the
full-text. Overall, we adopted a conservative procedure in that we
excluded studies for which ambiguity about the primary data source could
not be resolved or if the information necessary for effect size
calculation could not be acquired26. If studies
reported on the results of two independent treatment subgroups, effect
sizes for each subgroup were included in the meta-analysis and treated
as a separate study to enable moderator analyses for the type of
treatment.
2.4. Heterogeneity,
sensitivity, and publication bias
Based on the Cochrane Handbook Version 6.1.0, 2020, heterogeneity was
assessed using Cochrane’s Q-statistics to estimated the standard
deviation of the true effect sizes and theI2 -Index27. A significant
Q-test indicates that effect sizes of primary studies do not belong to
the same distribution of effect sizes. When performance qualification
statistics ≥0.05 (P ≥0.05), was considered no significant
heterogeneity among the included studies25.I 2 index is used to the estimated amount of
variability in the true effect sizes, and the proportion of observed
variability that can be explained by true heterogeneity. Values of 25%,
50%, and 75% for the I2 -index indicate low,
moderate, and high degrees of heterogeneity, respectively. To further
explain the heterogeneity of our main results.
The ”Leave-one-out” method is used in sensitivity analyses to check for
outliers that potentially influence the results of the meta-analysis
disproportionately25. All analyses were performed
repeatedly with each study removed once to detect whether overall
results depend on a single study.
Publication bias means that statistically significant results are more
likely to be published, while statistically insignificant results are
less likely to be published. Therefore, these studies with no
significant significance could be more likely to remain in the ”file
drawer”28. Publication bias was assessed by three
methods. Funnel plots illustrate the effect sizes of primary studies as
a function of study precision, operationalized by the standard errors of
studies. Asymmetry in plots can indicate publication
bias29. Egger’s regression test yields a statistical
verification of funnel plot asymmetry. If any bias could be assumed
based on these analyses, we planned to apply the
trim-and-fill
procedure to estimate the unbiased overall effect30.