Meta-analysis: formulating, evaluating, combining, and reporting

Two articles recommended that the review question state which study designs will be considered in the review.

“Two articles recommended that the review question state which study designs will be considered in the review [47, 48].”

Aspects frequently mentioned as important to address in the research question were the population, exposures or interventions, and outcomes.

“Aspects that were frequently mentioned as important to address were the population, exposures or interventions, and outcomes [38–41, 43, 47, 48, 50, 54].”

The majority of articles discussed aspects general to systematic reviews including advantages and limitations of different literature sources and databases, search tools, the importance of identifying unpublished studies, literature searching techniques including how to build a search string and reporting.

“The majority of articles discussed aspects general to systematic reviews including advantages and limitations of different literature sources and databases, search tools, the importance of identifying unpublished studies, literature searching techniques including how to build a search string and reporting [12, 29, 33–36, 46, 48, 50, 51, 53, 54, 57, 59–62, 67, 68].”

Five articles gave recommendations about how to assess eligibility: it should be blind, independent, and performed by two reviewers.

“Five articles gave recommendations about how to assess eligibility: it should be blind [29, 30, 46], independent [29, 33, 46, 48], and performed by two reviewers [29, 30, 33, 48].”

One article recommended using a panel of experts to decide on the inclusion status of a study.

“One article recommended using a panel of experts to decide on the inclusion status of a study [48].”

Nine articles (10%) made recommendations on how to perform data extraction, with publication dates between 1995 and 2016 (median year 2009).

“Nine articles (10%) made recommendations on how to perform data extraction [29, 33–35, 38, 43, 47, 48, 74], with publication dates between 1995 and 2016 (median year of publication 2009).”

The majority of articles advised against the use of scales but 12 articles recommended their use.

“The majority of articles advised against the use of scales but 12 articles recommended their use [29, 30, 48–50, 52, 53, 56, 66, 77, 86, 91].”

Twenty (22%) articles reported on item "publication bias", with publication dates between 1994 and 2012 (median year 2001).

“Twenty (22%) articles reported on item “publication bias” [16, 29, 33, 34, 36, 46, 48–50, 55, 56, 61, 70, 78, 85, 95–99], with publication dates between 1994 and 2012 (median year of publication 2001).”

Different articles advise to graphically examine the presence of publication bias using funnel plots, or to assess it with statistical tests.

“Since “journals are typically more likely to publish results that establish a difference than those that do not” [48], and publication bias is difficult to eliminate, different articles advise to graphically examine its presence using funnel plots, or to assess it with statistical tests [16, 36, 48, 55, 85, 97, 99].”

An explicit recommendation to search for and include unpublished studies was given in several articles, with the argument that the results of unpublished studies may differ systematically from published studies.

“An explicit recommendation to search for and include unpublished studies was given in several articles [34, 48, 50, 61, 96], with the argument that the results of unpublished studies may differ systematically from published studies.”