When and how to update systematic reviews: consensus and checklist

Updating of systematic reviews is generally more efficient than starting all over again when new evidence emerges, but to date there has been no clear guidance on how to do this. This guidance helps authors of systematic reviews, commissioners, and editors decide when to update a systematic review, and then how to go about updating the review.


APPENDIX 2. SEARCH STRATEGIES FOR A SYSTEMATIC REVIEW UPDATE
Newly emerging research is helping refine searching for studies for systematic review updates. This is based on an analysis of original searches and the yield in relation to databases searched, terms, and languages to improve the specificity of searches and reduce the burden of author screening. Further research will help clarify the effectiveness and efficiency of these innovations.

Refining the original searches
Database selection and search strategy optimization Database selection: While the original search may include multiple databases of many journals, it may be possible to limit the number of databases to the minimum set that would have identified all the original included studies. In some reviews, MEDLINE alone will suffice.
Strategies: Search strategies from the original review can be optimized. Investigators should confirm that the search strategies for the retained database would find the included studies from the original review and adjust the search strategies, if necessary, to improve recall. Examples of adjustments would be adding a key subject heading that was omitted from the original search or adding a newly introduced subject heading. 1 After maximizing recall, the searcher should endeavour to optimize precision, removing terms with a low yield of relevant records. Analytic tools such as GoPubMed and PubMed PubReMiner can provide useful analytics for optimizing both recall and precision. 2 Language: Whilst the original search may include multiple databases of many journals in languages other than English, again examination of the yield of additional studies from other languages can be examined in the original review. In a number of topic areas trials are published only in English language journals, so that a process of searching for non-English language trials may not increase sensitivity of the search. However, this is not currently Cochrane searching policy.
Text word terms: If text word terms are included in the search in the interests of identifying solely not-yetindexed material, restrict those terms to un-indexed records only, for example, in PubMed the string (pubstatusaheadofprint or publisher[sb] or pubmednotmedline[sb]) can be used. 3,4 Using a PubMed-only bespoke search strategy for an update: If the original included studies are all indexed in MEDLINE and the original search process was robust (for example, involved two or more databases and at least one non-database method for identifying relevant studies) a PubMed-only update can be considered. This would consist of two PubMed searches. The first would be a narrow Boolean search of the main MeSH for the population combined with the main MeSH for the intervention. The second would be a search using the Related Articles feature with the PubMed IDs (PMIDs) of the three newest and three largest included studies as the seed articles. 5,6 Can searches for updates be limited update by date?
Where the date of the original searches is known, and the original searches were well-designed and wellconducted, the update search should probably be limited to material added to the relevant databases since the original searches were conducted. 7 At a meeting with US National Library of Medicine (NLM) staff to discuss updating searches using PubMed, they advocated use of the Create Date [CRDT] field, i.e. the date the citation record was first created (McGhee M and Zipser J, oral communication, 20 th June 2014).
Where the date of the original search is unknown, one approach is to update the search to include the six months prior to the record creation of the newest included study. In all cases, however, care should be taken to ensure identification of (a) retracted studies (b) errata and corrected records. 8

Expanding searches to identify retracted studies or errata/corrections
In all cases, the bibliographic record for included studies from the review being updated should be checked to identify retraction and errata. 8

Additional update searches for trial registers and grey literature
In line with current Cochrane search guidance, as well as searching databases that contributed to the original review, update searches should query trial registers, typically ClinicalTrials.gov and the WHO portal. 8 As well as identifying studies with results, investigators should identify completed studies without posted results and in progress studies, which can be listed as "studies awaiting assessment".
If, in the original review, no studies were identified solely through grey literature searches, or they are small or appear only as early studies, it may not be productive to update the grey literature search.