A good literature search for a systematic review should be both comprehensive and reproducible. With this, the focus of the search should be on sensitivity rather than on specificity, i.e., to not potentially miss relevant studies and condoning to also identify irrelevant studies (which will be sorted for eligibility during step 5). The search strategy should be meticulously documented; this is not only important for other groups to potentially reproduce the search but it is also key for potential later update searches.
Databases
A broad variety of biomedical databases can be searched, most prominently:
- PubMed: Hosted by US National Library of Medicine, PubMed currently contains more than 35 million citations and abstracts of biomedical literature.
- EMBASE: A biomedical and pharmacological bibliographic Elsevier-hosted database containing over 32 million records.
- Web of Science: Paid-access platform providing access to multiple databases in a variety of academic disciplines.
- Scopus: Elsevier’s abstract and citation database with an interdisciplinary coverage.
| PubMed | EMBASE | Web of Science | Scopus | |
|---|---|---|---|---|
| Year of inauguration | 1997 | 1946/1972 | 2004 | 2004 |
| Description | PubMed is a free search engine that primarily accesses the MEDLINE database and hosted by the United States National Library of Medicine (NLM) at the National Institutes of Health. PubMed Central is a free online repository that archives open access full-text of scholarly articles that have been published in the fields of biomedicine/life sciences | A subscription-based biomedical database hosted by Elsevier, also covering PubMed/MEDLINE with an additional focus on pharmacology, medical devices, clinical medicine, and basic/translational science | A subscription-based platform providing access to multiple databases in a variety of academic disciplines | A subscription-based abstract and citation database covering life sciences, social science, physical sciences, and health sciences |
| Coverage | > 35 million records. Indexing > 6’000 journals | Almost 30 million records (including conference abstracts). Indexing around 8’500 journals | Indexing > 12’000 journals, 160’000 conference proceedings, and > 90’000 books | Indexing around 25’000 journals, around 5 million conference proceedings, 386 million scientific webpages, and 22 million patent records |
| Searching | Easy keyword searching and automatic mapping to MeSH terms | Advanced searches using Emtree (subjects with more natural language than MeSH) as well as basic search with synonym searching. Extensive filtering options available | Basic and advances search options. Allows citation tracking | Effective keyword/index term searching. Search results can be graphically analyzed |
| When to use | For quick searches or to gain a rough idea about the scope of certain keywords. Keyword mapping (MeSH terms). To explore MeSH terms | For guided Emtree subject searching (including MeSH). Excellent natural language searching. Drug/pharmacology topics. To systematically identify conference abstracts. For extensive filtering during the search | Besides searching for biomedical literature, it can be used to find information on authoring and citing activity by faculty | Broad coverage of journals published outside the U.S. and for broad coverage of non-English publications. For interdisciplinary field coverage |
Additional data bases with more dedicated content can be included, e.g., the Cochrane Central Register of Controlled Trials (CENTRAL). Because these data bases are only partly overlapping, it is important that the search encompasses at least two and ideally more of these data bases. Another caveat to consider is the use of multiple synonyms and search terms for each concept during data base search. And with the different functionalities and syntax used in each of these data bases, it is highly recommended to involve an information specialist for this task. Most academic libraries will offer such support for a comprehensive search for systematic reviews. Dedicated search filter, e.g., to identify randomized controlled trials or animal studies, can be applied to enhance specificity of the search.
Other sources of eligible references can be considered for systematic reviews such as screening of reference lists or grey literature. Screening reference lists of e.g., of reviews on related topics can identify additional eligible articles. Grey literature, i.e., material not published commercially or indexed by major databases, can also be included, for example conference abstracts or clinical trial registries. It has been suggested that grey literature should always be considered for meta-analyses on clinical trials to mitigate publication bias. However, grey literature has normally not undergone quality control by the peer-reviewing process. Thus, decision on whether to include such additional reference sources should be made on a case-by-case basis.
Craft your own search query
When you are crafting your own search string, we recommend the following heuristic approach:
- Come up with as many synonyms as possible for each of the main concepts of your PICO/research question. For example, consider the above research question: What is the effect of opicinumab on myelin regeneration in multiple sclerosis animal models? Check potential alternative names for the drug opicinumab (e.g., drug development codes or brand names) and make a list of multiple sclerosis animal models including potential (specific) abbreviations (e.g., experimental autoimmune encephalomyelitis, experimental allergic encephalomyelitis, cuprizone, lysolecithin etc.). Do not forget to consider UK and US spelling when coming up with synonmys.
- Combine these terms with Boolean operators “AND”, “OR” in a meaningful way. Also work with brackets if necessary. For example: (opicinumab OR BIIB033) AND (experimental autoimmune encephalomyelitis OR experimental allergic encephalomyelitis OR cuprizone OR lysolecithin) would search for all studies which tested opicinumab (with its synonym BIIB033) in multiple sclerosis animal models.
- You can further refine the search query by adding an animal keyword filter as developed by SYRCLE or by using the Boolean operator “NOT”, e.g., to exclude reviews (e.g., NOT reviews). Be aware that you might exclude relevant literature when using the Boolean operator “NOT”. Use quotation marks: “experimental autoimmune encephalomyelitis” will search for this exact term while experimental autoimmune encephalomyelitis will search for each of these three words individually. You can also add so-called wildcards: ischem* will search for ischemic and ischemia. In addition, exploring related MeSH terms (Medical Subject Headings) can be of help. The Polyglot Search Translator can assist you in translating the syntax of your search query across a variety of databases. We recommend to also get in touch with a librarian/information specialist from your local academic library. The University Library Zurich is happy to assist you.
- A guide to provide information on how to use different search operators in a variety of databases.
- A step-by-step guide to systematically identify all relevant animal studies (Leenaars et al., Laboratory Animals, 2012)
Combine and deduplicate your search
The search across different databases should be combined in a reference manager software such as Endnote, Zotero, or Mendeley. Most standard reference manager software also provide tools to deduplicate the pooled reference library which can substantially reduce the number of records to screen (since different databases are overlapping). A commonly used method for deduplication of reference libraries in EndNote is the Bramer method.
Pitfalls
A common pitfall during this step is an insufficiently sensitive search strategy which results in potential key studies being missed. An approach to mitigate this problem is to define a small set of benchmark publications. This is a set of publications which is likely eligible for the systematic review, and which can be tagged by the study authors. These benchmark studies should all be detectable by the respective search string.