Searching the Scientific Literature

Why Systematic Searching Matters

The scientific literature is the collective memory of the research enterprise. Every hypothesis tested, every method refined, every result that failed to replicate, every treatment that proved effective or harmful — all of it, to the extent that it has been published, is encoded in that literature and available to researchers who know how to find it. A researcher who fails to search that record thoroughly before beginning a new study risks duplicating work that has already been done, missing methodological innovations that would strengthen their own design, building on assumptions that have already been challenged, or — most consequentially — exposing research participants to risks that prior work would have shown to be unnecessary. The systematic literature search is not a preliminary formality to be dispatched as quickly as possible before the real work begins; it is itself a form of scientific practice, with standards of rigor and reporting that matter for the integrity of the research that follows NASEM 2018.

The word “systematic” is important here and should be understood precisely. A systematic search is not simply a search that is organized or careful, though it is both of those things. A systematic search is one that is conducted according to an explicit, pre-specified protocol; uses a defined set of databases and search strategies; applies consistent inclusion and exclusion criteria; and is documented in enough detail that another researcher could reproduce it. This level of rigor is required for formal systematic reviews and meta-analyses, where the credibility of the synthesis depends entirely on the credibility of the underlying search. For other kinds of literature reviews, a similarly rigorous approach is best practice, even if the formal documentation requirements are less stringent. The habits of systematic searching — using controlled vocabulary, combining search terms deliberately, recording what was searched and when — serve researchers well regardless of the type of review they are conducting.

Avoiding duplication is perhaps the most obvious reason to search thoroughly, but it is not the only one. A comprehensive literature search also helps researchers identify the most appropriate study designs for their question, understand the range of outcome measures that have been used in prior studies (and why some may be preferred over others), find validated instruments and protocols that can be adapted or directly adopted, and understand the methodological controversies and limitations that characterize a given body of literature. In this sense, the literature search is not only a quality-control measure but an act of scientific intelligence-gathering that shapes every subsequent decision in the research process.

Major Databases and Information Sources

The landscape of scientific databases is large and, for the uninitiated, bewildering. Different databases index different literatures, use different vocabularies, and are updated on different schedules. Knowing which databases are relevant for a given question, and understanding the distinctive strengths and limitations of each, is a practical skill that takes time to develop but repays the investment many times over.

PubMed, maintained by the National Library of Medicine, is the primary bibliographic database for the life sciences and biomedical research. It indexes more than thirty-five million citations from more than twenty-nine thousand journals and is freely accessible without a subscription. PubMed’s great strength is the Medical Subject Headings (MeSH) vocabulary, a hierarchical controlled vocabulary that assigns standardized subject terms to indexed citations, allowing searchers to retrieve literature on a given topic regardless of the specific terminology used by different authors. Understanding how to use MeSH terms effectively is one of the most valuable searching skills a biomedical researcher can develop, and it is discussed in more detail below. PubMed also offers powerful filtering capabilities, allowing searches to be restricted by publication type, date range, species, language, and many other parameters.

Web of Science and Scopus are multidisciplinary citation databases that index a broader range of scientific and scholarly literature than PubMed, including journals in engineering, the social sciences, and the humanities. Both databases track citation relationships — they record which papers have cited a given publication — which is essential for the citation-chasing techniques described later in this chapter. Web of Science is also the source of the citation metrics (impact factor, h-index) discussed in Chapter 10. Both databases require institutional subscriptions, and access varies by institution.

Embase, produced by Elsevier, is particularly strong for pharmacological and clinical literature, and it provides more comprehensive coverage of European journals and conference proceedings than PubMed. For pharmacological topics, drug safety questions, and systematic reviews requiring comprehensive coverage of the clinical literature, Embase is often essential in addition to PubMed.

Preprint servers represent a newer and increasingly important layer of the scientific literature. bioRxiv (for biology) and medRxiv (for health sciences and clinical research) host manuscripts that have been posted publicly before peer review. Preprints accelerate the communication of findings — a paper can be posted and read by the community within days of completion, rather than waiting months for peer review and publication — and they have become particularly prominent in fast-moving fields and during public health emergencies Polka et al. 2018. Searching preprint servers is increasingly important for comprehensive literature reviews, particularly in rapidly developing areas where the peer-reviewed literature may lag significantly behind the current state of research. The important caveat is that preprints have not been peer reviewed, and evaluating their quality requires the critical appraisal skills discussed in Chapter 4.

Constructing an Effective Search

A well-constructed database search combines controlled vocabulary with free-text terms and uses Boolean operators to define the logical relationships between search concepts. This sounds technical, and the mechanics do require some practice to master, but the underlying logic is straightforward and extremely powerful once internalized.

Controlled vocabulary terms — MeSH terms in PubMed, Emtree terms in Embase — are standardized labels assigned by professional indexers to the concepts discussed in a paper. The value of controlled vocabulary becomes clear when you consider that different authors, writing about the same phenomenon, may use different terminology: one paper about heart failure might use that exact phrase, while another uses “cardiac insufficiency,” another “congestive heart failure,” and still another “ventricular dysfunction.” A search that retrieves only papers using the phrase “heart failure” will miss all of those papers. A search using the MeSH term “Heart Failure,” which has been assigned to all papers dealing with that condition regardless of the specific phrasing used, will retrieve all of them. Using controlled vocabulary also helps with specificity: MeSH terms are organized hierarchically, allowing searchers to retrieve papers on a specific concept without being overwhelmed by papers on the broader category.

Free-text terms — also called keywords or title/abstract terms — search for the literal string of characters in specific fields of the database record. These terms are essential for several reasons. First, MeSH indexing has a lag: newly published papers may not yet have been assigned MeSH terms, so a search restricted to MeSH terms will miss the most recent literature. Second, some concepts are not captured by any existing controlled vocabulary term, particularly highly specific or emerging topics. Third, free-text searching allows the retrieval of papers that use specific technical terms that are not part of the controlled vocabulary. Best practice is to combine MeSH terms with free-text searches using the Boolean OR operator, so that the search retrieves papers indexed with the controlled vocabulary term as well as papers containing the relevant text strings in their titles and abstracts.

Boolean operators — AND, OR, NOT — define the logical relationships between search elements. OR is used to combine synonymous or related terms, broadening the search to capture papers using any of the specified terms. AND is used to combine distinct concepts, narrowing the search to papers that address all of the specified concepts simultaneously. NOT is used to exclude papers containing specific terms, but should be used with caution because it can inadvertently exclude relevant papers. A typical search strategy for a focused clinical question might use OR to combine synonymous terms for the population (e.g., “adolescents” OR “teenagers” OR “youth”), OR to combine synonymous terms for the intervention, and then AND to combine the sets of population terms and intervention terms. Truncation and wildcard characters (typically an asterisk in most databases) allow a single search term to retrieve all words sharing a common root: “pediatr*” will retrieve pediatric, pediatrics, pediatrician, and pediatricians in a single step.

Applying filters after a search is run can substantially sharpen the results. Common filters include date range (restricting to publications within a specified time window), publication type (restricting to randomized controlled trials, systematic reviews, or other specific study designs), species (human or animal), age group, language, and sex. Filters are valuable tools, but they should be applied thoughtfully. Restricting a search too aggressively — for example, by limiting to English-language publications only — can introduce systematic bias by excluding important work published in other languages.

Citation Chasing: Forward and Backward Searching

Database searching, however thorough, captures only the literature that is indexed in the databases being searched. Citation chasing — the practice of moving through the citation network of a relevant paper — is a complementary technique that can identify important sources missed by database searches, including older foundational works, papers in journals not indexed in the searched databases, and grey literature such as reports and technical documents.

Backward searching, also called reference mining or pearl growing, involves examining the reference lists of relevant papers to identify older literature they cite. A key paper on a topic will typically cite the foundational studies and major reviews that established the field; following those citations leads into the older literature systematically and efficiently. Starting with a high-quality recent systematic review and mining its reference list is often an excellent strategy for quickly building familiarity with a literature.

Forward searching — finding papers that have cited a given key paper — is enabled by citation databases such as Web of Science, Scopus, and Google Scholar. If a landmark paper published in 2010 has been cited by five hundred subsequent publications, examining those citations can identify all the major studies that have been conducted in the decade and a half since. Forward searching is particularly useful for finding the most recent work in a field, since the most recently published papers, by definition, will have cited the existing literature rather than being cited by it Higgins et al. 2019.

Types of Literature Reviews

Not all literature reviews serve the same purpose, and understanding the distinctions between review types helps researchers choose the approach appropriate for their question and their resources.

A systematic review is a formal synthesis of all available evidence on a focused question, conducted according to a pre-specified protocol that is typically registered prospectively in a registry such as PROSPERO. Systematic reviews use exhaustive search strategies, apply explicit inclusion and exclusion criteria to identify eligible studies, extract data from included studies in a standardized way, and assess the risk of bias in each included study. When the included studies are sufficiently similar in population, intervention, and outcome, a meta-analysis can be conducted to pool their results quantitatively. Systematic reviews sit at the top of the hierarchy of evidence for most clinical questions because their conclusions are based on the totality of the available evidence rather than on any single study. Conducting a rigorous systematic review is a major undertaking, typically requiring a team of researchers and a period of many months.

A scoping review maps the extent and nature of a literature on a topic without attempting to synthesize findings for decision-making purposes. Scoping reviews are appropriate when a field is heterogeneous, when the question is broad, or when the goal is to identify gaps in the literature and set priorities for future research rather than to answer a specific clinical or policy question. They typically use comprehensive but not exhaustive search strategies and do not assess risk of bias in included studies.

A narrative review, sometimes called a traditional literature review, summarizes and synthesizes a body of literature according to the author’s expertise and judgment without the formal methodological requirements of a systematic or scoping review. Narrative reviews are valuable for providing context, explaining complex concepts, and synthesizing a field for a broad audience, but they are more susceptible to bias — selective citation of evidence that supports the author’s views, for example — than systematic approaches. Readers of narrative reviews should be aware of this limitation and should look for evidence that the author has engaged with contradictory evidence as well as supporting evidence.

Staying Current with a Rapidly Moving Literature

Conducting an initial literature search is only the beginning. In an active research area, new papers appear continuously, and a researcher who conducts a single search at the outset of a project and never revisits it may find that important new evidence has appeared by the time they write up their findings. Several practical strategies help researchers stay current with a literature that is always in motion.

Most major databases offer saved search functionality: a researcher can save a search strategy in the database and receive automated email alerts whenever new citations matching the strategy are added to the index. PubMed’s “My NCBI” account, for example, allows users to create automatic email alerts from any saved search. Scopus and Web of Science offer analogous alert functions. Setting up saved searches for a project’s core questions at the beginning of the project, and reviewing the alerts regularly, is an efficient way to ensure that emerging literature is captured as it appears.

Citation alerts are available through tools like Web of Science, Scopus, and Google Scholar and notify a researcher when a specific paper they have identified as important has been cited by a new publication. Since important new work is likely to cite the same foundational papers as prior work in the field, citation alerts on key papers can efficiently surface relevant new publications. Journal table-of-contents alerts, available directly from most publishers, deliver notification when a new issue of a specific journal is published — useful for monitoring the most important journals in a field for directly relevant new work.

The Role of Your Library

Research libraries and research librarians are resources that researchers, particularly new researchers, consistently underutilize. A research librarian with expertise in the life sciences or biomedical research can do much more than help a researcher find a specific paper. Librarians can help design and execute systematic search strategies, recommend appropriate databases for a given question, assist with setting up reference management software, facilitate access to papers through interlibrary loan when the institution does not have a subscription, and advise on grey literature sources that are not indexed in the standard databases. For formal systematic reviews, involving a research librarian as a collaborator — rather than simply as a consultant — has become recognized best practice, and many published systematic review guidelines explicitly recommend professional librarian involvement in the search process.

Interlibrary loan (ILL) is a service through which a researcher at one institution can obtain a copy of a paper or book that their institution does not own, borrowed from another institution that does. The ILL process typically takes one to five business days and is available to students and faculty at most research institutions at no direct cost to the requester. For researchers at smaller institutions with limited journal subscriptions, ILL is an essential mechanism for accessing the full breadth of the scientific literature. Researchers should also be aware of legal mechanisms for open access, including PubMed Central, institutional repositories, author manuscript preprints, and direct requests to authors — all of which can provide legitimate access to papers that are otherwise behind a paywall.

The investment required to search the literature thoroughly and systematically pays dividends throughout the entire research process. A researcher who knows the literature deeply is better positioned to formulate a genuinely important question, design a study that builds credibly on prior work, interpret their results in context, write a compelling introduction and discussion, and respond to peer reviewers’ questions about how their work relates to existing evidence. The library, in the broadest sense — the accumulated record of what the scientific community has learned and how it has learned it — is not an obstacle between the researcher and the bench; it is the essential context within which bench work has meaning.