Literature reviews for medical devices need to be comprehensive and thorough1. This requirement places a lot of emphasis on getting the right keywords for setting up your searches, as the keyword selection directly determines the quality and amount of search results. This can be especially challenging when creating searches for a new type of device or in a domain you are not yet very familiar with.
Many people will start with explorative google or PubMed searches, read (systematic) review papers, consult textbooks or experts to get a good idea of the topics to search for. While these are valid, tried and tested methods, they are sometimes insufficient or too time-consuming.
Here we present you with an alternative approach to discover keywords and search terms for your review starting from a set of landmark ‘seed papers’ you know are relevant. Key to this method is to be aware that any seeding you use can introduce bias, so make sure to make your keyword discovery scope wide enough to prevent this.
Getting started: build an initial set of keywords
Domain-specific knowledge and expertise are of key importance for using the correct terminology and keywords in literature searches. If you are not an expert (yet) in the domain or topic of the literature searches you need to perform, contacting domain experts or reading up on the specific domain certainly are useful steps.
A reproducible and useful approach to building relevant keywords starts from a few landmark publications that you mine for keywords and Medical Subject Headings (MeSH) terms, which are then iteratively used to extend the keyword list. This process can either be executed manually, or use a number of tools or automation options, but conceptually, the process and steps are more or less the same. Either way, a set of landmark seed publications or a set of initial keywords are needed.
Extract keywords from seed publications
Look up your seed publications in PubMed. The MeSH terms are listed near the bottom of the Abstract view, but to see both MesH terms and author keywords, you need to change the display settings to PubMed view:
Then display or download the PubMed format of your seed papers and extract:
- Medical Subject Headings (MeSH) terms: lines starting with the MH tag
- Author keywords: lines starting with the OT tag (other terms)
Expand your initial list of keywords
Manual methods
- Get synonyms for your initial keywords: from a dictionary, thesaurus, controlled vocabulary such as MeSH or Emtree, sentinel articles, textbooks or webpages
- Extract keywords from abbreviations and acronyms
- Combine keywords into concepts, e.g. the concept “shoulder pain” as a combination of the shoulder and pain keywords
Use keyword discovery tools
Use keyword discovery tools on the internet. The list of available tools is long. The most useful ones for literature reviews on medical devices are discussed below.
Yale MeSH Analyser
The Yale MeSH analyzer 2 is a free tool to automatically create a MeSH analysis table, listing the associated MeSH terms and author keywords for up to 20 articles in an easy to scan grid format. The tool uses PubMed IDs (PMID) as input and can either a simple web page or excel file as output. The Yale MeSH Analyzer can be used from the web or directly in the PubMed web interface, after a very simple one-step browser setup.
PubMed PubReMiner
PubMed PubReMiner is a text mining tool that analyzes the results of a PubMed query it receives as input and returns frequency tables for the keywords, MeSH headings, substances, journals, authors, publication types and publication years in the results of that query.
PubReMiner can be used in multiple ways to expand or refine literature searches. To use it as a keyword discovery tool, you can either use it with a set of seed articles you know need to be included in your search and check the keywords and MeSH terms they are using in the PubReMiner output. Or you can enter a broad initial search and look for keywords and MeSH terms you know are relevant to the domain of your search.
CoreMine Medical
CoreMine Medical is an advanced text mining tool you can use to discover related concepts and keywords in the MEDLINE database starting from a number of input concepts. The output of a CoreMine Medical search is a graphical network representation of how these concepts are linked in biomedical literature, based upon the co-occurrence of biomedical terms such as MeSH terms, Gene/Protein, disease, molecular function, biological process, or cellular components. The tool is free to use, but you need to register for a free account, allowing you to save your work for future reference.
CoreMine Medical is especially useful to get a feel for the concepts that are relevant in a certain domain and will allow you to get an overview of a complex subject quickly. When using it as a keyword discovery tool, the possibility to drill down on every keyword in the concept network is very useful.
For those of you with programming skills
Tools such as LitsearchR 4 and its Python analogue Ananse 5 need the results of a naïve search with a limited set of highly relevant articles as input. These seed articles will then be used to extract keywords and terms and build a keyword co-occurrence network and provide a list of the most important keywords. These keywords can then be manually grouped into concept groups to use for building a final search. These tools unfortunately do not have a frontend application; they require use and knowledge of R or Python, making them less than useful for the typical medical writer.
Leverage AI Tools
Since OpenAI introduced ChatGTP in November 2022, the use of AI and large language models (LLM) in particular has taken flight and new AI tools for scientific research are proliferating like wildfire.
Below you will find a small selection of the large array of AI-based tools that can be used for discovering keywords related to scientific publications. Please take into account that, while very handy and intuitive to use, AI-based tools are not perfect and may produce inaccurate results or suffer from hallucinations. So use them with caution and make sure to double check the output of these tools.
|
Tool
|
Descripton
|
|---|---|
|
LitSuggest ⁶
|
A web-based machine learning application of the US National Library of Medicine where you can train your own model using seed publications to find additional references on a certain topic.
|
|
LitMaps
|
Find relevant related papers based on connections via citations and reference to a starting publication. The Pro version includes a specific keyword discovery tool.
|
|
ChatGPT plugins:⁷ SciSpace Scholar GPT, Scholar AI Consensus
|
ChatGPT-powered tools capable of analyzing PDFs from scientific articles. Using a suitable prompt, this tool can be used to extract keywords and domain-specific terms from a seed publication and suggest related MeSH terms.
|
|
Scite
|
AI tool for scholarly research. Primarily intended for discovery search and citation screening using deep learning to classify papers as providing supporting or contrasting evidence related to a topic.8 However, the Scite assistant can also be used to extract keywords and suggest keywords and MeSH terms if given the right prompt.
|
Links to tools: LitSuggest, LitMaps, ChatGPT plugins, Scite
References & Reading
- European Commission. MEDDEV. 2.7.1 Rev.4: Clinical Evaluation: A Guide for Manufacturers And Notified Bodies Under Directives 93/42/EEC and 90/385/EEC. MEDDEV 271 Rev4. 2009;(April 2003):1-65.
- Grossetta Nardini HK, Wang L. The Yale MeSH Analyzer. (Grossetta Nardini HK, Wang L, eds.). Yale University; 2018. https://mesh.med.yale.edu/.
- Scells H, Zuccon G. Searchrefiner: A query visualisation and understanding tool for systematic reviews. Int Conf Inf Knowl Manag Proc. 2018:1939-1942. doi:10.1145/3269206.3269215
- Grames EM, Stillman AN, Tingley MW, Elphick CS. An automated approach to identifying search terms for systematic reviews using keyword co-occurrence networks. Methods Ecol Evol. 2019;10(10):1645-1654. doi:10.1111/2041-210X.13268
- Kwabena AE, Wiafe OB, John BD, Bernard A, Boateng FAF. An automated method for developing search strategies for systematic review using Natural Language Processing (NLP). MethodsX. 2023;10(November 2021). doi:10.1016/j.mex.2022.101935
- Allot A, Lee K, Chen Q, Luo L, Lu Z. LitSuggest: A web-based system for literature recommendation and curation using machine learning. Nucleic Acids Res. 2021;49(W1):W352-W358. doi:10.1093/nar/gkab326
- Alshami A, Elsayed M, Ali E, Eltoukhy AEE, Zayed T. Harnessing the Power of ChatGPT for Automating Systematic Review Process: Methodology, Case Study, Limitations, and Future Directions. Systems. 2023;11(7). doi:10.3390/systems11070351
- Nicholson JM, Mordaunt M, Lopez P, et al. Scite: A smart citation index that displays the context of citations and classifies their intent using deep learning. Quant Sci Stud. 2021;2(3):882-898. doi:10.1162/qss_a_00146
