Evidence Based Medicine (EBM)
Choosing a full range of appropriate keywords for your search is not easy but several techniques can help.
- Use relevant articles or citations already found
These can reveal alternative ways of expressing the topic. Look at subject headings of relevant citations and the terms used by authors in describing their studies.
- Look at subject heading scope notes (indicated by the icon in Ovid databases)
When searching on subject headings in databases such as Medline and PsycINFO, look at the scope notes provided. Scope notes usually provide you with a definition and alternative terms for the same concept.
- Ask the subject experts
A group of experts in the field can be helpful in giving you a brainstormed list of synonyms to start with. They may also help you break down a broad concept into its component parts. For example, cardiac rehabilitation comprises exercise therapy, educational intervention and counseling. All three aspects should be searched.
- Test your search
Once you've constructed your search, test whether it can retrieve articles known to be relevant. If it can't, analyse what is being missed and revise your search.
Eg. The range of synonyms that apply to the concept 'Coronary Artery Disease'
Searching with subject headings (MeSH terms) and keywords
We recommend you search Medline using both subject headings and keywords for optimal efficiency. This will increase your retrievals but applying a methodological search filter (Clinical Queries) will make this results set manageable. (See 'Limiting by research methodology' tab on the Search construction page)
Subject headings are not available in Scopus and will be automatically included in your search of PubMed.
What's the difference between subject headings and keywords?
A quick explanation of the differences and the implications of using one or the other in your search
How to use Medical Subject Headings (MeSH) in a Medline search - why bother?
Using Boolean operators correctly in a search
Turning search terms into a search strategy using AND/OR, truncation etc.
An excellent short video from University of Oxford.
The most effective way of restricting a search strategy to studies or systematic reviews employing the research designs most relevant to your question type is to use the Clinical Queries function in PubMed, Medline or PsycINFO. Clinical Queries includes highly sophisticated built-in methodological search filters which retrieve appropriate study designs for the clinical categories of therapy, diagnosis, etiology, prognosis, clinical prediction, and others.
1. In PubMed, choose the Clinical Queries option from the PubMed front page
2. Enter a basic search query, employing uppercase AND/OR and brackets around synonymous terms.
E.g. (heart failure OR cardiac failure) AND statins
3. Choose the type of question you are asking from the Category menu (e.g. therapy, diagnosis, prognosis, etiology)
4. Choose a Scope for your question - from either 'Broad' or 'Narrow'.
- For a Therapy question, 'Broad' retrieves clinical trials which may or may not be randomised. 'Narrow' retrieves only randomised controlled trials.
- For Reviews, 'Broad' retrieves most kinds of reviews (including narrative reviews) while 'Narrow' retrieves only systematic reviews.
The actual search strategies behind these variations can be viewed at http://www.ncbi.nlm.nih.gov/books/NBK3827/#pubmedhelp.Clinical_Queries_Filters
A wider range of Clinical Queries options are available in the Ovid Medline and PsycINFO databases. In additional to those filters found in PubMed, these databases include broad (sensitive), narrow (specific) and balanced filters for Economics and Costs questions, as well as Qualitative (experiences, values, attitudes etc.) questions.
To access Clinical Queries in these databases:
1. Run your search first
2. Click on Additional Limits under the Search box
3. Scroll down to find the Clinical Queries search limit box on the LHS.
Limiting to single study designs in databases
Use this option when you want to isolate specific study designs (e.g. observational studies) rather than targeting all study designs relevant to a question type.
|Medline (OvidSP)||PubMed||Scopus or Web of Science|
|Systematic reviews||Clinical Queries Review option (maximises specificity version)||Clinical Queries Review option (narrow scope)||Combine the following search string with your own topic search using AND: “systematic review” OR “meta-analysis” OR metaanalysis|
|Randomised controlled trials||Clinical Queries Therapy option (maximises specificity version)||Clinical Queries Therapy option (narrow scope)||Combine the following search string with your own topic search using AND: Random* OR placebo|
|Observational studies (cohort, case-control, cross-sectional)||
Epidemiologic studies/ or Cross-sectional studies/ or exp case control studies/ or exp cohort studies/ or controlled clinical trial.pt. or (cross-sectional or case control or cohort analy* or longitudinal* or prospective* or retrospective*).tw. or (((cohort or follow-up or followup or observational) adj study) or studies).tw.
|(Epidemiologic studies[mh:noexp] OR case control studies[mh] OR cohort studies[mh] OR Cross sectional studies[mh] OR Controlled clinical trial[pt] OR case control[tiab] OR cohort stud*[tiab] OR cohort analy*[tiab] OR Follow up stud*[tiab] OR followup stud*[tiab] OR observational stud*[tiab] OR longitudinal*[tiab] OR Retrospective*[tiab] OR cross-sectional*[tiab])||Combine the following search string with your own topic search using AND:
"cross-sectional" or "case control" or "cohort analy*" or longitudinal* or prospective* or retrospective* or "cohort stud*" or "follow-up stud*" or "followup stud*" or "observational stud*"
|Case studies or case series||
case reports.pt. or (case$ adj2 (study OR studies or report$ or series)).tw.
|(case reports[pt] OR case study[tiab] OR case studies[tiab] OR case report*[tiab] OR case series[tiab])||Combine the following search string with your own topic search using AND:
case* W/2 (report* or study or studies or series)
|Qualitative studies||Clinical Queries Qualitative option (maximises specificity version)||"Interviews as topic"[mh] OR Focus groups[mh] OR Qualitative research[mh] OR interview*[tiab] OR themes[tiab] OR thematic*[tiab] OR experience*[tiab] OR qualitative*[tiab] OR focus group*[tiab] OR questionnaire*[tiab] OR ethnograph*[tiab] OR fieldwork[tiab] OR field work[tiab] OR psychology[sh:noexp] OR psycholog*[tiab] OR psychosocial*[tiab]||Combine the following search string with your own topic search using AND:
interview* OR themes OR thematic* OR interview* OR experience* OR qualitative* OR "focus group*" OR questionnaire* OR ethnograph* OR fieldwork OR "field work" OR psycholog* OR psychosocial*
Other methodological search filters
The goal of searching for answers to clinical questions is to find as many relevant citations as you can while keeping out as many irrelevant ones as possible. This will inevitably mean sacrificing a degree of search 'sensitivity' (favoured by systematic reviewers) in favour of 'precision'.
Sensitivity is the number of relevant citations retrieved as a proportion of all relevant citations in existence. As it is impossible to calculate the number of relevant citations in existence, maximal sensitivity is an aspiration rather than an achievable goal.
Precision is the number of relevant citations retrieved as a proportion of all citations (relevant and irrelevant) retrieved. An ideal search would achieve 100% sensitivity and precision. In reality, searches are always a trade-off between these two values.
Clinician searches should seek to maximise precision while not unduly sacrificing sensitivity. While low precision has practical implications (you have to sift through a larger number of citations), searches with low sensitivity will fail to find a substantial proportion of relevant, possible key, evidence.
Techniques for improving search sensitivity
The following search conventions will help you optimise search sensitivity, ensuring you get a higher proportion of relevant articles.
Don't include too many concepts in the search strategy
The search should include the minimum number of concepts required to describe the topic without jeopardising precision. The smaller the number of concepts included in the search, the broader the search.
Omit any redundant or implied concepts
E.g. A search on 'elderly' may be unnecessary when searching on 'dementia' AND 'aged care'. Similarly the concept 'females' is implied when searching on 'postpartum depression' or 'menopause'.
Both subject headings and keywords should be included in the search (where possible)
Subject headings are database-specific thesaurus terms and keywords are the terms authors use in the titles and abstracts of their studies. Keywords help where indexers omit to include relevant subject headings. Subject headings help when authors use unexpected, or less common terminology to describe a concept, e.g. 'apoplexy' instead of 'stroke'.
'Explode' subject headings where appropriate
This will include narrower, more specific subject headings.
Avoid using 'focus' or 'subheading' options
These are offered as ways to improve search precision in databases such as Medline, PsycINFO, and CINAHL. These elements are subjectively assigned by a human indexer which means they can be susceptible to error or bias.
Considerable effort should go into identifying an exhaustive range of synonyms for each concept in the research question. These should be combined in the search using the OR boolean operator.
Consider variant word endings
These can be easily incorporated into your search using the truncation symbol. This is commonly the asterisk * (e.g. child* finds child, children, childhood, etc.)
Consider alternative spellings and hyphenated forms of words
These can be unanticipated. Common variants include: paediatric/pediatric, healthcare/health care, socioeconomic/socio-economic, outpatient/out-patient
Avoid using filters (other than English) in PubMed
Applying the 'Filters' shown at the left hand side of the PubMed results screen will immediately eliminate a subset of PubMed results. This includes citations with the status 'in process', or 'as supplied by publisher.' These usually constitute the more recently published literature.
Techniques for improving search precision
Sometimes certain keywords can prove too sensitive in a search, bringing in a large number of irrelevant citations. Rather than immediately removing them, try the following techniques for improving their performance.
Add additional 'anchoring' concepts
Consider 'anchoring' overly sensitive terms to a more specific context by combining them with another term using AND. For example, if 'Indigenous' on its own is retrieving too many studies focused on Canadian Indigenous populations, search 'Indigenous AND Australia*'.
Employ proximity or adjacency operators to control the distance between terms
If anchoring with AND doesn't improve precision, replace AND with a proximity or adjacency operator. These 'commands' let you dictate how closely two search terms should occur in relation to each other. E.g. Heart adj3 failure means heart must occur within 3 words of failure. Note: These operators usually differ across databases. Use the Database Syntax Guide if translating a search for a different database.
A short visual guide to truncation and 'wildcard' search operators