Evidence-Based Health Care: A Tutorial Part 2

Evidence-Based Health Care:Evidence-Based Health Care:A TutorialA Tutorial

  Joyce Condon, MLS, AHIP

Reference Librarian

Exempla Saint Joseph HospitalDenver, Colorado USA

Medical Library

This is Part Twoand will cover

designing a search strategy

Part Three will coverselecting appropriate evidence resources

StrategiesStrategies

Once you have formulated a well-built question about a clinical problem you must create a workable search strategy. This involves converting the question to a format that is understandable by the search engine(s) of the database(s) you wish to use. The goal is to develop high-sensitivity and high specificity search strategies from the question.

For high sensitivity use a very broad search strategy to maximize retrieval and not miss any pertinent references.

For high specificity use a more precisely defined strategy to yield fewer, but more relevant results. The challenge is to find the point of balance between these two requirements, which seem to represent the extreme ends of a search strategy continuum.

Devise an initial strategy appropriate for your chosen database. Then evaluate, refine and adapt the strategy as you evaluate the retrieval from different resources until the question is answered or you feel reasonably certain that the information does not exist.

The first step is to format the question using the PICO paradigm to focus on essential elements and select terms. Next you decide on the type of question you wish to ask and link it to a study design that will provide the highest level of evidence incorporating an EBM filter if appropriate.

To illustrate this process look at the sample clinical issue: What is the best first-line therapy for treating blood pressure in the elderly?

PICO AnalysisPICO Analysis

Patients Hypertensive elderly

Intervention Beta-bockers

Comparision Diuretics or other drugs

Outcomes Morbidity or mortality

Question Type:  Therapy or what is the effectiveness of beta-blockers vs. diuretics or other drugs for the initial treatment of hypertension in the elderly?

Type of Study Design:   Look for systematic reviews, meta-analyses or randomized controlled trials.

Search Strategy SummarySearch Strategy Summary

First Concept hypertension

Second Concept elderly

Third Concept Beta-blockers

Fourth Concept randomized controlled trial

Strategies for EBM DatabasesStrategies for EBM Databases

Appraised resources are small databases containing a few thousand records in contrast to MEDLINE that contains a few million. Most of these databases are full text wherein every word of the document text is searchable.

When searching databases like the Cochrane Library, the ACP Journal Club and POEMs keep the strategy simple using single terms, short phrases and synonyms. These sources do not incorporate a system of approved subject terms to index or describe the contents of their articles.

Therefore, the best approach is to use text word searching techniques. The free text search will retrieve records containing a term or character string, for example "hypertension". This is not the same as retrieving records that are about the concept of hypertension. The text word "hypertension" will not retrieve a record where the text discusses high blood pressure and does not use the term "hypertension".

Since a free-text search is a non-intelligent, character string type of searching you need to cover all options with search term variants and truncation. A free text search strategy should include variations in spelling and terminology and incorporate synonyms, plural forms, acronyms and abbreviations if appropriate.

Sometimes you may need to use truncation when doing free text searching. Truncation allows you to look for a particular word stem with different endings. To use truncation, type your word stem followed by a symbol. Truncation symbols vary. It may be an asterisk (*), the dollar symbol ($), a question mark (?) or something else. Truncation may be automatic, non-existent, or available only at the end of a word. Truncation broadens your search and increases retrieval.

For example, you will retrieve variant forms and spellings of tumor using tumo*. Retrieved documents may contain the term tumor or tumour (British spelling), tumors or tumours, tumorous or tumoral. Check a database's instructions or help pages to learn about its use of truncation. For our search you may use truncation for the second concept - elderly. Truncate the term to elder$ to retrieve elder, elders, and elderly. You can also use truncated synonyms like geriat$ for this concept.

Wildcards are symbols (like # or ?) to replace or stand in for one or no characters WITHIN words. This is useful for British/American spellings such as tumo?r, and for inconsistently hyphenated terms such as beta?blockers.

Since every word of the text is searchable, for common terms like hypertension you may need to restrict retrieval. For example, restrict the search to the title field or combine hypertension with other terms or limits.

Otherwise, you will retrieve many irrelevant records. For rare or unusual terms or a newly discovered disease process (e.g., when AIDS was first being described) single term searching of the entire record may work well for you.

A free text search for our example would include the three terms (hypertension, elderly and beta-blockers) linked together using logical (Boolean) operators, AND, OR AND NOT. You can enlarge or restrict retrieval using these operators. OR increases retrieval while AND and NOT reduce retrieval. AND is used to join terms or concepts together and focuses the search retrieving only those records which contain ALL of the keywords or concepts you specify. It is generally used to combine the different or unrelated concepts in your search.

e.g. hypertension AND elderly AND beta-blockers

OR searches for either one term or another and broadens the search retrieving records which contain ANY of the search terms specified. It is generally used to combine synonyms and terms representing related concepts.

e.g. elder$ OR older OR geriat$

NOT will exclude terms and narrows the search. You should be cautious in using this option. You may eliminate important pertinent studies if you are not careful. For example, you could use a term like young or middle aged with the NOT operator in our search to restrict to elderly adults.

However, the text may describe the population of subjects as "the elderly, with young and middle aged adults excluded". Since the term young or middle aged appears in the text this article would be eliminated from the retrieval set.

For our example our initial search strategy will combine our three terms from the PICO analysis:

Hypertension AND elder$ AND beta?blockers

Review initial results for relevance to the focused question. Should the search be extended to pick up further references? Does it contain too few or too many results? Are the study designs appropriate to the question? The answer to such questions will determine whether a revised search strategy is required.

Searching MEDLINE

The MeSH SystemMEDLINE is a large general health-care database covering all health specialties including nursing. The National Library of Medicine (producer of the MEDLINE database) uses a controlled vocabulary (a list of authorized subject terms) or thesaurus to index or describe the content of articles included in the database. Therefore, the database record for each article citation is associated with a set of subject terms or descriptors called Medical Subject Headings (MeSH) that describe the content of the article. It is important to be aware of the structure of MeSH to formulate search strategies in MEDLINE. The MeSH thesaurus of subject terms is arranged hierarchically, with broader (more general) subject headings at the top and narrower (more specific) subject headings listed beneath them.

The MeSH vocabulary also includes subheadings or qualifiers that can be combined with a MeSH heading (descriptor) to make it more specific. Subheadings divide subjects into their particular aspects. For example, "hypertension" is a MeSH term.

If you are interested in effective therapies for hypertension you could combine the "therapy" or "drug therapy" subheadings with hypertension to narrow your retrieval to therapy articles on hypertension.e.g. hypertension/ therapy or hypertension/drug therapy. Articles are indexed and retrieved using the most specific MeSH terms available.

The Thesaurus Search:Is It Worth the Trouble?

With a thesaurus search you amend your strategy so that your search terms are matched to the corresponding MeSH terms. A sophisticated search engine will attempt to map your terms to the appropriate MeSH headings. MeSH terms apply consistent subject headings to articles that describe the same concept in a variety of ways. Therefore, thesaurus searching retrieves records about the term regardless of plurals, variable spellings, use of synonyms and acronyms, etc.

In addition, you can EXPLODE the MeSH term to include all the more specific, narrower terms listed beneath it. For example, the MeSH term for beta-blockers is adrenergic beta-antagonists. If you EXPLODE this term the search engine will OR together all the specific beta-blocker drugs together with the general MeSH term, adrenergic beta-antagonists. You need not type in each drug and connect it with the logical operator "OR". This is the power of the thesaurus search compared to the free-text search which retrieves only records containing the exact term or character string.

Searching MEDLINESearching MEDLINE

Thesaurus searching is a very effective way of searching a database, especially when combined with text words when it is appropriate. The MeSH structure also allows you to focus (or "major") a MeSH term. This is designated by an asterisk (*) placed by the term.

Majoring a term indicates this concept defines the major focus and concerns of the article and indicates what the article is really about. Focusing or majoring a term will reduce retrieval by ensuring your subject heading is the primary subject of the articles. Thesaurus or MeSH searching is precision searching.

Limiting the SearchLimiting the Search Limiting refines a search by applying select criteria. Much of the information available in large bibliographic databases such as MEDLINE is not evidence based. To make searching more effective and efficient, "filters" have been developed to create literature searches that yield a higher percentage of clinically relevant articles.

The objective of filtering is to reduce the retrieval to articles that have direct implications for patient care and that report clinical research conducted with specific methodologies. So filtering limits our subject search to retrieve only citations of studies with a specific methodology.

What Are Quality or Methodology Filters?What Are Quality or Methodology Filters?

Filters are generic search strategies incorporating research methodology terms as free text, thesaurus (MeSH) terms, publication types or truncated terms. These filters have been designed and tested to retrieve study types appropriate to diagnosis, prognosis, therapy, or etiology/harm/risk. Therefore, the filters conform to specific types of clinical questions (See Clinical Map) and help restrict the citations found to the types of studies that are likely to provide valid answers to those questions. These specially developed generic strategies or hedges are sometimes referred to as "quality filters" or "methodology filters".

To devise your own filter for a subject search begin by limiting it to the study design that usually provides the best evidence as listed in the evidence pyramid that is appropriate to the type of question you are asking. The Clinical Queries mode in PubMed uses stored search strategies that serve as quality filters for therapy, diagnosis, etiology, prognosis or systematic reviews.

What Are Publication Types?What Are Publication Types? Some large databases like MEDLINE allow you to search using designations of what a citation "is". This is often referred to as a "publication type". A publication type identifies the article as falling into one or more of several categories. Some of these categories conform to specific study designs.

For example, an article entitled, "Swedish Trial in Old Patients with Hypertension (STOP-Hypertension) analyses performed up to 1992", was indexed with the publication types "clinical trial" and "randomized controlled trial".

This study reports the results of a randomized controlled trial. To find this article in MEDLINE with a minimum of articles constituting commentaries, narrative reviews, or reports of weaker study designs, restrict the search to "randomized controlled trial" in the publication type field.

In this case only studies classified by the National Library of Medicine (NLM) as randomized controlled trials will appear. Quality or methodology filters incorporate publication types to match your question type to type of study. Publication types are very effective for identifying high-quality therapy studies and systematic review articles.

What Is Field Limitation?What Is Field Limitation? Documents can be divided into unique elements of information, such as title, author, abstract, publication type, language, date etc. These elements can be labeled in a database with two letter codes, such as TI for title or PT for publication type or AU for author, etc. These elements are often referred to as fields. You can narrow a search by restricting search terms to specific fields.

For example, you can search for "elderly" in the title field of records indexed in MEDLINE. In this case the search engine will check for this term in the title only. If the term appears in the abstract or subject headings, but not in the title of the article the record will not be retrieved.

Additional LimitsAdditional Limits

You may also restrict your retrieval by age, language, human studies, and date of publication or subject and journal subsets. Specific limits will vary by database and vendor.

OVID vs PubMedOVID vs PubMed

A number of vendors, each with their own search interface, distribute MEDLINE. How you access the database will influence how you construct your literature search. This is so because programs offered by different vendors for accessing MEDLINE may vary in their performance in important ways. Among the current distributors of software and online packages that include access to MEDLINE are the National Library of Medicine or NLM (PubMed which is free) and commercial vendors such as OVID Technologies which is fee-based.

MEDLINE via PubMed is usually the most current version since it is produced by NLM and is updated daily. Other differences between distributors include links to full text and related articles, integration of quality filters into the search options, the mechanics for text word searching, truncation and field limitation and inclusion of special utilities like Citation Matcher and Journal Browser in PubMed. The table below compares OVID and PubMed with respect to some common search features.

Comparison of PubMed and OVIDComparison of PubMed and OVID

Search Engine PubMed OVID

URL pubmed.gov gateway.ovid.com

Access Web: Free Web/CD-ROM: Fee-based

Text word searching Searches text word as entered. Maps to relevant subject heading. If no match is found, searches all fields.

Text word will search title, abstract, registry number word, and MeSH fields.

Search Engine PubMed OVID

Subject searching Uses automatic term mapping. Terms are matched against MeSH, Journal title table, Phrase list and Author index. MeSH browser allows selection of specific subject headings. Automatically EXPLODEs MeSH terms and subheadings. Can turnoff EXPLODE (eye[mh:noexp]) for subject headings, but not for subheadings. Can MAJOR (eye[major]) or FOCUS a MeSH term. Searching with MeSH excludes IN PROCESS and Publisher-Supplied citations

If selected, the system automatically maps to a MeSH term. Searcher must verify/select a term for it to be used. Can choose to EXPLODE (exp eye) or Focus (*eye) a MeSH term. Pre-MEDLINE available as seperate file - use text word searching.

Search Engine PubMed OVID

Phrase searching Searches from a predigested set of phrases. Does not do true adjacency searching. Enclosing a phrase within quotes bypasses automatic term mapping.

Yes. Statements should be field qualified: Pregnancy tests.tw. Adjacency is assumed. (information adjX retrieval).tw. Retrieves phrases in which "information" and "retrieval" are within X words of one another.

Truncation Right-handed truncation only (elder*). No internal truncation. Truncation turns off automatic mapping to MeSH term and turns off EXPLODE.

Statements should be field qualified:Use $ or : to retrieve unlimited suffix variations (computer$.ti.)Use # to replace exactly one character (wom#n.ti.)Use? to replace 0 to 1 characters (labo?r)

Order of Citations Most recent first. Variable

Search Engine PubMed OVID

Limits Limits: Language, year of publication, publication type, human or animal, document subset (AIDS, Bioethics, Cancer, Complementary Medicine, Core collection journals, toxicology, etc), age, searching only in a particular field (title word, title/abstract word, author, etc).Can only choose single entry from menu boxes.

Available on search page:English language, human only, review articles, articles with abstracts. Many additional limits available under the "limits" button on separate page. Can choose multiple entries from menu boxs. Can limit word or phrase search to particular field of document: title, author, abstract, publication type, and >40 other specific fields. Example is: eye.ti.

Search Engine PubMed OVID

Extra Features "Clinical Queries" and "Systematic Reviews" and Quality or Cost-related research filters (EBM quality filter search features.)Journals Database; Citation Matcher; Cubby; Links to related articles; Links to full-text articles (most subscription based). Includes citations to other related databases: PreMEDLINE,Toxicology, etc."History" permits refinement of search using a combination of previous search statements. Can email search retrieval.

Links to full-text journals and "EBM Reviews" with subscription. Continuously running history on search page permits combination of any number of search lines. Can save searches for 24 hours or permanently. Can email search. PreMEDLINE available as separate database.

Searching Web EBM ResourcesSearching Web EBM Resources

Most EBM Web databases use keyword search tools that search title and/or text. They are generally small databases. Before beginning a search read help screens to learn how to enter a query. Adapt your strategy by choosing the single most important concept or term and search on that term.

If necessary, refine this search by using synonyms or term truncation (if permitted) to increase retrieval. To restrict retrieval add the next important concept or term. Review the section on text word searching. You can adapt text word searching techniques according to the specific requirements of the Web resource search engine. For example, learn if truncation is permitted and what symbol to use or if truncation is automatic.

For our sample search, the most important concept is hypertension. Begin with this term. This is a common subject. If retrieval is large, restrict by adding elderly using truncation if permitted. If the retrieval is not relevant try combining elderly with the third term, beta-blockers.

The key is to remember that effective searching is an iterative or cyclical process. You must try a strategy, review the retrieval and adapt the strategy according to the results obtained.

What is the Clinical Issue?     

Example: What is the best initial first-line therapy for treating high blood pressure in the elderly?

Step 1: Do PICO Analysis

Use logical (Boolean) operators to show concept relationships.

Combine With

Combine With

Combine With

Patient Population

Interventions/

Comparison

Outcome

hypertension

(MeSH)

AND older (text word)

AND beta-blockers (text word)

AND Morbidity (MeSH)

OR OR OR

elder$ (truncated text word)

Adrenergic beta-antagonists(MeSH)

Mortality (MeSH)

OR OR OR

geriat$ (truncated text word)

diuretics (MeSH)

This table uses OVID truncation symbols. For PubMed searching use the asterisk (*).Note: An effective search strategy usually combines both subject terms and text words.

Step 2: Decide on Question TypeHere the focus is therapy. What we really wish to know is the effectiveness of different drugs for treating hypertension in the elderly.

Step 3: Type of EvidenceSince this is a therapy question we will look for RCT's, systematic reviews of RCT's, meta-analyses or EBM guidelines. To focus retrieval on these types of studies incorporate a therapy methodology or quality filter in the search strategy or use MeSH publication types and/or text words in the search strategy.

Step 4:Decide on Resources Strategy

Databases:First search appraised resources (ACP Journal Club, Cochrane Library, DARE).       Next search MEDLINE via OVID or PubMed using quality filters.

    Finally, check Web EBM sites: TRIP, SUM-SEARCH, Bandolier.

Step 4:Decide on Resources Strategy

Databases:First search appraised resources (ACP Journal Club, Cochrane Library, DARE).       Next search MEDLINE via OVID or PubMed using quality filters.

    Finally, check Web EBM sites: TRIP, SUM-SEARCH, Bandolier.

Sample Search Strategy for Searching MEDLINE via OVIDSample Search Strategy for Searching MEDLINE via OVID

Search each subject term separately. Use both MeSH terms and appropriate text words. Link the terms with the appropriate logical (Boolean) operator:

OR to retrieve records containing any of a set of terms. Generally used to combine synonyms and related concepts.

AND to retrieve records containing ALL of a set of terms. Generally used to combine different or unrelated concepts.

1.    EXPLODE hypertension/all subheadings       (Explode increases sensitivity by also searching the more specific terms under the general term. Searching all subheadings increases sensitivity.)

2.     older OR elder$ OR geriat$       (Use truncation to search the root of a word and retrieve all variations that begin with the root. Use dollar sign, $, to truncate in OVID and the asterisk, *, to truncate in PubMed.)

3.     Beta?blockers OR betablockers OR adreneric beta-antagonists       (? is the truncation symbol to replace 0 or 1 character in OVID. Cannot do this in PubMed. The term mapping feature in OVID and PubMed helps you find the right MeSH term for beta-blockers, i.e. adrenergic beta-antagonists.)

Combine the results of separate subject searches using appropriate logical operators:

4.    1   AND   2   AND   3

Use appropriate quality or methodology filter (for this search, a therapy filter) or publication types or text words pertinent to study design.

5.    Clinical trial (publication type) OR randomized controlled trial (publication type)

6.  Placebo.tw. OR blind$.tw. OR random$.tw.

Combine filter search statements:

7.  5 OR 6 Combine results of subject search (#4) with results of filter search (#7)

8.  4 AND 7

Evaluate retrieval and modify search strategy. Test different combinations, amending the initial strategy to widen or narrow the search as required. Remember searching is an iterative or cyclical process. Add limits for language, age, publication date, as appropriate.

Search Strategy TipsSearch Strategy Tips

PREPARE! Formulate a well-built question using the PICO model and clinical map checklist to determine search terms and question type.

Choose the relevant database(s) to search.

Before beginning a search read the help screens to learn about the search engine. Is there mapping to subject terms? Must you search each term separately or is phrase searching allowed? What logical operators to show concept relationships are allowed? Is truncation automatic or do you use a special truncation symbol?

Combine subject term (thesaurus) search with text word search techniques.

Search Strategy TipsSearch Strategy Tips

Aim for a high sensitivity search initially:

Use explode to search more specific terms related to the general search term.

Do not automatically major or focus a MeSH term.

For a text word search, include as many synonyms as possible. Be aware of variable spelling, plural form and acronyms.

Truncate text words when appropriate.

When using a MeSH term, also use its equivalent in text words.

Do not routinely use MeSH term/specific subheading combinations when you begin searching.

Combine subject search with appropriate quality filter or publication type.

Search Strategy TipsSearch Strategy Tips

7. Evaluate retrieval and modify search strategy, if necessary.

8.To restrict a high sensitivity search to make it more specific and relevant:

Major the most important MeSH term. AND in another concept or use more specific term. Limit text words to title fields. Use phrase searching where text words are adjacent. Combine MeSH term with a subheading to specify a particular aspect of

the subject. Limit by publication type, document subsets (peer-reviewed journals,

complementary medicine, nursing, etc.), language, age groups, and date.

This is the end ofPart Two

Part Three will cover selecting appropriate evidence resources