13 Research methods, statistics, critical review and evidence-based practice13.1 Introductory concepts

13.1.3 Electronic databases and searching

5 min read3 figures

Written by Resident Doctor Committee · Published

Electronic databases

Bibliographic databases

No single database covers all relevant literature, so sources are chosen to fit the question, the discipline and the document types needed.

Database

Scope

Subject headings

MEDLINE

Biomedical journals from 1946, compiled by the US National Library of Medicine

Medical Subject Headings (MeSH)

Embase

Biomedical and drug literature from 1947, including conference abstracts

Emtree, with detailed drug terms

PsycINFO

Psychology, psychiatry and behavioural science, including books and dissertations

Thesaurus of Psychological Index Terms

CINAHL (Cumulative Index to Nursing and Allied Health Literature)

Nursing, midwifery and allied health journals

Modelled on MeSH, with added nursing terms

  • AMED, the Allied and Complementary Medicine Database, is compiled by the British Library and indexes complementary therapies such as acupuncture, together with physiotherapy, occupational therapy and rehabilitation.

  • The Cochrane Library holds the Cochrane Database of Systematic Reviews and CENTRAL, the Central Register of Controlled Trials.

  • CENTRAL lists citations of trial reports, often with abstracts, partly found by handsearching, so it finds trials that conventional indexing misses.

MEDLINE and PubMed

  • PubMed is the free search interface to MEDLINE, but it also holds records outside MEDLINE, such as some preprints, so the two are not the same.

  • New PubMed citations arrive before MEDLINE indexing with MeSH headings is complete, so the newest records lack subject headings for a time.

  • A search on subject headings alone misses these records, so searchers combine headings with title and abstract synonyms.

  • PubMed Central is a full-text archive, separate from PubMed's citation database.

Grey literature

  • SIGLE, the System for Information on Grey Literature in Europe, indexed reports, theses and conference papers from 1980.

  • SIGLE stopped adding records in 2005, and its content passed to OpenGrey, which has since closed and survives only as an archive.

Searching across databases

  • Each database has its own vocabulary, so searchers translate subject headings rather than paste a MEDLINE strategy unchanged.

  • For a systematic review of treatment, MEDLINE, Embase and CENTRAL are the usual core sources, with PsycINFO added for psychological work and CINAHL for allied health or qualitative evidence.

  • Searchers record the database, platform, dates, full strategy and retrieval counts so that the search can be reproduced and updated.

Reading forest plots

  • The largest effect is the estimate lying furthest from the null value, provided its confidence interval excludes 1; for a protective ratio it is the smallest number.

  • The width of a confidence interval shows precision, not the size of the effect.

  • Precision depends on the information behind the estimate: the number of studies, participants and, above all, events.

  • For ratio measures, interval width is judged on the log scale.

  • The k column beside each row gives the number of separate studies pooled for that row, not the number of participants.

Amitriptyline, with an odds ratio of 2.13 and an interval that excludes 1, lies furthest from the null line and so shows the largest effect; milnacipran (1.74) and paroxetine (1.75) have almost the same estimate, but milnacipran's much wider interval shows less precision, not a different effect size. Cipriani et al. (2018). Comparative efficacy and acceptability of 21 antidepressant drugs for the acute treatment of adults with major depressive disorder: a systematic review and network meta-analysis. Figure 3, p. 7. doi:10.1016/s0140-6736(17)32802-7

Box-and-whisker plots

Feature

What it shows

Line inside the box

Median

Length of the box

Interquartile range, the spread of the middle half of the data

Distance between the whisker ends

Overall range

The overall range can look wide because of one long whisker even when the box is narrow.

Skew

Median lies nearer the

Long whisker

Positive

Lower quartile

Upper

Negative

Upper quartile

Lower

The line inside the box marks the median (6.1), the box spans the interquartile range from 3.5 to 9.6, and the whisker ends give the range; the long upper whisker, with the median nearer the lower quartile, is the pattern of positive skew. The Twists and Turns, CC0, via Wikimedia Commons

Searching PubMed and the Cochrane Library

Boolean operators

Operator

What it returns

Effect

OR

Citations containing at least one of the terms

Widens the search

AND

Citations containing every term

Narrows the search

NOT

Citations without the term

Excludes terms

Look at the shaded area in each pair: AND keeps only citations containing both terms and narrows the search, OR keeps citations containing either term and widens it, and NOT removes citations containing the second term. Jakub T. Jankiewicz, CC BY-SA 4.0, via Wikimedia Commons (cropped)

  • Synonyms for one concept are joined with OR inside brackets, and separate core concepts are joined with AND.

  • PubMed processes operators from left to right, so brackets make the OR group run before the ANDs: schizophrenia AND (olanzapine OR quetiapine).

  • Within each concept block, combine MeSH headings with free-text synonyms so that records lacking a suitable heading are still found.

Automatic term mapping

  • Automatic term mapping translates untagged terms using subject headings and other indexes, so the brand name Seroquel maps to quetiapine.

  • The search details and the Advanced Search history show what actually ran.

Phrases, truncation and field tags

  • Double quotation marks make PubMed search the words together and in that order, and they switch off automatic term mapping.

  • A trailing asterisk truncates a word stem to retrieve every term beginning with those letters, as in antipsychotic*.

  • Truncation also switches off automatic term mapping, so no MeSH headings are added, and it returns broad, loosely related hits; for these reasons truncation is discouraged.

Tag

Field searched

[mh]

MeSH headings

[tiab]

Title and abstract

[ti]

Title alone

[pt]

Publication type

[tw]

Text words

[dp]

Date of publication

Filters

  • Species, age and publication type filters depend on assigned indexing, so relevant records that lack those tags disappear.

  • Language restrictions may introduce bias.

The Cochrane Library

  • Title, abstract and keyword searching uses syntax such as clozapine:ti,ab,kw.

  • Reviews and protocols are distinct from trial records in CENTRAL, and a trial record may be only an abstract or a registry entry.

  • Do not add a randomised trial filter to CENTRAL, because it discards useful records.

Search design by question type

Question type

Blocks that carry the search

Design filter

Therapy

Population and intervention

Validated trial filter when the question requires trials

Diagnostic accuracy

Condition and index test, by heading and free text

Accuracy filters can miss relevant studies

Harm

Exposure, population and sometimes the adverse outcome

Do not restrict to randomised controlled trials when observational evidence is needed

  • Record the platform, exact strategy, date, limits and counts so that someone else can repeat the search.

Revision material for the MRCPsych examinations. Not clinical guidance, and not for use in patient care.

13.1.3 Electronic databases and searching | MRCPsychiatrist