Begin by defining your research question. You should identify who and what are the population, interventions, comparisons, outcomes, study design (PICOS) and study characteristics you are interested in, these are the beginning of your inclusion & exclusion criteria you will use to decide what articles you wish to include in your review. On the website of the Canadian Centre for Evidence-Based Medicine you can practice how to formulate an answerable question using PICOS for different types of clinical scenario: Therapy, Diagnostic, Prognosis, and Harm. Write down your keyword concepts, documenting them on a spreadsheet. Identify the MEDLINE MeSH headings used for your keywords - try the Yale MeSH Analyzer into which you can copy and paste up to 20 PMID numbers from search results in PubMed/MEDLINE and generate a MeSH Analysis grid to quickly scan the MeSH headings that were used to index those journal articles. Distribute the spreadsheet with your keywords and search terms to your research team colleagues and seek feedback from them and from your librarian. It is likely your spreadsheet will grow as you read around the topic and find new or related additional keywords and concepts.
SOURCES FOR SEARCH STRATEGIES
Someone may have already carried out a review on your topic and these can be a good source for finding a search strategy, some review guidelines such as PRISMA require the author to include an example of the search strategy for at least one database including the search keywords used, and in most cases this will be the MEDLINE search keywords and MeSH terms.
STUDY DESIGNS TO LOOK FOR
To help you think about what type of clinical studies you may want to include in your review read this description by staff at Duke/UNC medical libraries of the best study types for the type of question you are investigating. Think about what type of study design best collects the type of data you want to find and compare in your analysis e.g. if you want to analyse observational data what would be best: a time-series cohort study that measured data at several uniform time intervals, or a spatial data study that recorded the geographic locations where the observations were made, or a cross-sectional study that collected data at only one point in time such as a census? If you want to analyse the effect of an intervention or treatment what would be best: a randomized controlled trial, or would a non-randomized quasi-experimental study design be better such as an interrupted time-series where samples from the same population are taken before and after the intervention? To make a safety case certain types of study design are important to control for potential biases and confounding variables, however when you come to do your literature search you may find that to answer your question certain types of study design were not possible for reasons including ethics, time, cost, etc. If you can't find the data you're looking for in journal articles, might it exist in other public or private commercial sources? Try searching some of the supplementary data sources described on the next page of this guide such as clinical trials registers. Consider contacting authors you know have worked on this, identify them by doing an author search in Pubmed or an affilitation search in Scopus. Ask your professional colleagues or post questions to internet forums or listservs for professional societies and working groups for people working on this topic.
HOW TO REPORT YOUR SEARCH STRATEGY
In the final report in the methods section the PRISMA checklist Item 6 the PICOS representing the types of study you were looking for and study eligibility criteria should be reported as: