How to conduct a literature review

In recent decades, the ‘evidence-based medicine’ movement has become widely accepted and, consequently, decisions regarding what is best practice are informed by the best available evidence^[1]
. However, an ever-increasing quantity of evidence published in literature has meant that healthcare providers, researchers, and policy makers are inundated with unmanageable amounts of information, which can hinder, rather than inform, rational decision making^[2]
. It is therefore vital that healthcare professionals, including pharmacists, appreciate the importance of evaluating the available evidence when attempting to answer a clinical question^[3]
. A literature review can therefore be considered “the comprehensive study and interpretation of literature that relates to a particular topic”^[4]
.

When conducted correctly, a literature review can be viewed as more than simply a cursory overview of the literature on a given topic. Indeed, they are often viewed as a piece of research in their own right. A historical example of the value of literature reviews in informing evidence-based practice is that of streptokinase in the treatment of myocardial infarction (MI). In the 1970s, 33 separate clinical trials comparing streptokinase with a placebo for the treatment of MI had been conducted and their results published. Individually, these trials provided inconclusive evidence regarding the role of streptokinase. However, a re-analysis of their combined results clearly demonstrated the beneficial effect of streptokinase. As a result the medicine became part of the standard treatment following MI, thereby transforming care and saving lives^[2]
.

Narrative versus systematic literature reviews

Literature reviews can be broadly divided into narrative (descriptive) reviews and systematic reviews^[5]
. The most important difference between the two categories of reviews is the difference in their level of associated scientific rigour. Narrative literature reviews provide an overview of the literature, typically from the perspective of an ‘expert in the field’^[1]
. Such reviews often report findings in a format that includes a brief summary of each included study. While such reviews can be informative, particularly when there has been no previous attempt to assimilate the literature in a given area, narrative reviews are often criticised as having a high risk of various types of bias associated with them^[6]
. Narrative reviews typically rely on subjective and non-systematic methods of selecting and reviewing data. Consequently, authors may intentionally or unintentionally cite only those literature sources that reinforce the author’s preconceived hypotheses or promote their own views on a topic^[7]
.

In contrast to a narrative review, a systematic literature review aims to identify all the available evidence on a topic and appraise the quality of that evidence^[4]
. Therefore, such reviews have the ability to either answer the research question or, failing that, identify gaps in the existing literature which highlight the need for further high quality research into a specific area^[8]
. Established in 1993, the Cochrane collaboration is an international organisation that produces systematic reviews, often regarded as the gold standard of evidence, concerning the effectiveness of healthcare interventions.

The purpose of a literature review

Literature reviews may be conducted for a variety of reasons, for example, to form part of a research proposal or introductory section in an academic paper, or to inform evidence-based practice^[8]
. The original focus of evidence-based medicine concerned mainly clinical research questions, such as the streptokinase example. However, increasingly, more research is being conducted in the area of health services research, which incorporates social science perspectives with the contribution of individuals and institutions that deliver care. Pharmacy practice research (a sub-type of health services research) is concerned with investigating how and why people access pharmacy services, new roles for pharmacists, and outcomes for patients as a result of pharmacy services^[9]
. Whether conducting either a clinical pharmacy or a pharmacy practice literature review, the following basic principles outlined in this article should be followed.

Searching with a strategy

While time and resource constraints may not allow for a definitive systematic review to be conducted, those attempting to undertake a literature review should nevertheless adopt a ‘systematic approach’^[10]
. The first step in any literature review is to define the research question (i.e. the scope of the review). This will inform which studies or other forms of evidence are to be included and excluded from the literature review.

Questions that should be asked at this stage include:

• Which study designs are relevant to the review? For example, double-blind randomised controlled trials, observational studies or qualitative research.
• Is there a particular population of interest? For example, paediatrics, pregnant women or people living in care homes.
• Are there particular outcomes of interest? For example, mortality, costs or quality of life.

The literature search strategy is a fundamental component of any literature review, because errors in the search process could produce an incomplete and therefore biased evidence base for the review. Where possible, reviewers should consult with a pharmacy or healthcare librarian who can assist in developing and refining an appropriate search strategy^[11]
. The search should be conducted in more than one electronic database to help ensure all relevant papers are identified. The main medical literature databases are Medline and Embase. International Pharmaceutical Abstracts (IPA) is also an important database to consider because it indexes many pharmacy-specific journals not found in the larger databases. It should also be remembered that pharmacy literature often overlaps with other disciplines therefore consideration should be given to databases such as the Cumulative Index of Nursing and Allied Health (CINAHL), as well as condition-specific databases, for example PsycInfo.

When conducting searches on electronic databases, thought must be given to the search terms used. Several keywords can exist for each search term and all should be included in the search (e.g. older adult/elderly/aged, are all terms that could be used synonymously). Most databases also have the facility to identify all possible endings of the key terms; this is usually denoted with an asterisk. For example, by searching for ‘pharm*’, articles containing any of the following would be retrieved: pharmacist, pharmacists, pharmacy, pharmaceutical. Boolean operators (e.g. ‘and’, ‘or’, ‘not’) can also be used to combine search terms to refine the search results^[12]
.

Avoiding publication bias

It has been well documented that studies that report significant results are more likely than those reporting non-significant results to be published, cited by others and produce multiple publications, introducing what is known as ‘publication bias’. Consequently, such studies are also more likely to be identified and included in systematic reviews^[13]
.

For this reason, efforts should be made to identify all relevant literature on the review topic so the search should not be limited solely to electronic databases^[14]
. Additional search strategies include hand-checking relevant article reference lists and personal communication with experts in the field. Searching the ‘grey literature’ is of particular importance for pharmacy practice literature reviews because relevant articles written by non-academic pharmacists are often not published in traditional academic journals^[15]
.

The importance of such additional search strategies cannot be underestimated. Greenhalgh and Peacock reported that, in a literature review concerning innovations in healthcare organisations, 51% of the sources included in the eventual review were identified by hand-searching reference lists, while an additional 24% were identified through personal communications^[7]
. The search strategy, including databases searched, all search terms used, the limits of the search (for example, written in the English language, published in the last decade), the number of results retrieved from each resource and the date the search was conducted should be carefully documented. Although it can never be guaranteed that the entirety of relevant literature will be identified, conducting the search in a systematic manner will help avoid omissions, and where they do exist they can be said to be unintentional^[9]
.

Once the search has been performed, the next step is to screen the titles or abstracts (or both) of the identified articles against the predefined inclusion and exclusion criteria. If the article cannot be excluded on the basis of the information contained in the abstract, efforts must be made to access the full paper in order to reach a decision. Conducting a literature review can be a time-consuming process and the time taken to complete is directly related to number of citations identified by the initial search. Allen et al.^[16]
calculated that it would take a reviewer more than 1,000 hours to complete a literature review involving the screening of 2,500 articles. For this reason, it is worth considering the use of free citation management software, such as Refworks, Mendeley and Endnote, which can be used to manage references retrieved from the literature search.

Drawing conclusions

Studies identified as part of a literature review may report contradictory findings and the quality of the individual studies will have a direct impact on the overall findings of the literature review. When assimilating the evidence identified, the reviewer should attempt to assess the quality of the evidence provided by the individual studies to arrive at valid conclusions^[6]
. It is recommended that when conducting a literature review, the author should make a judgement of the quality of the evidence provided within the included studies or articles using appropriate assessment guidelines^[17]
. For example, the Cochrane Collaboration’s risk-of-bias tool attempts to aid quality assessment in terms of assessing potential sources of bias in studies.

In an effort to draw quantitative conclusions within a systematic review, data from individual studies may be pooled quantitatively and reanalysed. A meta-analysis is a statistical method used to combine the outcomes of individual studies to produce data with more power than the individual studies. By pooling the results of individual studies, the sample size is effectively increased, thereby increasing the statistical power of the analysis, which in turns narrows the confidence interval around the effect size, with the result that the overall estimate of effect is more robust^[9]
. Individual studies included in the analysis are assigned weights, with greater weight assigned to studies with larger sample sizes. The results of a meta-analysis are often displayed graphically in what is known as a “forest plot”^[18]
.