1. Clinically relevant question
A systematic review should always start with a clearly defined and focused question.
2. Protocol
All the methods for the systematic review should be clearly defined before the review is started. The protocol should define what studies will be included in the review – defining the patients, interventions and outcomes of interest. The methods of data synthesis and analysis should be defined in the protocol.
3. Search for studies
A good systematic review will have a comprehensive search strategy. The aim of the search should be to identify all studies relevant to the question. In most cases a good search will aim to find all relevant studies in the world, regardless of language of publication, year of publication, country of origin etc. Searching can include electronic databases (e.g. MEDLINE), databases of ongoing studies, hand-searching relevant journal publications and contacting known experts.
4. Select relevant studies
A search strategy is likely to bring up many false ‘hits’. An important part of the systematic review process involves determining which studies should be included. Studies are commonly selected based on the methods (e.g. is it a RCT?), participants (e.g. are they stroke patients?), intervention investigated (e.g. is the intervention a stroke unit?) and outcomes (e.g. do they assess death and disability?). A good systematic review will use 2 reviewers, who will both apply the selection criteria independently then come together to see if they agree, ensuring that the biases of one reviewer don’t influence the studies that are selected.
5. Assess study quality
A good systematic review should clearly document the quality of the included studies. ‘Risk of bias’ refers to whether a study is at risk of containing errors or inaccuracies as a result of the methods used. In RCTs, the main risks of bias arise from systematic differences in the groups that are compared (selection bias), the care that is provided (performance bias), withdrawals or exclusions of people entered into a study (attrition bias) or how outcomes are assessed (detection bias). Reviews of studies may also be particularly affected by reporting bias, where a biased subset of all the relevant data is available.
6. Synthesise data
A systematic review of RCTs will take the results from the individual RCTs and summarise these in an attempt to answer the systematic review question. Statistical methods (meta-analysis) may or may not be used to combine and analyse the results of the included studies. If meta-analysis is carried out this can give an estimate of the size of the effect of the intervention (when compared to no treatment, a control or placebo intervention or other intervention).
7. Implications for clinical practice
A good systematic review should conclude by providing a clear statement of the clinical implications arising from the systematic review, in relation to the question which was asked. Recommendations for future research should also be defined.
Page last reviewed: 14 May 2020
