How a Systematic Review Actually Works: The Systematic Review Process Our Team Follows, Step by Step

A systematic review answers a defined question by finding, selecting, appraising and synthesising all eligible studies using methods set out in advance [1]. That pre-specification is what separates it from a narrative review: anyone could repeat the search and the selection and reach the same set of studies.

The steps themselves are well documented. Where reviews fail is in how each step is carried out, so the table below pairs each step with the failure our team checks for before moving on.

We turn the topic into a question using PICO (population, intervention, comparison, outcome) for intervention reviews, or another framework when the question is about diagnosis, prognosis or experience [1]. The question then drives the review protocol, which we write to the PRISMA-P items: eligibility criteria, information sources, search strategy, screening, data extraction, risk-of-bias assessment and planned synthesis [2].

Where the review is eligible, we register the protocol in PROSPERO, ideally before screening starts. Eligibility criteria are written with worked examples, such as “trials in adults; trials that include children are excluded unless adult data are reported separately”, so two reviewers apply them the same way.

Our information specialist combines subject headings and free-text synonyms with Boolean operators for each database, and adds trial registries and reference checking where the topic needs them [1]. A second specialist then reviews the strategy against the PRESS 2015 guideline, which covers how the question was translated, the Boolean and proximity operators, subject headings, text words, spelling and syntax, and limits [4]. The full strategies are reported to PRISMA-S so the search can be reproduced [3].

After removing duplicates, two reviewers screen titles and abstracts independently, then full texts, with disagreements resolved by discussion or a third reviewer [1]. Each full text we exclude gets a recorded reason, which feeds the PRISMA flow diagram.

Data extraction uses a form piloted on two or three included studies before full use. It records study design, participants, interventions or exposures, comparators, outcomes and the data needed for synthesis, and a second reviewer checks the extraction [1].

Risk of bias is assessed per study with a tool that fits the design, for example RoB 2 for randomised trials and ROBINS-I for non-randomised studies of interventions [1]. For overviews that include other reviews, ROBIS assesses bias at the review level [5]. Our risk of bias guide covers tool choice in more depth. The judgements shape how much weight findings carry; they do not automatically exclude a study.

A meta-analysis is only run when studies are similar enough in population, intervention, outcome and design for a pooled estimate to mean something [1]. When they are not, we synthesise without pooling and report it with the SWiM guideline, which sets out how studies were grouped, which standardised metric was used and how certainty was judged [6]. Narrative synthesis is a method, not a fallback; done properly it is as transparent as a forest plot.

The final manuscript follows the 27-item PRISMA 2020 checklist and includes a flow diagram showing records identified, screened, excluded with reasons and included [7]. PRISMA tells readers what was done; it does not tell you how to do it, which is why the methods above come first.