How to write a systematic review, step by step

1. Frame an answerable question

A systematic review lives or dies by its question. "What is known about remote work?" is a topic; "Does remote work increase self-reported productivity in knowledge workers compared to office work?" is a question a review can answer. Most teams use a structuring device: PICO (Population, Intervention, Comparison, Outcome) for interventions, SPIDER for qualitative evidence, or PEO for exposures.

Write the question down before you search, and register it. For health research that means PROSPERO; in other fields, a preregistration on OSF does the job. Registration is what separates a systematic review from a narrative review with a methods section — it commits you to criteria before you have seen the results.

2. Build a search strategy you can defend

Reviewers will ask three things about your search: which databases, which query strings, and when. Search at least two databases (e.g. Scopus and Web of Science, or OpenAlex which is free and covers both well), plus one field-specific source. Derive the query from your question: synonyms for each concept joined with OR, concepts joined with AND.

Keep a log: the exact string, the database, the date, and the hit count. This becomes a table in your appendix and the "Identification" box of your PRISMA diagram. If you tweak the query later, log the new version too — undocumented query drift is the most common methodological hole in submitted reviews.

3. Screen in two passes with explicit criteria

Define inclusion and exclusion criteria before screening: study designs, populations, languages, date ranges, publication types. Then screen twice — first titles and abstracts, then full texts of survivors. Record a reason for every full-text exclusion; PRISMA requires it.

Two independent screeners are the gold standard. Where that is not feasible, a second screener on a random 10–20% sample with a reported agreement rate (Cohen’s kappa) is a defensible compromise most journals accept.

• Free PRISMA flow diagram generator
• Cohen’s kappa calculator

4. Extract data into a fixed template

Build the extraction sheet before reading: citation, design, sample, measures, effect estimates, limitations, funding. Pilot it on three papers and revise — every review team discovers a missing column in the pilot, and it is much cheaper to discover it there.

If you plan a meta-analysis, extract raw statistics (means, SDs, ns, correlations) rather than only computed effect sizes, so you can convert consistently later. Where papers report incompletely, note what you emailed authors for and what you received.

• Effect size calculator (d, r, odds ratio)

5. Write results before conclusions

Structure the manuscript around the reporting standard for your field — PRISMA 2020 for most reviews, MOOSE for observational meta-analyses, Cochrane conventions in health. Write the results section first, directly from the extraction sheet, before you allow yourself an interpretation: it keeps the discussion honest.

The flow diagram, the criteria table, and the search-log appendix are not decoration — they are the parts editors check first. Finish those before polishing prose.

• Reporting guidelines cheat sheet (PRISMA, CONSORT, MOOSE…)

Where Kahubi fits

Kahubi’s systematic review flow runs this exact pipeline in one place: define the question and criteria, search OpenAlex and Semantic Scholar, screen with AI-suggested include/exclude calls you confirm or overturn, auto-fetch open-access PDFs, and draft the review with every included paper as grounded context. The PRISMA numbers fall out of the screening log automatically.