A team submits a systematic review that includes 23 studies. A reviewer reruns their search in the same databases and finds 11 more eligible studies. The reason: the search used one term for the intervention and missed three other names in common use in the field. The authors are asked to redo the entire search phase.
Unlike a narrative literature review, a systematic review promises readers that all relevant evidence has been located through a transparent, repeatable process. That promise stands or falls on two stages: searching and screening.
Frame the question with PICO or an equivalent
A good search starts from a structured question. For intervention studies, PICO covers Population, Intervention, Comparison and Outcome. Qualitative reviews often use SPIDER or PICo (Population, phenomenon of Interest, Context).
Example: “Among first-year undergraduates (P), do peer mentoring programmes (I), compared with no programme (C), reduce one-year dropout (O)?” Usually only two or three elements go into the search, most often P and I. Adding the outcome tends to miss studies that report it in a table but never mention it in the abstract.
Build the string: synonyms, truncation, Boolean logic
- List synonyms for each concept. Peer mentoring may also appear as peer mentor, peer support, student mentor or buddy programme. Harvest extra terms from five to ten key papers you already know.
- Combine controlled vocabulary (MeSH in PubMed, the thesaurus in ERIC or PsycInfo) with free-text terms.
- Use truncation and wildcards: mentor* captures mentor, mentors, mentoring and mentorship.
- Join with Boolean operators: OR within a concept, AND between concepts, with brackets to group.
- Test sensitivity: the string must retrieve every known key paper. Missing one is a signal to add terms.
Record the full string for each database with the date run and number of hits. PRISMA 2020 asks for the full search of at least one database, usually in an appendix. If you can, ask a subject librarian to review your string; many university libraries offer exactly this service.
Choose sources: never rely on one database
| Source | Strong for | Notes |
|---|---|---|
| PubMed/MEDLINE | Biomedicine, public health | Free, uses MeSH |
| Scopus, Web of Science | Multidisciplinary, citation tracking | Needs institutional access |
| ERIC | Education | Includes grey literature such as reports |
| PsycInfo | Psychology | Its own controlled vocabulary |
| Google Scholar | Supplementary checks | Not fully reproducible; screen only the first pages |
| Thesis databases, trial registries | Reducing publication bias | Often forgotten |
Add backward citation searching (reference lists of included studies) and forward searching (papers that cite them). If your topic has strong regional literatures, look at regional and national journals too, because relevant studies are often missing from international indexes.
Register the protocol before screening
Register a protocol on PROSPERO for health-related reviews, or on OSF for other fields, before screening begins. It records the question, eligibility criteria, search strategy, quality appraisal and synthesis plan. Any later change must be reported with a reason. A growing number of journals ask for the registration number at submission.
Deduplicate and manage records
Four databases can easily return 3,000 records, a third of them duplicates. Export everything into Zotero or EndNote, run automatic deduplication, then check suspected pairs by hand, such as a preprint and the published version with different years. Move the clean set into a screening tool such as Rayyan or Covidence. Log the count at every step, because these become the numbers in your PRISMA diagram.
Screen in pairs, independently
Screening has two rounds: titles and abstracts, then full texts. The standard is two reviewers working independently and blind to each other’s decisions, then reconciling disagreements by discussion or a third reviewer.
- Pilot first on 50 to 100 records, compare decisions and reword any criterion people read differently.
- Be generous at title and abstract stage: when in doubt, keep it and let the full text decide.
- Record an exclusion reason at full-text stage for every paper, from a fixed list: wrong population, wrong design, no comparator, outcome not reported.
- Report agreement, for example Cohen’s kappa or percentage agreement, and how conflicts were resolved.
With limited resources, an acceptable compromise is one reviewer screening everything while a second checks a sizeable random sample. State this clearly as a limitation.
The PRISMA 2020 diagram and numbers that must add up
The PRISMA 2020 flow diagram shows records from each source, duplicates removed, records screened, records excluded at title and abstract, reports sought and not retrieved, full texts excluded with reasons, and the final number of included studies. The most common error is arithmetic that does not reconcile. Before submitting, add up every branch by hand.
Quick checklist
- Question written in PICO or an equivalent framework.
- Search retrieves every known key paper.
- Two to four field-appropriate databases, plus backward and forward searching.
- Protocol registered before screening.
- Two independent screeners, with a pilot round.
- Full-text exclusion reasons recorded for each paper.
- PRISMA numbers reconcile in every branch.
Your next move: pick the five papers you are certain must be in the review, write a first search string and see how many it retrieves. That single number tells you how much the string is missing before you spend weeks screening.
Câu hỏi thường gặp
What is the difference between a systematic review and a literature review?
A systematic review has a narrow question and reports its search strategy and eligibility criteria in full so others can reproduce it. A narrative literature review selects sources more flexibly and does not promise to find all the evidence.
How many databases should a systematic review search?
There is no universal minimum, but one database is not enough. Most reviews use two to four field-appropriate databases, plus citation searching and a search for grey literature.
Do systematic reviews need two screeners?
Independent dual screening is the recommended standard because it reduces errors and bias. If only one person screens, have a second person check a random sample and report this as a limitation.
Do I have to register my systematic review on PROSPERO?
Not every journal requires it, but prospective registration shows you did not change methods after seeing results. PROSPERO covers health-related reviews; reviews in other fields can be registered on OSF.
Can Google Scholar be the main database for a systematic review?
It is not recommended. Google Scholar does not give stable, reproducible results and does not support complex search strings. Use it as a supplementary check, such as screening the first pages of results.
