Systematic Literature Review Help — PRISMA-Compliant and Reproducible

A literature review selects sources by judgement and builds an argument toward a gap. A systematic review is a method: criteria stated before searching, an exhaustive documented search, dual independent screening, formal quality appraisal and a reproducible audit trail. The first is assessed on synthesis, the second on whether someone else could repeat it and reach your numbers.

For health-related reviews it is expected, and many journals now require a registration number before they will send a review out. Outside health, PROSPERO may not accept your topic and OSF is the usual alternative. Either way the point is the same: your criteria are fixed and timestamped before you see what the searches return.

No, but it changes what you can honestly claim. The right move is documenting what you actually did, stating that the protocol was not registered in advance, and reporting it as a limitation. We will not backdate a protocol and present it as prior — that is misrepresentation rather than paperwork, and it is exactly the kind of thing that unravels badly.

You can, and you must then declare it as a limitation, because single-reviewer screening reintroduces the selection bias the method exists to remove. Many Malaysian master's candidates have no second screener available, which is a legitimate reason to consider a scoping review instead. Where we act as second reviewer, your criteria govern every decision and the contribution is acknowledged accurately.

Almost always the gap is full-text exclusions recorded without reasons, or a screening file that was overwritten. The diagram has to be built from your screening records rather than typed to look plausible — identification, duplicates removed, screened, assessed and included must reconcile arithmetically. Send your numbers and the file and we will find where it breaks.

It depends on the design of each included study, not on your field. Randomised trials need Cochrane RoB 2, observational studies suit Newcastle-Ottawa, qualitative work suits CASP or JBI, and mixed designs are handled by MMAT. If your included studies are themselves reviews, you need AMSTAR-2. Applying one tool to every study is a visible error.

Often yes, and we say so more than students expect. A scoping review suits broad mapping questions, follows PRISMA-ScR and the PCC framework, and does not require formal quality appraisal — which makes it achievable on a master's timeline without a second screener. A well-executed scoping review scores better than a systematic review that could not be completed properly.

Six to ten weeks at master's level and eight to fourteen for doctoral or publication work, with a scoping review at four to seven. Screening is the bottleneck — 1,500 records take real hours twice over with dual screening. Anyone quoting a full systematic review in ten days is describing something else.

Only if your included studies report comparable effect sizes on comparable outcomes. Pooling incomparable results produces a number that means nothing, and reviewers will say so. Where pooling is not justified, a well-argued narrative or thematic synthesis is the correct answer rather than a weaker option.

Yes, at the same rates. Sources remain mostly English-language indexed journals, so the methodological terminology — kriteria kemasukan, penyaringan tajuk dan abstrak, penilaian kualiti — has to stay consistent across every section. That consistency is where most translated reviews come apart, not the translation itself.

AI-assisted prioritisation is now built into tools like Rayyan and it can genuinely speed up title and abstract screening. What it cannot be is your second independent reviewer, because the whole point of dual screening is two people applying the criteria separately and disagreeing where they disagree. If you use any automation, PRISMA 2020 expects you to report it in your selection process — so declare it rather than quietly relying on it.

No, and this is the most dangerous place to try. A fabricated reference in a narrative review is embarrassing; a fabricated study inside a systematic review corrupts your extraction table, your appraisal scores and your PRISMA counts at once. Your included studies must come out of a documented database search you can rerun — if a study cannot be traced back to a logged search, it does not belong in the review.

Two separate declarations, and students usually miss one. Your university will have a policy on generative AI in the written thesis. Separately, PRISMA 2020 asks you to report any automation tools used in the selection and data collection stages — that is a methodological disclosure, not a misconduct question, and it belongs in your method section whether or not your faculty asks for it.

Run it. That settles the question in about a minute, and it is what your supervisor is really asking you to do. Generated strings tend to look correct while using field tags or operators the database does not accept, so they return zero results or silently ignore half the string. Paste each one into the database it was written for, record the hit count and the date, and put those in your appendix — a string with a logged result behind it stops being a suspicion.

You can draft with it, but what comes back is almost always untestable. Criteria like "studies of good methodological quality" or "recent and relevant research" read fine and cannot be applied consistently by two reviewers, which is the only thing criteria have to do. Yours need to be rules — population, design, language, date range, publication type — specific enough that two people screening separately reach the same verdict on the same record.

Not reliably, and the errors are the kind nobody notices until a viva. Models misread table values, report a sample size from the wrong subgroup, and confuse a cohort study with a cross-sectional one. Your extraction table goes into an appendix where an examiner can open any row and compare it against the original paper, so every cell has to come from someone who actually read the study.

No. Appraisal is a judgement about how a study was conducted, made domain by domain after reading the full text — whether allocation was concealed, whether confounders were handled, whether the sampling was defensible. A model given an abstract produces scores that look plausible and correspond to nothing in the paper. If your appraisal scores cannot be justified out loud, they will not survive being questioned.

Treat the two halves differently. Your discussion and background can be rewritten normally — rebuild the argument and put your own included studies back into it. Your method section cannot, because it has to match your logs exactly: the database names, the string, the hit counts, the screening numbers, the appraisal tool. Do not paraphrase those into something that reads better than it is true. Fix the prose around them and leave every figure exactly as your records say.