A systematic review sets out its question, its search strategy and its inclusion rules before looking at any results, then applies them to everything it finds.
A systematic review sets out its question, its search strategy and its inclusion rules before looking at any results, then applies them to everything it finds. The protocol is the part that matters: it is what stops the authors from selecting the studies that happen to agree with them. Without one, a review is a reading list with a conclusion attached.
A narrative review is an expert summarising a field as they see it. It can be excellent, and it is not reproducible: a second expert reading the same literature may describe it differently, and there is no way to adjudicate. A systematic review publishes its method so the search can be run again by someone who disagrees.
A good one names the databases, the date range and the exact search terms used. If a review does not publish these, a reader cannot tell whether it missed half the field, and its conclusion is an opinion with a bibliography. Reproducing the search is the single most informative check available to a non-specialist.
Every review discards studies, and the reasons are where bias can still arrive after a careful search. A review that excluded forty papers should say why, grouped into categories, with counts. Vague exclusion criteria are the most common weakness in otherwise rigorous work, and they are usually visible in the flow diagram rather than the text.
Because it reduces the chance that one unusual result drives the conclusion. A single study can be wrong by accident; a dozen pointing the same way is harder to explain away. That is the whole of the claim being made for the format, and it is a claim about reliability rather than certainty.
A review cannot be better than the studies inside it. Pooling weak trials produces a confident-looking summary of weak evidence, which is more misleading than the weak trials were individually, because the format carries authority the inputs do not deserve.
Three things are visible without reading the body. Whether the protocol was registered in advance, and where. Whether the search strategy is reproduced in full. And whether a risk-of-bias assessment was applied to each included study rather than asserted for the set as a whole. A review carrying all three has made itself checkable.
A protocol registered after the authors had seen their results is not a protocol. Registration platforms timestamp submissions precisely so a reader can tell, and the interval between registration and publication is worth a glance. Reviews that changed their primary outcome partway through sometimes record it in an amendments log, and the honest ones do.
The claims on this page about why one well-made review outranks a dozen individual studies can be checked against sources outside this site: