How to Read a PRISMA Flow Diagram in a Systematic Review

A PRISMA flow diagram is the map of how a systematic review went from thousands of database hits to the handful of studies it actually analyzed. Read it top to bottom: identification (what the search returned), screening (what survived a title and abstract pass), eligibility (what survived full-text reading against the inclusion criteria), and inclusion (what made the final synthesis). The numbers only have to do one thing to be trustworthy, and it is arithmetic: every record removed has to be accounted for, with a reason, so that the count entering each stage minus the exclusions equals the count leaving it. When those subtractions balance and the exclusion reasons are specific, you are looking at a search you can audit. When they do not, you are looking at a selection process you have to take on faith.

What the diagram is actually tracking

PRISMA stands for Preferred Reporting Items for Systematic Reviews and Meta-Analyses, and the flow diagram is one piece of a larger reporting standard updated in 2020. Its job is narrow and useful. It does not tell you whether the included studies were any good, whether the effect estimate is large, or whether the conclusion is sound. It tells you how the authors found and filtered the evidence. That makes it the first thing worth reading, because every downstream result depends on the pool of studies that survived the funnel. A flawless meta-analysis built on a selectively assembled set of studies is still a selectively assembled answer.

The diagram is a set of boxes connected by arrows, and the arrows carry counts. Records go in at the top. At each level, some are removed, and the removed ones branch off to the side with a tally. What remains flows down to the next box. The discipline the format imposes is that you should be able to trace a single number all the way down and never lose track of where records went.

Stage one: identification

The top box reports how many records the search returned, usually split by source. You will see counts from bibliographic databases (such as MEDLINE, Embase, or the Cochrane registers) and, in the 2020 version, a separate stream for records found through other methods: citation searching, reference lists, contact with experts, or registries of trials.

Two things are worth checking here. First, are multiple databases searched? A review resting on a single database has a structural blind spot, because no one index covers the whole literature. Second, is there a deduplication step, and does its count make sense? The same paper appears in several databases, so records removed before screening as duplicates is expected and healthy. What you want to see is a stated number, not a silent collapse from a big figure to a smaller one.

Stage two: screening

Screening is the title and abstract pass. A human (ideally two, working independently) reads each record's summary and decides whether it is plausibly relevant. Most records die here, and that is normal. A broad search of a common condition can return tens of thousands of hits, the vast majority of which are obviously off topic on a five second read.

The number that matters is how many were screened and how many were excluded. At this stage a bulk exclusion without individual reasons is acceptable, because you are discarding the clearly irrelevant. The signal to watch is the ratio of what enters full-text review. If a search returns fifteen thousand records and only twelve reach full-text reading, that funnel is either extraordinarily precise or quietly narrow, and the reader deserves the search string to judge which.

Stage three: eligibility

This is the stage that separates a rigorous review from a loose one. Reports that survived screening are read in full and checked against the pre-specified inclusion and exclusion criteria. Every report excluded here must come with a reason: wrong population, wrong comparator, wrong outcome, wrong study design, no usable data, and so on.

Read those reasons closely. They are the most informative numbers on the whole diagram. Specific, categorized exclusion counts tell you the criteria were applied consistently and transparently. A single lumped figure ("excluded, n = 47") with no breakdown tells you nothing and should lower your confidence, because it hides whether studies were dropped for principled reasons or for inconvenient results. The exclusion reasons are also where you check the criteria themselves. If a review of a drug's effect excludes trials that measured a relevant safety outcome, that is a decision worth understanding before you trust the summary estimate.

Stage four: inclusion

The bottom box gives the number of studies included in the qualitative synthesis and, where relevant, the number contributing to the quantitative synthesis or meta-analysis. These two can differ legitimately, because a study can be eligible but lack data in a poolable form. When they differ, the diagram should say by how much and, ideally, why.

What a suspicious diagram looks like

A few patterns should make you slow down. The clearest is arithmetic that does not close: numbers that enter a box do not equal the numbers that leave it plus those excluded. That is a bookkeeping failure at best and, at worst, a sign that records were handled off the page. Missing exclusion reasons at the eligibility stage is another, because it removes your ability to audit the most consequential decisions. Watch too for a single narrow database with no attempt at other sources, which raises the chance that relevant work simply never entered the funnel. Last, be wary of an oddly precise search that yields a tiny included set from a large base, without a search string you can inspect to see how the sieve was cut. None of these prove misconduct. Each one shifts the burden of trust from the diagram back onto your own skepticism, which is the opposite of what good reporting should do.

The habit worth building is to read the flow diagram before the forest plot. A forest plot tells you what the included studies said. The flow diagram tells you whether the right studies were included in the first place, and whether the reasons for leaving evidence out were stated plainly or left for you to guess. A search that shows its work is not automatically correct, but it is auditable, and auditable is the precondition for everything else.

This article is educational and not medical advice; for decisions about your own care, talk with your own clinician.