PRISMA 2020 vs the 2009 flow diagram

PRISMA 2020 replaced the 2009 statement and redrew the flow diagram. Most journals in health research now expect the 2020 version. Here is what actually changed and what it means for your record keeping.

Registers are counted separately from databases

The 2009 diagram had one bucket for database searching. PRISMA 2020 splits it, so trial registries such as ClinicalTrials.gov or the WHO ICTRP are counted on their own line rather than lumped in with MEDLINE and Embase. If your search log does not distinguish them, you will need to go back and separate the counts.

Records removed before screening are broken out

The old diagram collapsed everything before screening into deduplication. PRISMA 2020 asks for three numbers: duplicates removed, records rejected by automation tools, and records removed for other reasons. The automation line was added because screening classifiers became common; zero is a normal answer if you did not use one.

Studies found by other methods get their own column

In 2009, studies found through citation chasing or hand-searching were folded into the same funnel as database results, which never described what actually happened. PRISMA 2020 gives them a parallel column that skips the title-and-abstract screening step, because you assess those reports directly.

Reasons for exclusion are required

Reporting why full-text reports were excluded, with counts per reason, is now expected rather than optional. If you are reconstructing an older review, this is usually the hardest part to recover, which is a good argument for logging reasons as you screen rather than afterwards.

Studies and reports are distinguished

PRISMA 2020 is careful about the difference between a study and a report of that study. One trial published across a main paper and two secondary analyses is one study and three reports. The diagram asks for both counts at the included stage.

If your review started under PRISMA 2009

You can usually still produce a 2020 diagram, since the extra fields are mostly splits of numbers you already have. The two that commonly need rework are separating registers from databases, and recovering exclusion reasons. If a number genuinely cannot be reconstructed, say so in the methods rather than inventing one; reviewers accept a documented gap far more readily than a figure that does not add up.

Open the generator to build a 2020 diagram, or read which of the four templates you need.