Managing Papers for Meta-Analysis:Trace Any Forest Plot Row to Its Source

For an ordinary review, a forward flow — paper → summary → manuscript — is enough. A meta-analysis is stricter: it repeatedly demands the reverse route, from a numerical result → extraction table → verification notes → the exact passage in the original.

During verification. When explaining to collaborators. When a peer reviewer asks "what's the basis for this value?" Only management that survives reverse lookup can carry a meta-analysis to the finish.

Reverse lookup means connecting information across four layers.

1. Analysis data — the values entered into your statistical software. 2. The extraction table — the record of where each value came from. 3. Judgment and verification notes — inclusion reasoning and traces of what you checked. 4. The original PDF. With links between layers — study ID, table number, page number — you can move from any layer to the next. Management usually breaks exactly where one of those links breaks.

The backbone of those links is a single ID per study. And here's the meta-analysis-specific point: think one ID per study, not one ID per paper.

The same trial may have produced several papers — main results, long-term follow-up, subgroups — and pooling them as separate studies counts the same patients twice. Group reports of the same study by registration number, authors, and study period, and assign the ID at the study level.

There are two distinct kinds of exclusion to track: papers excluded during the review (ineligible), and eligible studies excluded from pooling anyway (incompatible data format, no matching time point, and so on).

Mix them up and your PRISMA flow diagram will disagree with the number of studies in the analysis. Record "excluded from review" and "excluded from analysis" in separate fields.

A meta-analysis runs months to a year, and interruptions are inevitable. To avoid re-loading the whole project into your head each time, leave a brief "you are here" note whenever you stop.

"Extraction done for 12 of 23 studies. Next: Sato 2021. Two queries unresolved." Three lines like that are enough to restart in minutes, even a month later.

Layer 1 belongs to statistical software (R, RevMan); screening records belong to dedicated review tools or spreadsheets. Paperfy replaces neither.

Paperfy covers layers 3 and 4 — the place where the original PDF and your verification and judgment notes are linked and stored, one page per paper. Edit the summary to add notes like "extracted; verified in Table 2; possible time point mismatch," and the endpoint of every reverse lookup — the original text — sits right next to the notes that lead to it. Figures and tables are on the same page, so checking "which table this came from" is quick too.

Paperfy today is primarily a personal library, so the realistic setup is this: keep the team's official records — screening and extraction tables — in files on a shared drive, and use Paperfy notes as your personal verification layer.

Copy what the team needs from your notes into the shared table. Keeping the personal working layer and the team record layer separate keeps both clean.

There's an easy way to test whether your management holds up: pick a value from the extraction table and see if you can reach the corresponding page of the original within 3 minutes.

If you can't get there, a link is broken somewhere. Run this test on a few values before the analysis gets serious, and you'll never lose half a day at the end.