01Choose the unit represented by a row

Cochrane distinguishes a study from the reports describing it. One study may produce several papers or other records, and those reports need to be linked rather than counted as separate studies.Cochrane Handbook — Chapter 5: Collecting data (opens in a new tab)

For a compact overview, one row per study can work. For a detailed extraction, one row per outcome or comparison may be necessary. Whichever structure you use, retain a study identifier so that multiple rows do not become an inflated study count.

In an original example, Study A has a conference abstract, a full paper and a correction. These are three source records for one study. The correction may change a value while the abstract preserves useful history; none creates another independent experiment.

Do not link records solely because their authors or peptide names match. Document the evidence for the connection, such as a shared registration identifier, sample description or explicit reference to the earlier report.

02Keep the fields that change interpretation

Add columns only when they serve the question. An evidence table about binding measurements may need the receptor and affinity measure. A table about reporting quality may instead need whether allocation and masking were described.

Avoid a single generic dose column if the underlying papers describe different exposure concepts. Keep the reported basis and context together; an experimental concentration and an administered amount are not interchangeable table entries.

Store exact extracted values separately from any transformed values you calculate. Record the transformation and its assumptions so that a reader can reconstruct the path back to the source.

03Define missing and conflicting information

Use a small data dictionary. For example, not reported means the checked source did not provide the detail; not applicable means the field does not fit the design; unclear means information exists but cannot be interpreted confidently.

Cochrane’s data-collection guidance discusses linking multiple reports and deciding how to handle conflicting information. A transparent extraction process records the source used rather than silently selecting whichever value suits the summary.Cochrane Handbook — Chapter 5: Collecting data (opens in a new tab)

Suppose a figure caption gives n = 8 while the methods describe ten samples. Enter the discrepancy and its locations. Do not decide without evidence that two samples were excluded or that one number is a typographical error.

04Check the table before using it as evidence

Trace a selection of entries back to their source locations, especially central numerical results and any manually calculated values. Check units, decimal places, direction of the comparison and whether error bars were identified correctly.

Review the study identifiers for duplicate counting. Then examine whether one paper contributes several correlated comparisons that your prose might otherwise describe as several independent confirmations.

Keep interpretation in a separate field or accompanying paragraph. The extracted observation might be a change in a fluorescent signal; calling it improved biological function is an additional claim that needs its own support.

Finally, date the extraction and record the source version. When a correction or new report appears, this structure allows a precise update instead of rebuilding the table from memory.