Extract an Evidence Table After Screening, Not From Titles
Data extraction is reducing included records to inspectable fields. It is not a results section, not a meta-analysis, and not something you complete from a title when full text is missing.
Screening decides which records enter the evidence base. Extraction decides what you are allowed to say about them. If you write results from titles, or fill design and outcomes because the abstract “probably” contains them, you are not extracting. You are decorating.
LitSynth’s evidence table is a card per included record: who was studied, what was done, compared with what, on which outcomes, at what follow-up, with which main estimate, plus design and limitations. The citable page is data extraction methods. This essay is how to use that table without turning it into fake certainty.
The method, in one paragraph
Data extraction was performed on records included after title and abstract screening. For each included item, study design, population, intervention or exposure, comparator, outcomes, follow-up, main estimates, and author-reported limitations were recorded when the available source supported them. Failed or source-insufficient rows were retained rather than silently dropped. Dual independent extraction and full-text retrieval for every record were not completed inside the software unless separately described.
If you did extract from full text, say so. If you only had abstracts, say that too. Missingness is a method, not an embarrassment.
Extraction starts after screening, not beside it
Do not extract the whole retrieved set “in case.” That mixes ineligible records into the table and then tempts you to discuss them as evidence. The screening essay is the gate. Appraisal and the Research Gaps Matrix come after the table exists.
Order is not decoration. A gap derived from titles is a guess. A RoB judgement derived from titles is theatre. Both essays later in this series assume you already have extracted fields.
A minimum card is reconstructable PICO plus honesty fields
A row you cannot reconstruct is not an evidence table. At minimum:
- population — who was actually enrolled, not who the review wished for;
- intervention or exposure;
- comparator, or an explicit note that there was none;
- outcomes and follow-up;
- study design and a conservative evidence level;
- main estimate only when the source reports one;
- limitations the paper itself gives you a basis for.
LitSynth’s card is built around those fields. Extra columns are fine. Empty required columns with invented numbers are not.
Failed rows are part of the method
Full text will be missing. Abstracts will omit the comparator. PDFs will not parse. The correct behaviour is to mark the card failed or source-insufficient and keep it. Deleting failed rows makes the table look complete and makes the review irreproducible.
Two rules that save you in peer review:
- Do not complete design and outcome fields from the title alone and then discuss them as extracted evidence.
- Do not treat extracted effect sizes as pooled estimates. A column of numbers is not a meta-analysis. You still need a separate analysis if you want a summary effect.
The evidence matrix is an inspection layer, not the results section
LitSynth can roll cards into themes, supporting versus conflicting items, and common limitations. Use that to see the shape of the set. Then write the narrative and cite. A matrix is not a forest plot, and it is not a substitute for synthesis.
If two cards are extensions of the same parent trial, they are not two independent rows of support. Extraction should make that visible; the matrix should not launder it.
What to report so a methods reviewer can follow you
- which fields you extracted;
- that extraction was AI-assisted and human-inspected;
- how failed extractions were handled;
- whether a second extractor checked a sample or the full set;
- whether full text informed the card, or only title/abstract.
The appraisal essay starts from these fields. If they are empty, you do not have inputs for risk of bias. You have a spreadsheet-shaped hope.