Extraction methods

Data extraction that produces a table you can check, not a summary you have to trust

Extraction is where included studies become comparable. If a field cannot be traced to a paper, it does not belong in the synthesis. The method is structured cards, retained failures, and a human pass before drafting.

Last reviewed 2026-09-23. Written to be cited in a Methods section, a protocol, or a procurement review—not as a feature tour.

Open the review workspace All methods pages

Language you can paste into Methods

Included studies were extracted into structured evidence cards covering study design, population, sample size, intervention or exposure, comparator, outcomes, reported quantitative results, limitations, and supporting quotes. Extraction was assisted by LitSynth and inspected before synthesis. Failed extractions were retained with reason codes and not silently dropped. Independent dual extraction was not performed inside the software.

Adapt the sentence if you later completed dual screening, full-text review, RoB 2, or GRADE outside LitSynth. Do not delete the boundary to sound more systematic. Last reviewed 2026-09-23.

Procedure

Do the work in this order

Step 01
Extract only papers that survived screening
Do not extract the full retrieved set “just in case.” Extraction on excluded records wastes time and contaminates the table. The included set is the sampling frame for this stage.
Step 02
Use one card per study, with fields a second reviewer could audit
LitSynth cards capture design, population, sample, intervention, comparison, outcomes, findings, effect estimates, quantitative results as reported (measure, value, CI, p-value, follow-up, quote), limitations, evidence level, and source quotes. Empty is better than invented.
Step 03
Keep failed rows in the table
Source unavailable, insufficient text, no question-aligned finding, or extraction failure should remain visible. A missing row is a decision. Silent omission is a methods error.
Step 04
Inspect before synthesis
Read the assembled evidence table. Correct fields, drop studies that should not have been included, and only then generate the draft. Synthesis cannot repair a table nobody checked.

What LitSynth keeps

Audit trail for this stage

One evidence card per included paper, or a failed card with a reason code
Study design, population, sample, intervention, comparator, outcomes
Quantitative results as reported, with quotes where present
Limitations listed on the card, not only in the narrative
An evidence matrix grouping themes, supporting vs conflicting items, and common limitations

What this is not

Boundaries that belong in print

Automated extraction is not dual independent extraction.
Reported numbers are not re-analysed or meta-analysed in-tool.
Quotes must come from the source text; unverifiable quotes should not drive claims.
Extraction of limitations is not a Cochrane risk-of-bias judgement.

Reporting checklist

If you cannot tick these in the manuscript, the software run is not yet a method. It is only a draft aid.

01List the fields extracted.
02State that extraction was AI-assisted and human-inspected.
03Report how failed extractions were handled.
04Say whether a second extractor checked a sample or the full set.
05Do not treat extracted effect sizes as pooled estimates unless you ran a separate analysis.

Questions methods committees actually ask

Which fields should a minimum evidence table contain?: Enough to reconstruct PICO-or-equivalent plus design and limitations: who was studied, what was done, compared with what, on which outcomes, at what follow-up, with which main estimate, and which validity problems the paper itself reports. LitSynth’s card is built around those fields.
Can I use the evidence matrix as the results section?: Use it as an inspection layer: themes, supporting versus conflicting items, and common limitations. The results narrative still has to be written and cited. A matrix is not a substitute for synthesis, and it is not a forest plot.
What if full text is missing?: Mark the card as failed or source-insufficient. Do not complete design and outcome fields from the title alone and then discuss them as extracted evidence.

Screening methods

Risk-of-bias and appraisal methods

Extraction product page Screening product page Citation audit Evidence-table extraction essay

Data extraction that produces a table you can check, not a summary you have to trust

Last reviewed 2026-09-23. Written to be cited in a Methods section, a protocol, or a procurement review—not as a feature tour.

Do the work in this order

Step 01

Extract only papers that survived screening

Do not extract the full retrieved set “just in case.” Extraction on excluded records wastes time and contaminates the table. The included set is the sampling frame for this stage.

Step 02

Use one card per study, with fields a second reviewer could audit

LitSynth cards capture design, population, sample, intervention, comparison, outcomes, findings, effect estimates, quantitative results as reported (measure, value, CI, p-value, follow-up, quote), limitations, evidence level, and source quotes. Empty is better than invented.

Step 03

Keep failed rows in the table

Source unavailable, insufficient text, no question-aligned finding, or extraction failure should remain visible. A missing row is a decision. Silent omission is a methods error.

Step 04

Inspect before synthesis

Read the assembled evidence table. Correct fields, drop studies that should not have been included, and only then generate the draft. Synthesis cannot repair a table nobody checked.

Audit trail for this stage

One evidence card per included paper, or a failed card with a reason code

Study design, population, sample, intervention, comparator, outcomes

Quantitative results as reported, with quotes where present

Limitations listed on the card, not only in the narrative

An evidence matrix grouping themes, supporting vs conflicting items, and common limitations

Reporting checklist

If you cannot tick these in the manuscript, the software run is not yet a method. It is only a draft aid.

01List the fields extracted.

02State that extraction was AI-assisted and human-inspected.

03Report how failed extractions were handled.

04Say whether a second extractor checked a sample or the full set.

05Do not treat extracted effect sizes as pooled estimates unless you ran a separate analysis.

Questions methods committees actually ask

Which fields should a minimum evidence table contain?

Enough to reconstruct PICO-or-equivalent plus design and limitations: who was studied, what was done, compared with what, on which outcomes, at what follow-up, with which main estimate, and which validity problems the paper itself reports. LitSynth’s card is built around those fields.

Can I use the evidence matrix as the results section?

Use it as an inspection layer: themes, supporting versus conflicting items, and common limitations. The results narrative still has to be written and cited. A matrix is not a substitute for synthesis, and it is not a forest plot.

What if full text is missing?

Mark the card as failed or source-insufficient. Do not complete design and outcome fields from the title alone and then discuss them as extracted evidence.