From material to a reviewable table
How to build a literature evidence table: link outcome definitions and results to sources
Connect each field to its source before comparing studies. A title, abstract and “significant result” are insufficient: identically named outcomes may use different time origins, coverage and analysis samples.
For readers with candidate papers who need study characteristics and outcome definitions. Selection, citation identity and evidence extraction are separate tasks; this guide does not replace a systematic-review protocol. Supply authorized full reports, supplements or clearly labelled abstracts, never identifiable patient records or confidential material.
Request a field dictionary, report–study map, field-level evidence table and discrepancy list. The excerpt below is synthetic teaching and the task is suggested: no model execution, real-study extraction or customer outcome occurred.
What the table should retain
Keep provenance for every field and distinguish interpretation from source facts.
| Field group | Record |
|---|---|
| Provenance and identity | Report ID, study ID or unresolved link, citation/version and access date; section, table or physical PDF page for each field |
| Design and comparison | Population, eligibility, design, exposure/intervention and reference group; separate source facts from interpretation |
| Outcomes and results | Original definition, time origin/window, coverage, analysis denominator, effect measure/unit/interval and adjustment variables |
| Missingness and review | Separate not reported, not applicable and cannot tell; retain extractor, date, disagreements and decision rationale |
Make an evidence row reopen the same place in the PDF
Add a reviewable locator to the evidence table: report ID, file version and source, physical PDF page counted from 1, printed page label if present, section or table, a necessary short verbatim excerpt, a separate interpretation and review status. A physical page is its position in the file, not necessarily the printed number. “Page 3” alone is ambiguous. This is a practical suggestion, not a universal reporting standard or a model run.
Confirm permission to obtain, upload and quote the material; record the actual version and access date. Free reading in PMC does not make every article freely redistributable. Check the article licence and third-party images separately. Do not bypass paywalls, batch-download full texts from the main PMC site or publish restricted full texts. Label abstract-only work and unread supplements honestly.
Check that text in the original PDF can be selected and searched reliably. Locate relevant methods, results and limitations, and retain only necessary short excerpts with context. Keep author text, your interpretation and unknowns separate. Flag scanned pages, broken reading order or unreadable tables for human inspection; do not have AI invent numbers from a caption. Locating a sentence is not full-text verification or proof of study quality.
Self-written synthetic locator: report A / file v1 / physical page 5 / printed page 3 / Methods; excerpt: “Readmission to this hospital within 30 days after discharge was recorded.” Separately: “This sentence does not establish outside-hospital coverage or an effect estimate.” The numbers 5 and 3 illustrate a locator, not a universal PDF offset. This is not a real paper or AI output.
Suggested complete request: “Use only my authorised [file/version and actually readable scope] to organise [fields]. For each field return report ID, physical PDF page counted from 1, printed label, section/table, a necessary short verbatim excerpt, a separate interpretation and open questions. Without supporting text, mark unknown; do not invent numbers or turn not reported into 0. Separate abstract, main text and supplements. Stop the affected field and explain unreliable text or pagination; do not claim full-text verification, independent human double extraction or a meta-analysis.”
Reopen the same file at the listed physical page and check the excerpt, context, numbers, units, denominator and strength of claim. In Zotero, an annotation’s “Show on Page” can return to its context. Its documentation warns that external page deletion, reordering or rotation can affect annotation placement. Recheck affected locations after a version change; an old annotation link does not verify a new version.
From material to a reviewable table
Specify the question and pilot fields
Define the question and outcomes, trial the form on a few reports and revise instructions. Keep versions; do not choose which result to extract because it is significant.
Link reports without guessing study independence
Check registration, setting, recruitment period and participants. One study may have several reports, and one report may cover several studies. Similar titles do not establish identity; keep uncertain links unresolved.
Preserve original values and locations
Record the source meaning for each field, with interpretation separate. Keep result-specific denominators, time points, scales and reference groups. Missing is not zero; an incomplete abstract is not verified full text. Do not silently convert values or select favorable results.
Resolve discrepancies before synthesis
Retain conflicting abstract, text, table or report versions and apply prespecified decisions with reasons. Cochrane intervention reviews require at least two people to independently extract outcome data; two AI answers are not duplicate human extraction.
Synthetic output: two “30-day” outcomes differ
Teaching input A: “Readmission to this hospital within 30 days after discharge.” Input B: “Readmission within 30 days after surgery, including outside-hospital records.” No study identity, sample size or effect is supplied.
| Report/source | Origin and window | Coverage | Effect and denominator | Review decision |
|---|---|---|---|---|
| A / teaching input A | 30 days after discharge | This hospital | Not supplied | Do not equate the outcomes |
| B / teaching input B | 30 days after surgery | Includes other hospitals | Not supplied | Study linkage and comparability unresolved |
A defensible output is “different origins and coverage; inspect methods before judging comparability,” not “different treatment effects.” These teaching rows represent neither real papers, independent-study counts nor analysis results.
Adapt this suggested task
“Use my authorized reports and supplements to pilot an evidence table for the supplied question. Give report IDs, study-linkage evidence, source locations, source facts and separate interpretation. Retain design, population, comparison, outcome definition/window, denominator, effect name/unit/interval and adjustment. Distinguish not reported, not applicable and cannot tell. List conflicts, missing items and human decisions; invent no data, run no statistics and infer no causation. Return field instructions, linkage map, evidence table and gap list for review.”
Check the delivered draft
Return to the obtained sources and actual review records.
- Can key fields be located in material actually obtained?
- Are multiple reports linked without hiding uncertain relationships?
- Are windows, denominators, references, units and adjustment retained?
- Are discrepancies, missing items and actual human review traceable?
A completed table is not a risk-of-bias assessment, GRADE certainty judgement, Summary of findings table or approval for meta-analysis. Pooling still needs an appropriate protocol and methodological assessment. Do not make association causal or give clinical advice.
Questions
Only an abstract?
Build an explicitly abstract-level preliminary table and leave missing fields unresolved. Do not claim full-text extraction.
Does CSV or Excel export mean results can be pooled?
No. Format establishes neither independent studies, comparable outcomes nor reliable extraction.
Verified sources
The official sources below were checked on 2026-10-15. Access conditions and update dates are rechecked before publication.
- Cochrane Handbook chapter 5: collecting data
- Cochrane Handbook chapter 4: searching and selecting
- PMC copyright and licence notice
Read 2026-10-20: distinguish free access, reuse permission and systematic-download restrictions.
- Zotero PDF reader and annotation locations
Read 2026-10-20: returning to an annotation and limits after external page changes; no RigorMed integration claim.
Start with materials and a deliverable
Describe your evidence-organization task in RigorMed
Bring authorized material, the question and fields to review; adapt the task above. Confirm available capability and credits before execution, then check actual deliverables. No guarantee covers arbitrary papers, automated dual extraction or a completed meta-analysis.
