🧬 Aging Wiki

extracting-evidence

Properties1
aliasesreading a paper, study extraction

4 min read

SOP: extracting evidence from a primary source

When the wiki cites a paper for a non-trivial claim, create a corresponding studies/ page so the evidence layer is auditable.

When to create a study page

  • Always when the paper is the sole source for a claim on the wiki.
  • Always when extracting multiple distinct claims (so each footnote can deep-link).
  • Optional when the paper is one of many corroborating sources for a well-established fact (a bare DOI footnote is fine).

File naming

studies/{first-author-lastname}-{year}-{topic-slug}.md

Examples:

  • studies/yousefzadeh-2018-fisetin-senolytic.md
  • studies/lopez-otin-2023-hallmarks-expanding-universe.md
  • studies/calerie-2017-caloric-restriction-2yr.md (consortium / acronym OK for multi-author)

Required frontmatter

---
type: study
doi: 10.1016/j.ebiom.2018.09.015
title: "Fisetin is a senotherapeutic that extends health and lifespan"
authors: [Yousefzadeh MJ, Zhu Y, McGowan SJ,...]   # truncate after ~5
year: 2018
journal: EBioMedicine
study-design: in-vivo                                # see allowed values below
organism: mus-musculus                               # or: homo-sapiens, multi
n-subjects: 80                                       # total across groups
intervention: ["[[fisetin]]"]                        # what was tested
hallmarks-tested: ["[[cellular-senescence]]"]
human-evidence: false                                # true if any human arm
peer-reviewed: true
preregistered: false                                 # true if NCT/OSF/EU-CTR registered
key-findings: [senescent-cell-clearance, lifespan-extension]
local-pdf: null
---

Auto-populating frontmatter

Most fields can be pulled from :

  • title, year (publication_year), journal, oa_status, download_status/download_path

Two fields the archive doesn’t currently provide (as of 2026-05-04) — fetch from Crossref:

DOI="10.1016/j.ebiom.2018.09.015"
curl -s "https://api.crossref.org/works/$DOI" | jq '{
  authors: [.message.author[] | "\(.family) \(.given[0:1])"],
  abstract:.message.abstract
}'

Crossref’s abstract field is JATS-XML — strip tags before pasting. Some publishers don’t provide it; fall back to the paper itself.

The remaining fields (study-design, organism, n-subjects, intervention, hallmarks-tested, human-evidence, preregistered, key-findings) require reading the paper. They are the whole point of the study page — don’t guess.

study-design allowed values: rct | observational | cohort | case-control | cross-sectional | meta-analysis | systematic-review | in-vivo | in-vitro | in-silico | mendelian-randomization | review

What to extract into the page body

Most study pages should have these sections (omit any that don’t apply):

# {Title}
 
## TL;DR
 
One paragraph: what they did, what they found, why it matters for aging.
 
## Design
 
- Subjects: species, age, sex, n per group
- Intervention: dose, route, duration, controls
- Endpoints: primary and secondary
 
## Key results
 
- Quantitative outcomes with effect sizes and p-values
- Confidence intervals where reported
- Note any failed/null endpoints
 
## Extrapolation to humans
 
[Use the rubric from [[_extrapolation-guide]]]
 
| Dimension | Status | Notes |
|---|---|---|
| Pathway conserved in humans? | yes/partial/no/unknown |... |
| Phenotype conserved in humans? |... |... |
| Replicated in humans? |... | NCT number if applicable |
 
## Limitations
 
- Authors' acknowledged limitations
- Limitations the wiki notes (small n, single lab, conflict of interest, etc.)
 
## Cited by wiki pages
 
[Backlinks; auto-populated by Obsidian]

What NOT to do

  • Don’t paraphrase the abstract verbatim — extract structure (n, design, p-values).
  • Don’t omit null/negative findings. They’re often the most informative part.
  • Don’t trust press releases, news coverage, or science-blog summaries over the paper itself.
  • Don’t extrapolate beyond the paper’s actual claims (e.g., “this would also work in humans” if the paper didn’t test humans).

Quality flags to surface

If you notice any of the following while reading, add the corresponding tag and consider noting it in the wiki page that cites this study:

  • Single lab, never replicated → #gap/needs-replication
  • Industry-funded with author conflicts → note in “Limitations”
  • p-hacking suspected (n < 10 per group, many comparisons, no correction) → flag as low confidence
  • Retraction or correction notice → CRITICAL — update the page and notify user
  • Preprint not yet peer-reviewed → set peer-reviewed: false and qualify claims

See also

Graph View

Backlinks