Interactive · Prism research corpus
What the evidence actually links
to a disease
Prism extracted 55,545,265 statements linking a condition to a factor — a cause, a treatment, a protective habit, a mechanism — from 25,990,000 research and news documents, and classified every one by study design and evidence strength. This page lets you walk that record: pick a condition, see what is linked to it, then click any factor to make it the centre and see every other condition it touches.
Ranking is by evidence strength, not by volume. Those two answers are often completely different, and the page shows you where they part company.
14,862 conditions and 14,879 factors are explorable — each carries at least 120 extracted claims including 8 at high evidence strength.
Sized by claims at high or very high evidence strength. A dashed ring marks a link whose claims since 2023 outnumber 2019–2022 by more than two to one.
Written about most, supported least
Links with the most coverage behind the least support: the share of each link's claims that come from a study strong enough to carry it. A low share is not a verdict that something does not work — it means the volume of writing about it is not evidence that it does.
| Linked term | Claims | High-strength | Share |
|---|
The full record, lane by lane
Every linked term the corpus records, ordered by the measure selected above. Select any of them to make it the centre of the graph.
What kind of study the claims come from
Every source document was classified by study design. Most of the corpus is news and general reporting that states no design at all, which is why the strongest tiers are small even for heavily researched conditions.
Attention over time
How to read this
- What a claim is
- One statement, extracted from one document, linking a condition to a factor with a relation (cause, risk factor, treatment, protective, mechanism, association) and a direction. It is a record that something was stated, not a finding this page endorses.
- Evidence strength
- Each source document was classified by study design — randomised trial, other clinical, observational, review or meta-analysis, preclinical, or unspecified — and given a strength grade weighing design, sample size and replication. "High-strength" means high or very high. It describes the design behind a claim, not a clinical guideline.
- Why evidence, not volume
- The corpus includes heavily syndicated health content. One supplement carries 595,402 claims across four conditions and 85 at high strength. Ranked by volume it leads; ranked by evidence it does not appear. Both orderings are available so the difference stays visible.
- Time
- Counts run to 2026, which is a partial year and is always drawn dashed. The trend is a share of all claims recorded that year, because the corpus roughly doubles between 2019 and 2020 and a raw count would read that as a research surge.
- What is not merged
- Spellings that differ only in typography or a self-initialism are merged. Terms that mean the same thing in different words — vitamin D and cholecalciferol, or type 2 diabetes and type 2 diabetes mellitus — are not, so a subject's claims can be split across more than one entry. Check neighbouring spellings before reading a total as complete.
- What this is not
- Not medical advice, not a treatment recommendation, and not a systematic review. It is a map of what a very large body of text asserts, with the strength of the sources behind each assertion attached.
Corpus built 2026-09-18. 55,545,265 extracted claims · 25,990,000 documents · 25,369 explorable subjects · 928,349 condition–factor links.