Skip to content
About Evidentia

Interactive demonstration · no sign-up

Is your next study necessary?

Explore three real evidence sets compiled from PubMed. Include or exclude studies, add a hypothetical next study, and see how Evidentia’s model measures support, uncertainty and expected information gain.

Step 1

Start with a question

A scientific claim is broken into its core elements, so that every study can be judged against the same question.

“Vitamin D supplementation reduces the risk of acute respiratory tract infections.”

Population
Children and adults
Intervention
Vitamin D supplementation
Outcome
Acute respiratory tract infection

Step 2

Listen to the studies

84 PubMed records were screened; 40 did not directly test the claim and are not counted. Each remaining study supports or contradicts the claim. Untick a study to see how the picture changes.

Step 3

See the bigger picture

Supporting and contradicting studies update a uniform prior into a posterior for the study-support rate, with its probability above one half and a 95% credible interval.

Contradiction-dominant pattern

Beta(1,1) → Beta(14,32)

13 support · 31 contradict · N = 44

p · study-support rate

P(p > 0.5)

0.3%

95% credible interval

18.2%–44.3%

Posterior mean

30.4%

Bayes factor BF₁₀

0.00332

Posterior odds ÷ prior odds for p > 0.5

The curve describes plausible values of the study-support rate p. The dashed line is the uniform prior and the shaded band the 95% equal-tailed interval. It is not the probability that the claim is true, and each informative study counts once regardless of size or quality.

Step 4

Is the next study necessary?

Expected information gain estimates how much one more eligible, informative study would change what we know, averaged over whether it supports or contradicts.

Expected information gain · one more informative study

0.0107 nats

REIG5.6%

Relative to the maximum for one study, ln(2) − ½ ≈ 0.1931 nats, reached with no prior evidence.

REIG is descriptive. It is not a measure of certainty or a rule to conduct research; a research recommendation also requires source review, substantive importance, feasibility and ethics.

If the next study supports, P(p > 0.5) becomes0.3% → 0.6%
If the next study contradicts, P(p > 0.5) becomes0.3% → 0.2%
Predictive probability of a supporting study30.4%
Expected absolute change in P(p > 0.5)0.14 pp
K₁ · K₀ (KL divergence per outcome)0.0245 · 0.0047

How this demonstration was built

Every study is a real PubMed record, linked by its PMID. Records were retrieved on 2026-11-18 with the search below and ordered by PubMed relevance.

SRI screened each abstract once with an automated classifier as supporting, contradicting or not informative, and quoted one verbatim sentence from the abstract. Classifications were not reviewed by Evidentia or by peer review, and this is not a systematic review. Use the PubMed links to check any study.

Statistics use Evidentia’s published model: p ∼ Beta(1,1) updated to Beta(1 + y, 1 + n); P(p > 0.5); a 95% equal-tailed credible interval; EIG = m·K₁ + (1 − m)·K₀ with m = α/(α + β) and K the KL divergence after one hypothetical outcome; REIG = EIG ÷ (ln 2 − ½).

PubMed search used(vitamin D[tiab] OR cholecalciferol[tiab]) AND supplement*[tiab] AND (respiratory tract infection*[tiab] OR acute respiratory infection*[tiab] OR upper respiratory infection*[tiab] OR influenza[tiab] OR common cold[tiab] OR pneumonia[tiab]) AND randomized controlled trial[pt] NOT (review[pt] OR meta-analysis[pt])

Run this on your own scientific claim.

In Evidentia you define the claim, add your studies and receive a traceable report.

Start with Evidentia (opens in a new tab)