Evidence appraisal · rigor, not a recap
How solid is the evidence behind this result?
TrialReviewer doesn’t just recap a paper — it appraises the evidence like a clinical epidemiologist: recomputes absolute benefit, surfaces methodology flaws, grades quality, and shows whether the evidence supports the claim. The decision stays yours.
Includes ACCORD, SPRINT, and RECOVERY examples
The ACCORD trial randomized 10,251 patients with type 2 diabetes at high CV risk to intensive glucose lowering (HbA1c<6.0%) or standard therapy… the intensive arm had higher all-cause mortality (HR 1.22, 95% CI 1.01–1.46) and was terminated early.
| Endpoint | Direction |
|---|---|
| Nonfatal MI | Favors intensive |
| CV death | Favors standard |
| All-cause mortality | HR 1.22 · favors standard |
| Severe hypoglycemia | ~3× excess |
Intensive control did not reduce MACE; all-cause mortality rose (HR 1.22) with ~3× more severe hypoglycemia — the evidence does not support tighter glucose targets in this population.
Every threshold judgment carries an inline [Author Year] citation, e.g. [Walsh et al. 2014].
Why not just use ChatGPT?
Same number, two very different treatments. Say a trial reports a 25% relative risk reduction:
“The trial reports a 25% relative risk reduction, suggesting the intervention may reduce the outcome compared with control.”
But it does not check absolute benefit, harms, or whether the result supports the clinical claim.
- Absolute risk reduction ARR 1.6% (not 25%)
- Number needed to treat NNT 61
- All-cause mortality actually increased
- Appraisal confidence: B
Appraisal: the evidence does not support the claimed benefit.
What it does
Doesn't parrot relative risk (RRR) — it reconstructs absolute risk reduction and number needed to treat. Absolute benefit is what the clinic acts on.
Selective reporting, immortal time bias, composite-endpoint decomposition — checked one by one, with reverse signals never ignored.
Starts from the RCT / observational baseline and states, criterion by criterion, why and how far it downgrades (GRADE-informed).
An A/B/C/D appraisal-confidence rating plus a one-line Bottom Line — the conclusion professionals read first.
The appraisal rests on a framework
Every appraisal is grounded in established evidence-appraisal methods and 60+ core references — not the model's intuition.
Downgrades criterion by criterion from a high-quality start.
A fatal flaw sends the rating straight to D.
An explicit checklist for whether the evidence clears the bar to change practice.
Benefit and harm quantified side by side — net value.
How stable the result is — how few events would flip significance.
Whether the improvement clears a clinically meaningful threshold.
Sources: Nuovo 2002 · Walsh 2014 · Balshem 2011 · Schulz 2010, and 60+ more.
Coverage
Two modes — pick how you read
Assess whether the evidence supports changing practice. A/B/C/D appraisal confidence, GRADE, itemized Concerns, every threshold judgment backed by an inline [Author Year] citation, full References at the end.
Read a published paper or diagnose your own draft. Teaching Mode walks through the methods, explains concepts, and surfaces the questions your teacher, journal club, or committee may ask.
Ready to appraise a trial?
Start with a sample report or paste your own abstract.