TrialReviewer-Medical Literature Appraisal

TrialReviewer — Appraise medical literature like a clinical epidemiologist. Know whether a result is real, stable, and worth changing your practice.

Do any of these sound familiar? You read an RCT labeled "statistically significant" but aren't sure whether to change your prescribing. Meta-analyses on the same question contradict each other and you don't know which to trust. A paper reports RRR=50% but never mentions NNT. A subgroup analysis looks compelling but feels like data mining.

TrialReviewer fixes this.

Paste an abstract or upload a PDF, and TrialReviewer works through it the way an experienced clinical epidemiologist would — systematically, without shortcuts.

WHAT IT DOES

Statistical interpretation: Calculates ARR, NNT, and NNH automatically — never just reports RRR. Flags p-value traps and interprets results in the context of effect size and confidence intervals.

Research quality assessment: Checks baseline balance, dropout asymmetry, multiple comparisons, subgroup interaction tests, and Fragility Index. Every factor that affects whether a conclusion holds up.

GRADE evidence rating: High / Moderate / Low / Very Low, with explicit reasoning for every downgrade.

Final Confidence score (A / B / C / D): A direct answer to whether this result should change clinical practice — or whether it belongs in the hypothesis-generating pile.

Domain-specific appraisal: Oncology (PFS ≠ OS; surrogate endpoint traps); Cardiology (MACE decomposition; nonfatal MI inflation); Psychiatry (placebo response inflation; HAMD small-effect traps).

Automatic report saving: Every analysis generates a timestamped Markdown report, ready to archive or share.

STUDY TYPES COVERED
RCT, cohort, case-control, meta-analysis, network meta-analysis, survival analysis, Bayesian design, Mendelian randomization, diagnostic studies, and real-world evidence (RWE).

WHY NOT JUST ASK CLAUDE?
General Claude translates statistics. TrialReviewer makes judgments: calculates absolute risk instead of stopping at relative risk; identifies fatal flaws and issues A/B/C/D ratings; applies MCID to distinguish true negatives from underpowered studies; estimates event counts to make "small sample" concrete; catches immortal time bias, collider bias, and subgroup traps; and benchmarks findings against landmark trials to give clinical context. This is a clinical epidemiologist's appraisal framework — not a statistics textbook.