METRIDA

trial evidence infrastructure

Filter and find all evidence in seconds

58,363
PubMed Abstracts Screened
328
Confirmed RCTs
826
Doctor Decisions
93.5%
Classifier Accuracy
13,033
Open Access Papers

smartData.json

Structured trial data for every confirmed RCT. Each file contains PMID, study design metadata, SNOMED-CT intervention codes, fusion classification tier, and ClinicalTrials.gov linkage. One trial, one file, machine-ready.

Aion Database

A living evidence database of 58,363 ADHD abstracts, continuously enriched via EuroPMC and ClinicalTrials.gov. 19,911 papers with full text. 578 linked to registered trials.

METRIDA Trial
Bank

328 doctor-validated ADHD RCTs with 93.5% classifier agreement. Each trial labeled by clinical experts with full audit trail. Covering publications from 1960 to 2026.

Log-in New user

24 SNOMED-CT Intervention Categories

Every trial classified by standardized clinical intervention codes

Stimulants

  • Methylphenidate
  • Amphetamine
  • Lisdexamfetamine

Non-stimulants

  • Atomoxetine
  • Guanfacine
  • Clonidine
  • Viloxazine
  • Bupropion

Behavioral

  • CBT
  • Behavioral therapy
  • Mindfulness
  • Parent training
  • Psychoeducation

Neuromodulation & Lifestyle

  • Neurofeedback
  • tDCS
  • TMS
  • Exercise therapy
  • Dietary intervention
  • Digital therapeutics
Our team

4-Generation Classification Pipeline

Each abstract passes through four independent classifiers before a calibrated fusion decision

G1

Pattern Recognition

SNOMED-CT coded regex patterns across 24 intervention categories. Identifies trial design keywords including randomization, blinding, and placebo control.

G2

Context-Aware NLP

Negation detection, section weighting, and implicit signal extraction. Filters out review papers and non-interventional mentions.

G3

Semantic Similarity

Sentence-transformer embeddings comparing abstracts against intervention anchor texts. Dual-score filtering with positive and negative anchors.

F

Calibrated Fusion

GOLD/SILVER/RESCUE/REJECT tier system combining Gen 2 and Gen 3 scores. Calibrated thresholds rescue unusual language patterns.

Fusion Decision Tiers

Tier Gen3 Score Decision Logic
GOLD > 0.4 Both keyword and semantic signals agree — high confidence
SILVER >= 0.2 Design keywords detected with weak semantic support
RESCUE > 0.65 Unusual language rescued by strong semantic similarity
REJECT < 0.2 Neither signal found strong evidence of RCT