Clinical evidence, not promises.
45 studies published with hospitals, insurers and universities across 6 countries. Every result is one click from its original paper.
- publications: peer-reviewed papers, preprints and evidence briefs
- 45
- million patients analyzed
- 68
- countries with studies
- 6
- research lines
- 5
We validate every model with real-world data and blinded specialists before it reaches the clinic.
- Hospitals
- Insurers
- Universities
Research lines
Five fronts where we measure, with real data, whether AI improves clinical decisions.
Kidney and nephrology
Early detection of chronic kidney disease.
4 studies
Medical imaging
X-ray, CT and OCT combined with clinical data.
14 studies
LLMs and clinical search
Assistants that answer with cited evidence.
14 studies
Clinical data and EHR
Reading clinical records at scale, nuance intact.
8 studies
Population health
Predicting risk across millions of patients at once.
5 studies
Latest research
Filter by research line or evidence type. Every card leads to the summary and the paper.
- Featured
AI in primary care and internal medicine: only 5% of 519 health-care LLM studies used real patient-care data
5% Health-care LLM studies using real patient-care data
LLMsWhitepaper / evidence briefZenodo · 2026 · 1 min AI in neurology: stroke AI detects 62% of large-vessel occlusions versus up to 96% for human readers
0.62 Sensitivity · AI for large-vessel occlusion
ImagingWhitepaper / evidence briefZenodo · 2026 · 1 minAI in surgery and anesthesiology: 8 randomized trials and no proven general benefit
8 Randomized-trial reports · hypotension AI
PopulationWhitepaper / evidence briefZenodo · 2026 · 1 min
AI in mental health: chatbots eased depression in 18 trials, but not at 3 months
18 Randomized trials · AI chatbots
LLMsWhitepaper / evidence briefZenodo · 2026 · 1 min
AI in pediatrics: only 6% of models in 112 studies are interpretable and externally validated
6% Models both interpretable and externally validated
PopulationWhitepaper / evidence briefZenodo · 2026 · 1 min
AI in obstetrics and gynecology: cervical cytology AI reaches AUC 0.947, but regional evidence rests on two Mexican cohorts
0.947 AUC · AI cervical cytology grading
ImagingWhitepaper / evidence briefZenodo · 2026 · 1 min
AI in pulmonology: assistance lifts chest-radiograph AUC to 0.840 in a randomized trial, without changing clinical decisions
0.840 Chest-radiograph AUC with AI assistance
ImagingWhitepaper / evidence briefZenodo · 2026 · 1 min
AI in ophthalmology: diabetic retinopathy screening leads at 87.2% sensitivity; Latin American evidence only from Mexico
87.2% Sensitivity · more-than-mild diabetic retinopathy
ImagingWhitepaper / evidence briefZenodo · 2026 · 1 min
AI in dermatology: skin-cancer classification is most mature, with zero prospective clinical validations in Latin America
0 Prospective clinical validations of dermatology AI found in Latin America
ImagingWhitepaper / evidence briefZenodo · 2026 · 1 minAI in emergency and critical care: no Latin American evaluation in 28 sources; a sepsis model reached AUC 0.63 externally
28 Sources reviewed
EHRWhitepaper / evidence briefZenodo · 2026 · 1 min
AI in digital pathology: 24-source review finds detection and grading most mature, molecular prediction still retrospective
24 Sources reviewed
ImagingWhitepaper / evidence briefZenodo · 2026 · 1 min
Physicians using conversational AI score 2.83 vs 2.46 on answer validity in a randomized vignette experiment in Latin America
2.83 Answer validity out of 3
LLMsPreprintmedRxiv · 2026 · 1 min
AI in gastroenterology and hepatology: 18-source review finds assisted colonoscopy most mature, but outcome benefit unproven
18 Sources reviewed
ImagingWhitepaper / evidence briefZenodo · 2026 · 1 min
AI in radiology: 27-source review finds AI chest X-ray flags actionable nodules in 0.59% vs 0.25%
0.59% vs 0.25% Actionable nodules detected on chest X-rays · with vs without AI
ImagingWhitepaper / evidence briefZenodo · 2026 · 1 min
AI-built physician portraits from clinical searches identify 78.4% of physicians, but raw search text reaches 85.0%
78.4% Top-1 identification · portrait (TF-IDF)
LLMsZenodo · 2026 · 1 min
Framework sorts clinical-document identifiers into 9 categories before AI medical record review in Colombia
9 Sensitive-variable categories
EHRZenodo · 2026 · 1 min
AI in oncology: 27-source review finds AI mammography the most mature use, with 17.6% more cancers detected
+17.6% Cancer detection with AI-assisted mammography
ImagingWhitepaper / evidence briefZenodo · 2026 · 1 min
Medical students using Arkangel AI give more valid clinical answers in 55% less time: a randomized trial
2.73 Clinical validity out of 3
LLMsPeer reviewedIntelligence-Based Medicine · 2026 · 5 min
AI in cardiology: 25-source review finds ECG and imaging AI validated, but clinical benefit rarely shown
25 Sources reviewed
ImagingWhitepaper / evidence briefZenodo · 2026 · 1 min
LLM-supported grant workflow screens 741 funding sources and advances 5 applications to interview or finalist stage
741 Funding sources scanned
LLMsPreprintResearch Square · 2026 · 1 min
67.6% of Latin American physicians' queries on Arkangel AI seek management, diagnosis, or work-up support across 20 countries
67.6% Queries on management, differential diagnosis, or work-up
LLMsPreprintmedRxiv · 2026 · 1 min
Payer-side SOAT traffic accident claims audit structures review into 135 versioned audit elements
135 Versioned audit elements
EHRPreprintResearch Square · 2026 · 1 min
Governed AI manuscript workflow corrects 8 of 9 governance gaps within one development cycle
8 of 9 Governance gaps corrected
LLMsPreprintJMIR Preprints · 2026 · 1 min
Governed machine learning with 18 human checkpoints builds a CKD screening model at AUROC 0.7479
0.7479 AUROC · CKD screening in type 2 diabetes
PopulationPreprintJMIR Preprints · 2026 · 1 min
Arkangel AI, OpenEvidence, ChatGPT, and Medisearch evaluated against medical standards
100% Reference authenticity · Arkangel AI-Deep
LLMsPeer reviewedMayo Clinic Proceedings: Digital Health · 2026 · 1 min
AI detects bipolar disorder in EHRs of patients with affective diagnoses with AUC 0.93
0.93 AUC · bipolar disorder
EHRPreprintmedRxiv · 2026 · 1 min
From free text to SOFA score reconstructs sepsis severity from clinical notes
100% Respiratory variable recovery
EHRPreprintmedRxiv · 2025 · 1 min
Recursive learning architecture improves zero-shot clinical coding F1 from 0.318 to 0.605
0.605 F1 · zero-shot clinical coding
EHRPreprintJMIR Preprints · May 2026 · 1 minDecision tree detects CKD risk with 90.2% sensitivity from eight routine variables in 203,067 Latin American records
90.2% Sensitivity · CKD risk (eGFR < 60)
KidneyPreprintResearch Square · 2025 · 1 min
Arkangel AI real-time multi-LLM agent answers clinicians' medical questions with 90% accuracy
90.26% Accuracy · MedQA
LLMsPeer reviewedIntelligence-Based Medicine · 2025 · 5 min
PANDORA LLM Automates COPD Risk Detection with Near-Perfect Extraction and 94% PUMA Accuracy
94.5% PUMA scoring accuracy · MIMIC
EHRConference paper2025 IEEE Conference on Artificial Intelligence (CAI) · 2025 · 6 min
Minimal-resource AI Detects CKD in Type 2 Diabetes Patients Across Six LMICs with 90% Sensitivity
90.05% Sensitivity · CKD in type 2 diabetes
KidneyAug 2025 · 5 min
Patients and clinicians: LLMs achieve high QA accuracy but require human evaluation for clinical safety
90.2% MedQA accuracy · GPT-4 + Medprompt
LLMs2024 · 6 min
Vitruvius conversational AI achieves 90% accuracy on USMLE-style clinical queries for patients across specialties
90.26% Accuracy · MedQA
LLMsPreprintmedRxiv · 2024 · 6 min
PANDORA AI Extracts EHR Data, Identifies COPD Risk in Patients with 98% PUMA Accuracy
98% PUMA scoring accuracy · MIMIC-IV
EHRPreprintAug 2025 · 5 min
Clinicians Using Arkangel AI Conversational Search Answer Questions 79% Faster, 34% Fewer Searches
−79.8% Time to answer
LLMs2024 · 6 min
CNN Identifies Exudative Macular Disease on OCT in AMD and DME Patients with 97% Accuracy
96.4% Accuracy · exudative macular disease
ImagingRetina Today · 2024 · 4 min
GPT-4o Conversational Agent Achieves 100% Guideline Accuracy in Alzheimer’s Disease Care
100% Concordance · Alzheimer's knowledge scales
LLMsPreprintmedRxiv · 2024 · 6 min
AI Chest X‑Ray + Clinical Data Predicts ICU Admission (AUC 0.92) in Hospitalized COVID‑19 Patients
0.92 AUC · ICU admission
ImagingPeer reviewedERJ Open Research · 2022 · 5 min
Ensemble AI Detects CKD with 91% Sensitivity in Diabetics and 92.5% in Non‑Diabetics
91% Sensitivity · diabetic patients
Kidney2024 · 5 min
Arkangel.AI predictive models reduce hospital admissions up to 45% across 68 million chronic disease patients
−45% Hospital admissions (up to)
PopulationAug 2025 · 5 min
Automated AI Chest Imaging Analysis Increases Lung Nodule Detection Sensitivity 13% in Screening Patients
+13% Nodule detection sensitivity
ImagingAug 2025 · 5 min
No-code Hippocrates AutoML Builds Pediatric Leukemia AI Models Tenfold Faster
10× Faster model development
PopulationAug 2025 · 5 min
Noninvasive AI Algorithm Boosts Early CKD Detection by Up to 90% in Latin American Patients
+90% More at-risk patients identified (up to)
KidneyAug 2025 · 5 min
Arkangel AI Tool Improves Early Breast Cancer Detection in Latin American Women to AUC 0.90
0.90 AUC (up to) · breast cancer detection
ImagingArkangel AI whitepaper · Aug 2025 · 5 min
Want to validate Arkangel at your institution?
We design the study with you: protocol, ethics committee, de-identified data and joint publication.