🔍AnalyseAdvanced4 steps

AI Medical Literature Synthesis Agent

This agent analyzes a corpus of scientific medical articles to produce a structured literature synthesis on a given clinical question. It evaluates evidence levels, identifies consensus and controversies, and generates actionable recommendations for clinical practice.

recherche médicalesynthèse littératureevidence-based medicineanalyse critiquerecommandations cliniques

For who

Physicians, clinical researchers, hospital pharmacists, and medical students looking to quickly produce a rigorous literature review on a precise clinical question.

Input

Type: text

Format: libre

Question clinique à explorer (ex : efficacité d'un traitement, comparaison de stratégies thérapeutiques) accompagnée des résumés ou textes intégraux des articles scientifiques à analyser (abstracts PubMed, PDF, ou références bibliographiques)

steps (4)

Clinical Question Framing

prompt

Structures the research question according to PICO format and defines inclusion criteria

Critical Article Analysis

prompt

Evaluates each article using critical appraisal grids and evidence levels

Evidence Synthesis

prompt

Cross-references results to identify consensus, divergences, and gaps in the literature

Recommendations and Final Report

prompt

Formulates graded clinical recommendations and produces the complete synthesis report

Output

Type: text

Format: structuré

Rapport complet de synthèse de littérature comprenant : cadrage PICO, analyse critique graduée de chaque article, synthèse narrative des preuves, recommandations cliniques graduées (A/B/C/AE), tableau récapitulatif des études et perspectives de recherche

Example

Input

Clinical question: What is the efficacy of SGLT2 inhibitors (gliflozins) in reducing cardiovascular mortality in type 2 diabetic patients with heart failure with preserved ejection fraction? Articles provided: abstracts of the EMPEROR-Preserved trial (Anker 2021), DELIVER trial (Solomon 2022), Vaduganathan 2022 meta-analysis, French observational registry FASTTRACK-HF 2023.

Output

Executive summary: The synthesis of 4 studies (2 RCTs, 1 meta-analysis, 1 registry) involving 15,842 patients shows that SGLT2 inhibitors significantly reduce the composite endpoint of cardiovascular death/heart failure hospitalization (HR 0.80, 95%CI 0.73-0.87) in T2D patients with HFpEF...

Recommendation 1 (Grade A): It is recommended to initiate SGLT2 inhibitor treatment in all type 2 diabetic patients with heart failure with preserved ejection fraction (LVEF > 40%), in the absence of contraindication...

Recommendation 2 (Grade B): It is suggested to introduce treatment early after heart failure diagnosis, as DELIVER data suggest a benefit from the first weeks...

Customization

Parameter	Description	Default
Système de gradation	Échelle de niveaux de preuve et grades de recommandation utilisée (HAS, GRADE, Oxford CEBM)	HAS (Haute Autorité de Santé)
Profondeur d'analyse	Niveau de détail de l'analyse critique — résumé rapide ou évaluation méthodologique exhaustive avec grilles validées	Exhaustive avec grilles CONSORT/STROBE/PRISMA
Langue des sources	Langues acceptées pour les articles du corpus analysé	Français et Anglais

Technical Notes

This agent is designed to assist — not replace — the clinical judgment of the healthcare professional. Generated recommendations must be validated by a domain expert before any application in clinical practice.For optimal results, provide complete structured abstracts of the articles rather than simple references. The agent can process between 3 and 15 articles per session. Beyond that, divide the corpus into sub-themes and run multiple analyses.The agent uses international critical appraisal frameworks (CONSORT, STROBE, PRISMA, AMSTAR-2) and the HAS classification by default. You can switch to the GRADE system (Grading of Recommendations Assessment, Development and Evaluation) via customization parameters if your institution requires it.Prompts are optimized for interventional biomedical literature (therapeutic trials, diagnostic studies). For literature reviews in health humanities or public health, adapt the PICO framing to the PICo format (Population, phenomenon of Interest, Context) at step 1.

Glossary Terms

Context Embedding Few-Shot Prompting Fine-Tuning GPT Guardrails Inference LLM