Barie searches live regulatory databases and medical literature simultaneously. It compiles adverse event types, frequency, severity, and patient demographics into a structured safety report — with a direct search link to every FAERS report, published case study, and regulatory signal alert. Not a training-data summary of historical safety data.
Why adverse event data from training memory is unusable
A pharmacovigilance analyst asked an AI assistant to summarize adverse event reports for a current-generation GLP-1 agonist. The output covered the expected categories: gastrointestinal events, injection site reactions, and the small number of serious cardiac signals from early clinical trial data. The summary was coherent and well-structured. It also completely missed the FDA’s October 2025 update to the prescribing information for semaglutide that added a new gastrointestinal obstruction warning following post-market ahrefs signal analysis, and a 2024 case series published in the New England Journal of Medicine describing aspiration events in patients undergoing elective procedures while on GLP-1 receptor agonists.
Both of those updates were material to the safety assessment she was conducting. Neither was in the training data. The AI tool acts like a historical lens. It had simply described the safety profile as it was known at the time its training data was collected, without any indication that the picture had changed significantly since then.
Adverse event databases are updated continuously. FDA FAERS receives hundreds of thousands of new reports every year. EMA EudraVigilance is updated on a rolling basis. Published case studies appear in PubMed daily. A safety analysis built on training data is a snapshot of an earlier version of the evidence base. In pharmacovigilance, that old picture has regulatory and patient safety consequences.
💡
Barie queries FAERS and PubMed at the moment you submit the prompt: FDA FAERS data is retrieved live from the OpenFDA API, which is updated quarterly with the latest submitted reports. PubMed case study retrieval reflects publications indexed up to the query date. The safety profile in Barie’s output describes what is known about the drug today, not what was known when any model was last trained.
Your prompt
Task prompt
“Research adverse event reports for a specific drug across FDA FAERS database and published case studies.”
One sentence. Barie fires parallel retrieval threads across FDA FAERS, EMA EudraVigilance, PubMed, and the WHO VigiBase, applies severity classification, extracts patient demographics, identifies disproportionality signals, and delivers a structured safety report with every source linked. Here is the complete workflow.
1
Live Multi-Database Retrieval
Step 1: Four databases queried simultaneously — regulatory reports and medical literature together
Barie Deep Research fires four parallel threads at the moment the prompt is submitted. The regulatory thread queries the FDA FAERS OpenFDA API and EMA EudraVigilance for all submitted adverse event reports for the specified drug. The literature thread queries PubMed for published case reports, case series, and observational studies describing adverse events. The WHO VigiBase query adds global spontaneous reporting data beyond the US and European markets. All four threads run simultaneously so the complete dataset is assembled before any analysis begins.
🇺🇸 FDA FAERS
Query OpenFDA API for reports where the suspect drug is active ingredient (e.g. semaglutide, tirzepatide).
4,218 reports
🇪🇺 EudraVigilance
EMA database retrieval. Extracts reporting rates, severity classification, and outcome parameters (recovered, recovering, fatal).
1,847 records
📘 PubMed
Retrieves clinical case reports from the last 12 months. Query: (“drug name” [Title/Abstract]) AND (“adverse event” OR “side effect” OR “case report”). Cross-references with retrieved database signals.
814 publications
🌐 WHO VigiBase
Global spontaneous reporting database query. Adds IC (Information Component) and ROR (Reporting Odds Ratio) data to establish statistical signal strength.
3,840 records
⚡
Web Research checks for recent label changes and regulatory communications: After the primary database retrieval, Barie Web Research checks the FDA Drug Safety Communications, EMA PRAC alerts, and recent product label updates. This catches rapid changes that reflect new safety findings identified through FAERS signal analysis — exactly the kind of update that would not appear in pre-training data responses.
2
Structured Safety Analysis
Step 2: Adverse events classified by type, severity, frequency, and patient demographics
4,218
US FDA reports retrieved
34
Distinct event categories
identified
6
Published case series
cross-referenced by signal
100%
Signals linked to primary data
The combined dataset from all four sources is classified using MedDRA System Organ Class groupings. Events are ranked by reporting frequency, severity (e.g., hospitalization, life-threatening, disability, other serious, non-serious), and patient demographic patterns. Disproportionality analysis using Reporting Odds Ratio and Information Component methodologies identifies events reported more frequently for this drug than for the overall FAERS database population — which is how novel safety signals are detected before they reach the level of a regulatory label change.
The highest-volume serious event category across FAERS and EudraVigilance. Represents severe delayed gastric emptying events beyond normal drug mechanism of action in a subset of patients. Disproportionality signal confirmed (PRR > 2.0). A 2025 FDA updated label warning was precipitated by this cluster. High severity rating; hospitalized in over 45% of cases. Elevated rates in patients with delayed gastric emptying on concurrent medications.
Median age: 54
Females: 72%
Hospitalization in 45%
Rare but severe hepatobiliary events form a distinct clinical cluster. Well-documented in prior GLP-1 literature, with a consistent signal across FAERS, EudraVigilance, and VigiBase. The PRR signal is lower than gastroparesis but remains positive. Elevated occurrence rates with extended long-term continuous use rather than early in the dose titration phase. Symptoms typically present with acute abdominal pain; cholecystectomy was frequently required.
Median age: 58
Females: 68%
Hospitalization in 60%
An emerging and complex post-market signal in 2024 and 2025, driving regulatory investigations. The PRR signal reached the minimum threshold for review but remains relatively low. The EMA initiated a review that concluded there is no direct causal link currently. FDA has similarly stated the evidence does not show a clear relationship. However, the signal volume persists in spontaneous reporting systems.
Active monitoring
EMA PRAC inquiry initiated
FDA evaluating safety signal
The most frequently reported non-serious events. Injection site reactions are expected and correlate with mechanism of administration. Alopecia was historically not strongly associated but has seen increasing spontaneous reporting volume in FAERS in the 2024 onwards. Non-serious outcomes, associated with rapid weight loss demographic. It is likely a sequela of rapid extreme weight loss in this patient population rather than a direct toxic drug effect on hair follicle development.
Mild severity
Alopecia 18%, Site reactions 82%
3
Structured Safety Report
Step 3: The structured safety report — every data point linked to its source
Total records
4,218 reports from FDA FAERS, 1,847 from EudraVigilance, 3,840 from WHO VigiBase. Deduplicated unique patients: 7,422.
Most reported event
Nausea (3,412 reports), vomiting (2,810) — consistent with known GLP-1 side effect profile. Mostly non-serious, short duration, early in dose escalation.
Serious events with signal
Disproportionality in reports for gastroparesis, intestinal obstruction, gallstones, pancreatitis. Signal criteria met for cholelithiasis in continuous long-term use.
Patient demographics
Median report age: 56 years. Female: 69%. Demographics skew reflects broader GLP-1 prescription demographics for obesity. No distinct sub-cohort signals isolated for hepatobiliary events.
Label changes detected
EMA added warning for intestinal obstruction. FDA updating labeling for aspiration risk under general anesthesia (Oct 2025).
Similar regulatory action
MHRA initiated safety review on suicidal ideation. EMA concluded no causal link (April 2026). FDA continues to evaluate post-marketing surveillance data.
📄
The live label changes read data absent from any training data response: The October 2025 gastroparesis warning and the March 2026 peri-operative aspiration risk were added to the prescribing information after most AI training datasets were collected. A pharmacovigilance analyst relying on a training-data answer would have an incomplete safety picture and would need to verify separately. Barie’s Web Research layer retrieves the current DailyMed database and flags any updates executed against the base label text directly.
Step 4: Report delivered to your pharmacovigilance and medical affairs tools
The structured safety report and the full citation database exports to the tools your medical affairs, regulatory affairs, or pharmacovigilance team uses. Notion receives the complete report as a linked database with every source entry connecting to its primary FAERS record, PubMed PMID, or regulatory communication page. A Word document version is available for inclusion in a PSUR, DSUR, or regulatory submission package. The list disproportionality signal summaries go to Slack for team awareness before the next safety review meeting.
🔄
Quarterly re-runs align with the FAERS update cycle: FDA FAERS is updated quarterly. Configure Barie to re-run the same adverse event query each quarter and push new signal findings to Notion automatically. When a new event category crosses a disproportionality threshold or a new label change is detected via the Web Research layer, the team receives an alert before the next scheduled signal review. Continuous safety monitoring rather than point-in-time research.
What you get
A structured adverse event safety report covering 4,218 FAERS reports, 1,847 EudraVigilance records, 3,840 WHO VigiBase ICSRs and 47 included PubMed case report/case series. Adverse events classified by MedDRA System Organ Class, ranked by frequency, severity, and disproportionality signal strength. Patient demographic analysis across age, sex, and concomitant medication cohorts. Two label changes detected via the Web Research layer that would not appear in any training data response. The complete report exported to Notion, Word, Google Docs, and Slack in one session. Configured for quarterly re-runs aligned to the FAERS update cycle.
The Verdict
FDA FAERS receives over two million new reports per year. The most consequential safety findings for any drug are often the most recent ones — the signals that have just crossed a threshold, the label changes that were issued last quarter, the case series that was published in the last three months. A pharmacovigilance review built from training data describes the safety profile as it was understood at an unknown historical point. It cannot tell you about the October 2025 gastroparesis warning. It cannot tell you about the active suicidal ideation signal review. It cannot tell you what the current label says. Barie queries the live databases at the moment the prompt is submitted. That is not a convenience. That is the standard the task requires.
Barie features used in this task
Feature
ChatGPT
Perplexity
Barie
Live FAERS/VigiBase OpenFDA Query — current quarterly database retrieved at prompt time via official API
✗
✗
✓
Web Research Label Change Detection — DailyMed current label checked for updates absent from training data
✗
✗
✓
Disproportionality Signal Analysis — ROR and IC methodology applied to combined regulatory report dataset
✗
✗
✓
Primary Source Citations — every event category links to FAERS records, EudraVigilance data, or PubMed PMIDs
✗
✗
✓
Windows App 
MacOS (sillicon) 