A pharmacovigilance analyst opens a single FAERS record and sees a patient on semaglutide admitted to hospital with vomiting, dehydration, and acute kidney injury, coded to three MedDRA Preferred Terms and a single "Suspect" drug role. The temptation is to read that record as a mini case report proving the drug caused the injury. It does not. That one record is an Individual Case Safety Report (ICSR) — a structured but unverified narrative submitted voluntarily to a database that exists to detect signals, not to adjudicate them. Knowing how to read that record — which fields carry weight, which are routinely missing, and which conclusions the format itself forbids — is the difference between competent postmarket surveillance and a misleading safety slide.
This is a reading guide for one FAERS report, written for pharmacovigilance, medical-affairs, regulatory, and market-access teams who pull ICSRs from the openFDA API, the FDA FAERS Public Dashboard, or the quarterly ASCII extracts. It covers the structure of a report, the fields a reviewer reads in order, the MedDRA coding that makes reactions searchable, and the reporting-bias limits that define what a FAERS report can and cannot tell you. It is the methodological companion to the FAERS database analysis of 20 million reports and the GLP-1 receptor agonist class analysis; it is not medical advice and does not assess any individual patient.
What FAERS actually is: a spontaneous-reporting signal detector
The FDA Adverse Event Reporting System — now referenced on FDA's surveillance pages as the FDA Adverse Event Monitoring System (AEMS), though the openFDA data and most published literature still use the name FAERS — is a database of adverse event, medication error, and product-quality reports submitted to FDA for approved drug and therapeutic biologic products. It exists to support FDA's postmarket safety surveillance: its job is to surface potential safety signals for FDA reviewers to investigate, not to establish incidence rates or causality.
Reports enter FAERS through two channels. Roughly 95% arrive through manufacturers, who are legally required to submit ICSRs they become aware of (from clinicians, patients, the literature, and their own programs) within regulated timelines. The remaining roughly 5% arrive directly from healthcare professionals, patients, and consumers through the MedWatch voluntary reporting program — online via the FDA Safety Reporting Portal, by Form FDA 3500 for health professionals, or the consumer Form 3500B. This direct-versus-manufacturer split matters enormously when you read a report, because the two channels produce systematically different records: manufacturer ICSRs are typically more complete and MedDRA-coded by trained staff, while direct consumer reports more often lack dose, lot, medical history, and concomitant medications.
The structural format is international. FAERS follows the International Council for Harmonisation E2B standard, the same individual case safety report format used by the EU's EudraVigilance and the WHO's VigiBase. That common format is why a FAERS record looks the way it does and why MedDRA coding is non-optional.
The fields, in the order a reviewer reads them
A FAERS record is one row in the public extract, but it carries the same fields FDA reviewers see. Reading them in the right order prevents the most common mistake — anchoring on the reaction term before establishing who reported, how seriously, and in what drug role.
The identifiers and dates
The safetyreportid is the unique case identifier; the receivedate is the date FDA received the report, not the date the event occurred, and it is the date that assigns a report to a calendar year in any trend analysis. Serious reports carry a version flag because manufacturers file follow-up ("alert") reports within 15 calendar days of receiving new information on a serious case, so a single patient can appear as several versions of the same safetyreportid. De-duplication — collapsing those versions into one case — is a mandatory preprocessing step before any aggregate count, and it is the single most common source of inflated numbers in amateur FAERS analyses.
Serious and seriousness
serious is a yes/no field. seriousness carries the regulatory reason, pipe-joined: death, life-threatening, hospitalization (initial or prolonged), disabling (persistent or significant disability or incapacity), congenital-anomaly, and other (other medically important events that interrupt treatment or require intervention). This is the ICH serious-adverse-event definition verbatim, and it is the most important single filter in FAERS. A report coded serious=yes with seriousness=other is a very different animal from one coded seriousness=death, and a class-level "serious %" that lumps all six reasons conceals that. When you read a report, read the seriousness reason before the reaction term.
The reporter field
reporter identifies who submitted the case — Physician, Pharmacist, Other health professional, Consumer, or Lawyer. The reporter field is the single best predictor of report quality and completeness. A physician- or pharmacist-filed ICSR with manufacturer follow-up is a document you can build a case series on. A consumer-filed direct report with no lot number and a free-text narrative is a signal that something happened, no more. Lawyer-filed reports cluster in litigation windows and can transiently spike a drug's report count without any change in real-world incidence — a confound that has distorted more than one FAERS-based safety claim.
The drug list and the role field
This is where most misreadings happen. A FAERS report carries a list of drugs (drug_substances, drug_brands, drug_generics) and, in parallel, a list of roles:
- Suspect — the drug the reporter believes caused the reaction.
- Concomitant — a drug the patient was taking that is not implicated.
- Interacting — a drug suspected of interacting with the suspect drug to cause the reaction.
The role field is the entire basis for attributing a report to a drug. A report that names semaglutide as concomitant while listing metformin as suspect is not a semaglutide report for attribution purposes, even though semaglutide appears in the drug list. Published studies and the FDA dashboard typically distinguish "primary suspect" and "secondary suspect" from concomitant, and a defensible drug-level count requires role attribution, not mere mention. (In the openFDA extract, role fields are sometimes de-duplicated and do not always map one-to-one to the drug list, which is why careful studies parse the parallel arrays rather than assuming alignment.)
Reactions and MedDRA coding
reactions carries the adverse event terms, pipe-joined, coded to MedDRA Preferred Terms. MedDRA — the Medical Dictionary for Regulatory Activities — is the ICH-maintained standardized vocabulary with a five-level hierarchy: System Organ Class (SOC), High-Level Group Term, High-Level Term, Preferred Term (PT), and Lowest-Level Term (LLT). The Preferred Term is a single medical concept ("Nausea," "Acute kidney injury," "Pancreatitis acute"), and the LLT captures the reporter's original wording mapped to that PT.
Two consequences follow. First, the same event can be coded to several PTs (a single case of severe vomiting with dehydration and renal decline may carry three PTs), so PT counts are event counts, not patient counts, and they exceed report counts for any multi-symptom case. Second, MedDRA is updated twice a year, and PT casing and granularity drift over time in public extracts, which is why case-normalization is necessary before aggregating. For signal searches, reviewers use Standardised MedDRA Queries (SMQs) — predefined groupings of PTs that capture a broader medical concept such as "gastrointestinal nonspecific symptoms and therapeutic procedures" — rather than single PTs, because single-PT searches miss synonymous coding.
Outcomes
reaction_outcomes records the resolution status per event: Recovered, Recovering, NotRecovered, RecoveredWithSequelae, Fatal, and Unknown. Outcome is not the same as seriousness: a serious hospitalization report can have outcome Recovered, and a Fatal outcome is what makes a report count toward death statistics. The high prevalence of Unknown outcomes in consumer-filed reports is itself a quality signal.
Indications, demographics, and geography
drug_indications carries the reasons drugs were given, pipe-joined and parallel to the drug list. For access teams, the indication field is how you separate a diabetes-prescribed GLP-1 report from an obesity-prescribed one — a confound that matters when comparing agents. patient_sex, patient_age, and patient_weight are frequently blank, especially in consumer reports. occurcountry is the country of event occurrence and drives the US-versus-international split that affects any reporting-rate comparison.
What FAERS can defensibly tell you
Given those fields, a well-read FAERS record or case series supports a narrow but real set of conclusions.
It can tell you that a reaction was reported in temporal association with a drug, with whatever detail the reporter supplied. It can tell you how many reports name a drug and how that count has changed over time, which is useful for monitoring and for anticipating FDA attention. It can tell you which reaction PTs cluster with a drug, which is how FDA reviewers build the case lists they then investigate. And it can tell you who is reporting — the consumer-versus-clinician-versus-lawyer mix that contextualizes everything else.
Within FDA, reviewers do not rely on the public dashboard or quarterly files alone for signal evaluation; they access the full FAERS database, retrieve all ICSRs coded to the MedDRA terms of interest, and read the individual narratives. The public extract reproduces the structured fields but not always the full narrative, which is one reason public FAERS analyses are inherently less conclusive than FDA's internal work.
What FAERS cannot tell you — the four hard limits
Everything defensible in FAERS sits inside four limits, and a report read without them is a report misread.
No causation. FAERS is a spontaneous-reporting system. The existence of a report naming a drug and a reaction does not establish that the drug caused the reaction. Patients on medication have underlying disease, comorbidities, and concomitant drugs; the report records co-occurrence, not attribution. FDA states this directly: a report does not establish causation.
No incidence and no denominator. FAERS has no reliable exposure denominator. You cannot compute the rate of an event from report counts alone, because you do not know — at the level of a single PT — how many patients took the drug, for how long, at what dose, and with what background risk. A drug with ten million users will accumulate more reports of a common background event than a drug with one hundred thousand users, with no difference in true risk.
Underreporting and reporting bias. Spontaneous systems are subject to severe underreporting (estimates historically suggest the great majority of serious events are never reported) and to reporting bias — a drug in the news, in litigation, or newly approved receives disproportionate reporting that has nothing to do with its inherent hazard. The "Weber effect," litigation-driven batches, and direct-to-consumer campaigns all distort counts. These biases mean report counts track attention and usage as much as they track risk.
Duplicates and quality variance. The same case can be submitted multiple times by different sources (physician, manufacturer, patient), creating duplicate records that inflate counts unless de-duplicated. Report completeness and coding quality vary by reporter and by year. These are mechanical errors that must be cleaned before any serious analysis.
Signal detection: why disproportionality, not raw counts
Because of those four limits, the discipline of converting FAERS reports into a safety signal uses disproportionality statistics, not raw counts. A signal asks whether a drug-event pair is reported more often than expected given the drug's overall reporting volume — the classic 2×2 contingency of drug-of-interest versus all other drugs and event-of-interest versus all other events. The standard statistics are the Reporting Odds Ratio (ROR), the Proportional Reporting Ratio (PRR), the Bayesian Confidence Propagation Neural Network information component (used by WHO VigiBase), and the Empirical Bayesian Geometric Mean (EBGM, used in FDA's own data mining).
A positive disproportionality signal — say, ROR lower-bound above 1 with adequate case count, or an EBGM above a threshold — is a hypothesis that the drug-event association is stronger than background. It is still not proof of causation. It is, however, the minimum bar for treating a FAERS observation as something worth investigating, and it is why no competent safety review ranks drugs by raw report count or by serious percentage.
The reading workflow, before quoting a FAERS number
Before you cite a FAERS figure in a briefing, dossier, or policy submission, run it through this checklist.
- Define the denominator question. State what the count is — reports naming the drug in any role, reports with the drug as suspect, reports with a specific PT, de-duplicated cases — and state explicitly what it is not (an incidence rate, a causal estimate).
- De-duplicate. Collapse
safetyreportidversions to one case. If you cannot de-duplicate, say so and treat the count as an upper bound. - Attribute by role. Decide whether you are counting suspect, suspect-plus-interacting, or any mention, and apply it consistently. Do not compare a suspect-only count to an any-mention count.
- Read the seriousness reason, not just the serious flag. Death and hospitalization are different signals; "other" is weaker still.
- Stratify by reporter. A 90% consumer-filed case series reads differently from a 90% physician-filed one, and a litigation-driven lawyer cluster should be flagged as a temporal artifact.
- Normalize MedDRA. Case-normalize PTs, consider an SMQ rather than a single PT, and note the MedDRA version if precision matters.
- State the bias caveats. Underreporting, stimulated reporting, no denominator, no causation — in the same document as the number, not in a separate disclaimer.
What pharmacovigilance and access teams should take from a FAERS report
A single FAERS ICSR is a lead, not a finding. Read in aggregate, with de-duplication, role attribution, and disproportionality statistics, FAERS is the front end of FDA's postmarket surveillance — the system that surfaces the hypotheses FDA then investigates with the full database, epidemiologic studies, and label-change reviews. For market-access teams, FAERS literacy matters because payer policies, label changes, and risk communications increasingly cite FAERS-derived numbers, and the difference between a defensible and an indefensible citation is almost always one of the four limits above. Anchor claims to current FDA labeling; use FAERS to describe what is being watched, not what has been proven.
Sources
- US FDA, FDA Adverse Event Reporting System (FAERS) / FDA Adverse Event Monitoring System (AEMS). https://www.fda.gov/drugs/surveillance-post-drug-approval-activities/fda-adverse-event-monitoring-system-aems
- openFDA, "Data from FAERS (FDA Adverse Event Reporting System)." https://open.fda.gov/data/faers
- "FAERS Essentials: A Guide to Understanding, Applying, and Interpreting Adverse Event Data Reported to FAERS," PMC. https://pmc.ncbi.nlm.nih.gov/articles/PMC12393772
- IntuitionLabs, "The FDA Adverse Event Reporting System (FAERS) Explained" (MedDRA coding, MedWatch channels, causation limits). https://intuitionlabs.ai/articles/fda-adverse-event-reporting-system
- ICH, "MedDRA Use at FDA" (serious AE definition, 15-day alert reports, AERS/FAERS limitations). https://admin.ich.org/sites/default/files/inline-files/Regulatory_Perspective_SBrajovic.pdf
- US FDA, MedWatch: The FDA Safety Information and Adverse Event Reporting Program (Form FDA 3500 / 3500B, Safety Reporting Portal). https://www.fda.gov/safety/medwatch-fda-safety-information-and-adverse-event-reporting-program
- US FDA, FAERS Public Dashboard. https://www.fda.gov/drugs/questions-and-answers-fdas-adverse-event-reporting-system-faers/fda-adverse-event-reporting-system-faers-public-dashboard
- "A real-world data analysis of tirzepatide in the FDA Adverse Event Reporting System (FAERS) database" (ROR, PRR, BCPNN, EBGM signal-detection methodology), Front Pharmacol, 2024. https://www.frontiersin.org/journals/pharmacology/articles/10.3389/fphar.2024.1397029/full
