A medicine can pass a well-run randomized trial and still have important unanswered safety questions. Trials are designed around defined outcomes, selected participants, protocol visits, and a limited period. They may be too small to reveal a very rare event, too short to show a delayed problem, or too selective to represent people with several illnesses, many medicines, pregnancy, frailty, or impaired kidney or liver function.
Pharmacovigilance is the continuing science and set of activities used to detect, assess, understand, and prevent adverse effects or other medicine-related problems, and it begins before approval and continues while a product is used. Its job is not to declare every reported event a drug reaction. Its job is to turn incomplete observations into questions that can be examined, then to support proportionate action when the total evidence warrants it.
Why approval cannot answer every safety question#
Randomization is powerful because it tends to balance measured and unmeasured differences between treatment groups. Yet even a large trial can observe only the events that occur among its participants during follow-up: a harmful event occurring once in 50,000 users may never appear in a program enrolling several thousand people. A risk that emerges after years can remain invisible in a trial lasting months.
Trial populations also differ from routine care. Eligibility rules commonly exclude people at higher risk of complications or those whose other conditions could make results hard to interpret. Adherence and follow-up are usually more structured than ordinary care. After approval, use expands across ages, diagnoses, medicine combinations, and health systems. That larger and more varied population supplies information that a preapproval program could not contain.
The purpose of postmarket work is therefore complementary. Trials estimate benefits and common harms under controlled conditions. Postmarket systems look for rare, delayed, interaction-related, population-specific, or use-related problems and revisit whether benefits still outweigh risks.
What counts as an adverse event report#
An adverse event is an unfavorable medical occurrence after a product was used. The reporter does not have to prove causation. You, along with caregivers, clinicians, and manufacturers, may submit reports through national systems. In the United States, MedWatch is the FDA program for reporting serious problems with regulated medical products. Manufacturers also have mandatory reporting duties under applicable rules.
A report may describe symptoms, diagnoses, or laboratory abnormalities. It may describe hospitalization, disability, a birth outcome, or death. Useful reports include the sequence of events, indication, dose history when known, other medicines, relevant illnesses, test results, what happened after the medicine was stopped, and whether the problem returned after reuse. Even a detailed report remains an observation, not a randomized comparison.
The word "adverse" can be misunderstood as a finding that the medicine was responsible. In surveillance terminology, the event may be caused by the underlying disease, another medicine, infection, chance, diagnostic error, or the product, and reporting with uncertainty is appropriate because early detection would fail if proof were required before submission.
Why spontaneous reports are valuable#
Spontaneous reporting systems cast a broad net. They can bring attention to a striking clinical pattern, an event rarely seen without a particular mechanism, or an unexpected cluster across different locations. Reports can arrive sooner than a formal study can be planned and completed. This makes them especially useful for hypothesis generation.
Case details can also reveal patterns hidden by a simple count. A consistent time course, improvement after withdrawal, recurrence after reuse, or concentration in a susceptible group may strengthen concern. Similar reports from several data sources can add coherence. Pharmacology and toxicology can supply a plausible mechanism.
These strengths explain why a small number of compelling cases may matter. They do not erase the method's limits. A surveillance system is designed to notice possible signals, not to provide a clean risk estimate for every reported event.
Why a report does not establish causation#
The most important missing quantity is often the denominator. A database may show 500 reports, but the number of users, duration of use, dose patterns, and background rate of the same condition may be uncertain; five hundred reports among 10,000 users mean something different from the same count among 50 million users.
Reporting is incomplete and selective. Common or expected symptoms may go unreported. Publicity, a regulator's warning, or litigation can increase submissions. So can social media or the launch of a new product. This is called stimulated reporting. The pattern can create a rise in counts even if the underlying event rate has not changed.
Reports may be duplicated, lack a confirmed diagnosis, omit other medicines, or contain no usable timeline. A person may take a medicine because of a condition that itself raises the outcome risk, and that problem, known as confounding by indication, can make an association look drug-related when disease severity is the main driver.
Background events also matter. Heart attacks, cancer, miscarriage, headache, and fatigue occur among people who take no particular product. If millions use a medicine, some events will happen soon afterward by coincidence. Temporal sequence is necessary for many causal explanations, but it is not sufficient.
So do not divide counts from a reporting database casually to calculate incidence, compare them as if prescribing volumes were equal, or treat them as a ranking of product safety. The FDA explicitly warns that an adverse event report does not mean the product caused the event.
From report to safety signal#
A signal is information suggesting a new potentially causal association, or a new aspect of a known association, that deserves further evaluation. Signal detection can combine clinical review with statistical methods that look for drug-event pairs reported more often than a database comparison would predict.
Disproportionality is not proof. A high reporting ratio can reflect publicity, indication, or co-medication. It can reflect channeling of high-risk patients, coding practices, or duplicate cases. A low ratio can reflect underreporting. Algorithms help prioritize review; they do not replace it.
Assessment asks several connected questions. Is the diagnosis credible? Did the timing fit? Is there a dose or duration pattern? Did the event improve after stopping? Is recurrence after reuse documented and ethically interpretable? Is a mechanism plausible? Do trials, observational studies, toxicology, class effects, or international reports point in the same direction? Is the event more common than expected among comparable nonusers?
Some signals weaken with better evidence. Others become established adverse reactions. Both outcomes show the system working: an early warning deserves examination even when the eventual conclusion is reassuring.
What active surveillance adds#
Active surveillance starts with a defined question and searches structured data rather than waiting only for voluntary reports; the FDA Sentinel Initiative, for example, uses electronic health data from participating data partners to evaluate medical-product safety. Claims, electronic health records, registries, and linked datasets can supply denominators and comparison groups. They can supply medication records and outcome timing.
These systems can estimate rates and compare people who received different treatments under specified designs. A regulator might examine whether a defined outcome is more frequent after starting one medicine than after starting an alternative for the same condition. Repeated analyses can test sensitivity to definitions, follow-up windows, and confounder adjustment.
Large data do not automatically produce a causal answer. Prescriptions do not guarantee ingestion. Diagnoses may be miscoded. Important factors such as frailty, over-the-counter use, or disease severity may be poorly measured. People receiving different treatments may differ before treatment begins. Sound protocols, transparent assumptions, validation, and sensitivity analyses remain essential.
Registries and prospective postauthorization studies can answer questions that routine records cannot. Pregnancy registries, disease registries, or studies built around a suspected risk can gather focused clinical detail. The ICH E2E guideline frames planning around important identified risks, potential risks, and missing information so that surveillance matches the uncertainty.
How regulators decide what to do#
Regulatory action is based on benefit-risk assessment, not the existence of a report count alone. Reviewers consider seriousness, preventability, and reversibility. They consider certainty, vulnerable groups, and therapeutic benefit. They consider alternatives and whether risk can be reduced without removing a useful treatment.
Possible responses span a wide range. Agencies can request more data, require a postmarket study, or issue a safety communication. They can revise warnings or contraindications, add monitoring recommendations, or limit distribution. They can change packaging or require a risk-management program. Labels may change as knowledge improves. If risk is severe and cannot be managed while preserving a favorable balance, use can be suspended or the product can be withdrawn.
The response may change over time. An initial communication can precede a refined estimate. Later evidence may narrow a concern to a specific group, identify an interaction, confirm a class effect, or show that an early signal was not causal. Updating is a feature of responsible surveillance, not evidence that earlier evaluation was pointless.
In March 2026, the FDA announced its Adverse Event Monitoring System, a public dashboard intended to unify and modernize access to adverse-event data as older interfaces are migrated. Easier access can support transparency, but the interpretation rules remain the same: a visible report is not a verified causal case, and dashboard counts are not incidence rates.
What a reader can do with a safety concern#
If you suspect a problem, it deserves clinical attention in proportion to its severity. Emergency symptoms require emergency services. For a nonemergency concern, a clinician or pharmacist can review timing, other medicines, and kidney or liver function. They can review alternative causes and whether testing or a treatment change is appropriate. Abruptly stopping some medicines can itself be harmful, so a public database is not a substitute for advice about you.
Reporting can still be valuable when causation is uncertain. A complete MedWatch submission may contribute to a pattern that becomes visible only when many observations are assessed together. Describe what happened accurately, without overstating what you know.
Ask what kind of evidence supports a safety claim. Is it a single case, a cluster, a disproportionality signal, a controlled observational analysis, or a randomized finding? Is there a denominator? Were comparable groups used? Has a regulator concluded that the risk is causal, or only announced an investigation? Those distinctions keep you from dismissing an early warning and from panicking at a raw count.
References#
- WHO pharmacovigilance overview
- FDA MedWatch safety reporting program
- FDA postmarket drug safety information
- FDA Sentinel Initiative
- FDA drug safety-related labeling changes
- FDA Adverse Event Monitoring System announcement, 2026
- ICH E2E pharmacovigilance planning guideline
Questions and answers
Does an FDA adverse event report prove that a medicine caused harm?
No. It shows that an event was reported after a product was used. The event may be related, unrelated, or impossible to classify from the available information. Reports are inputs to signal detection and assessment.
Why are reports accepted without proof?
Requiring proof would suppress early warnings. Reporters can provide observations and relevant details while regulators and researchers assess patterns across multiple evidence sources.
Can a reporting database tell me which medicine is safest?
Usually not by comparing raw counts. Product use, duration, patient risk, publicity, mandatory reporting, duplicate submissions, and data completeness differ. Comparative safety requires a design with suitable denominators and comparison groups.
What is the difference between a signal and a confirmed adverse reaction?
A signal is a credible question that merits evaluation. A confirmed adverse reaction has stronger evidence supporting a causal relationship. The evidence can evolve, and regulators may act before certainty is complete when potential harm is serious.
Should someone stop a medicine after finding reports online?
Not on the basis of report counts alone. Urgent symptoms need urgent care; other concerns should be reviewed with a clinician or pharmacist who can weigh the suspected risk, treatment benefit, alternatives, and risks of stopping.