Who Watches a Medicine After It Reaches the Market?

A medicine enters ordinary care with a dossier full of evidence and an incomplete biography. Its trials may have involved thousands of people, while routine use can expose millions. Trial participants are selected, monitored closely, and followed for a defined period. Patients in clinics are older or younger, take interacting medicines, become pregnant, miss doses, change treatments, and live with conditions that exclusion criteria kept out of the pivotal studies. Rare harms may first become visible only at this scale. Delayed effects require time. The science built to watch this larger and messier world is pharmacovigilance [1].

The word covers the detection, assessment, understanding, and prevention of adverse effects or other medicine-related problems. Responsibility is deliberately distributed. Patients and carers notice what happens after a dose. Doctors, pharmacists, nurses, coroners, and other professionals report suspicions. Manufacturers have legal systems for collecting cases, reviewing literature, maintaining risk plans, and informing regulators. National agencies such as the UK's Medicines and Healthcare products Regulatory Agency assess their own data. In the European Union, national authorities, the European Medicines Agency, and its Pharmacovigilance Risk Assessment Committee share work through a regulatory network. Researchers test signals in clinical and population data.

In Britain, the most visible doorway is the Yellow Card scheme. Anyone can report a suspected side effect through its website or app. A reporter does not need proof that the medicine caused the event. That low threshold is essential. The first person to encounter an unexpected reaction rarely has the denominator, comparator, or specialist information required to establish causality. A report supplies a clue: which product was used, what happened, when it happened, and what other circumstances might matter [2]. Reports can also concern medication errors, defective products, falsified medicines, and medical devices, though the analytical pathways differ.

Reporting is one stream within a plan that begins before launch. An authorisation dossier identifies known risks, possible risks, and gaps in knowledge, then specifies routine or additional measures for studying and reducing them. Companies periodically aggregate worldwide safety data so that a pattern too diffuse for one clinic can be reviewed across the product's life. Regulators inspect pharmacovigilance systems and can demand clarification or new analysis. This planned work matters because passive reporting is shaped by chance and awareness. A suspected risk already visible in trials may call for targeted monitoring, a pregnancy registry, laboratory checks, or a database study instead of waiting for another clinician to notice it [3].

A spontaneous report is best thought of as a clinical observation offered for investigation. Timing can strengthen or weaken suspicion. So can a known mechanism, improvement after stopping the medicine, recurrence after restarting it, an unusual pattern of illness, or several similar cases. Yet coincidence is unavoidable. People have heart attacks, miscarriages, seizures, rashes, and infections whether or not they recently took a medicine. The underlying disease may cause the event. Another treatment may be responsible. The European Medicines Agency therefore states plainly that a safety signal does not by itself mean that the medicine caused the reported problem [3].

Spontaneous reporting has a second limitation: most events are never reported. A systematic review of 37 studies found a median under-reporting rate of 94 percent, though estimates varied widely with the event, setting, and method [4]. Serious, unusual, or newly publicised events may attract more attention than familiar ones. Media coverage can produce a burst of reports. Reports can contain missing details or duplicates, and prescribing volume is usually absent from the reporting database. Ten reports cannot reveal whether the rate was one in a hundred or one in a million unless the number and characteristics of exposed people are known.

These weaknesses do not make the system futile. Spontaneous reports are unusually good at spotting rare, distinctive, and previously unsuspected patterns across many products and countries. They also include the voices of people often absent from trials. A systematic review found that patient reports can add information about the nature, impact, and timing of suspected reactions, complementing reports from health professionals [5]. The point is breadth and sensitivity. The database is designed to raise questions early, accepting false alarms as the price of noticing a real hazard before it grows.

Large collections of reports are screened for disproportionate reporting. Analysts ask whether a particular event appears with one medicine more often, relative to other events for that medicine and the pattern for other medicines in the same database. Methods include reporting odds ratios, proportional reporting ratios, and Bayesian measures that stabilise estimates when counts are small. A high score is a flag for review rather than an incidence rate or proof of causation. Different statistical measures can rank combinations differently, and the quality of the underlying cases still matters [6].

Human review begins where the algorithm stops. Assessors read case narratives, check diagnostic definitions and dates, look for duplicate reports, examine dose and duration, and consider alternative causes. They search clinical trials, observational studies, toxicology, product quality investigations, and the scientific literature. They may compare related medicines or ask whether the suspected effect fits the drug's action. A single well-documented case can be persuasive when the event is extraordinarily specific. A common event such as myocardial infarction usually requires comparative population evidence.

An observed-versus-expected analysis adds a rough denominator and a baseline. Suppose 20 cases of a neurological condition are reported within six weeks of a large vaccination campaign. Analysts estimate how many cases would normally occur in a comparable population over the same amount of person-time, adjusting where possible for age, sex, calendar period, and other determinants. If the observed count exceeds the expected range, concern rises. The result is sensitive to under-reporting, inaccurate exposure counts, choice of risk window, and background rates drawn from a population that may differ from the exposed group [7].

That approach is useful for rapid triage, particularly during mass vaccination, but a historical comparison can mislead. Healthcare use and diagnosis may change during a campaign. The people receiving a product first may be older, frailer, or healthier than the population used to calculate the expected count. Pandemic infection can itself alter the incidence of clotting or neurological disease. An observed excess therefore needs verification in designs that create a more credible comparison, while an observed count close to expectation cannot exclude a risk hidden by incomplete reporting.

Electronic health records, insurance claims, disease registries, and purpose-built cohorts provide the next layer. Researchers can identify exposed and unexposed groups and estimate rates with denominators. They can compare one medicine with another used for the same condition, match people on measured characteristics, or use a self-controlled case series in which each person's risk period after exposure is compared with other times in that same person's record. Each design tackles some biases and leaves others. Confounding by indication is a recurring problem: patients who receive a medicine may differ in health from those who do not, for reasons related to the outcome.

Regulators can require a post-authorisation safety study, known as a PASS, to identify or quantify a hazard, confirm a safety profile, or test whether risk-minimisation measures work. A PASS may be a clinical trial or a non-interventional study using care data. In the EU, imposed study protocols and results are assessed by the Pharmacovigilance Risk Assessment Committee, and information about studies and outcomes is made public through regulatory records and the European catalogue of real-world data studies [8]. A company may sponsor the work, while regulatory oversight, protocol transparency, and reproducible analysis remain essential to its credibility.

The investigation of thrombosis with thrombocytopenia syndrome after the Oxford-AstraZeneca COVID-19 vaccine demonstrates several parts of the system at once. In early 2021, spontaneous reports described unusual blood clots together with low platelets, including clots in veins draining the brain or abdomen. The combination and timing were distinctive enough to trigger urgent European review. EMA's committee examined case reports, exposure data, expected rates, clinical and laboratory evidence, and expert advice. It concluded that the syndrome should be listed as a very rare side effect while judging that the vaccine's overall benefits still outweighed its risks at that stage of the pandemic [9].

That conclusion was not a declaration that every clot reported after vaccination was caused by the product. Common forms of thrombosis did not show the same pattern, and the syndrome required a specific clinical combination. Product information was updated so patients and professionals could recognise warning signs, and regulators introduced restrictions or preferences that varied with age, infection risk, vaccine supply, and national circumstances. Pharmacovigilance changed practice because the signal acquired clinical coherence and supporting evidence. The action was calibrated to a benefit-risk balance that could differ across populations and over time.

Regulatory responses form a ladder. A signal can be closed when evidence does not support it, or kept under monitoring. Authorities may request better case data, further analysis, a periodic safety update, or a formal study. They can add warnings and contraindications, change doses, require laboratory tests, limit prescribing to specialists, issue direct communications to health professionals, suspend supply, or revoke an authorisation. EMA publishes recommendations on assessed signals, and marketing authorisation holders are expected to update product information when the regulatory process requires it [10]. The strongest action is not automatically the safest; removing an effective medicine can also harm patients who need it.

Communication creates its own risk. Public databases list suspected reactions, and the numbers are easy to detach from their warnings. A raw count cannot show incidence, compare risks between products, or prove causation. At the same time, excessively soothing language can erode trust when authorities later strengthen advice. Good communication distinguishes event from reaction, signal from conclusion, relative from absolute risk, and known evidence from continuing uncertainty. It also explains the expected benefit. A one-in-many-thousands harm has a different meaning for a life-saving treatment than for a medicine offering marginal relief.

The system depends on reporting, so silence is ambiguous. It can mean that a harm is absent, that nobody connected it to the medicine, or that people did not know how to report. Patients and professionals should submit serious or unexpected suspicions even when other explanations are possible. Reports are especially useful when they include timing, test results, diagnoses, other medicines, relevant medical history, and what happened after treatment stopped. Urgent clinical care comes first; the Yellow Card scheme is a surveillance route, not an emergency service or a substitute for medical advice.

Who watches a marketed medicine, then? No single sentinel does. Surveillance works as a chain: observation at the bedside, aggregation in national and international databases, statistical screening, clinical assessment, targeted epidemiology, committee judgement, and action that can itself be evaluated. Each link corrects a weakness in another. Trials offer cleaner comparisons but limited scale. Reports offer scale and surprise but weak denominators. Health databases provide comparison while introducing confounding. Regulation joins these imperfect views into a decision. A medicine's safety profile remains provisional because use keeps generating evidence, and watching is part of what makes continued prescribing responsible.