Adverse Drug Reactions

Adverse drug reactions (ADRs) are unintended, harmful responses to medications that occur at doses normally used for prophylaxis, diagnosis, or therapy. They are the most common cause of iatrogenic injury in healthcare, responsible for more hospital admissions than most diseases themselves, yet they remain systematically underreported, underanalyzed, and undertheorized. The standard framing treats ADRs as a problem of pharmacology: a drug interacts with a patient in an unexpected way. But the deeper problem is institutional: the feedback topology that would detect, report, and prevent ADRs is structurally broken.

The Detection Gap

The majority of ADRs are never detected. Patients do not report symptoms they cannot attribute to a drug. Clinicians do not report reactions they cannot prove. The pharmacovigilance systems that exist are passive — they wait for reports rather than actively seeking signals. This is the opposite of the surveillance systems used in aviation or nuclear power, where near-misses are aggressively investigated. The medical system treats ADRs as inevitable side effects rather than as preventable system failures, and the feedback loop that would drive improvement is attenuated by the very culture that normalizes harm.

The Reporting Failure

Even when ADRs are detected, they are rarely reported. The reporting burden falls on clinicians who are already time-constrained, and the incentives are perverse: reporting an ADR creates documentation work, potential liability exposure, and no tangible benefit to the reporter. The result is a massive underreporting bias that makes pharmacovigilance databases systematically incomplete. The FDA's Adverse Event Reporting System (FAERS) contains a fraction of the actual events, and the fraction it contains is not representative. The most serious reactions may be the least reported, because they produce the most legal exposure.

The Prevention Paradox

The deepest problem is that ADR prevention requires information that the system cannot generate. To predict which patient will react to which drug, we would need comprehensive data on genetics, comorbidities, drug interactions, and environmental factors — data that is fragmented across institutions, encoded in incompatible formats, and legally inaccessible. The federated learning architectures that could, in principle, enable privacy-preserving analysis of ADR patterns are not deployed for pharmacovigilance, because the institutional incentives for data sharing are weaker than the institutional incentives for data hoarding. The result is that ADRs continue to occur at predictable rates, and the system continues to treat them as unpreventable.

Adverse drug reactions are not a pharmacological problem. They are a systems problem — a failure of detection, reporting, and prevention that no single drug or patient can solve. The solution is not better drugs. It is better feedback.