Why do medicines have side effects?

An intended target can have several jobs

A medicine can affect several processes in different tissues. Receptors, enzymes, transporters and ion channels may be present in many organs, not only the one named in a treatment claim. An unwanted effect may come from the intended target working elsewhere, or from another target at a different exposure. A mechanism diagram can suggest how this might happen; it cannot predict how often an effect occurs or how important it is to patients.1

Exposure is part of the picture

Effects depend on what reaches tissues and for how long, not only the amount given. Absorption, distribution, metabolism, removal, organ function, formulation and interactions can change exposure. Age, genetics, illness and other treatments can also change the concentration at a target or the response.

A broad safety statement is incomplete without the product, dose, people, comparison and time period. Drug may still be measurable after an important effect ends; a breakdown product or later biological response can also persist after the original drug level falls.3413

Event does not automatically mean cause

An adverse event is an unwanted medical occurrence after someone receives a treatment, whether or not it is thought to be related. The term alone does not establish a cause. In pre-approval research, an adverse drug reaction is narrower: it indicates at least a reasonable possibility that the medicine caused the response, though not necessarily proof.

Headaches, infections and falls can also occur without the treatment. A comparison group helps show whether events are more frequent with it. Timing, dose, biological plausibility, what happens after stopping or restarting, and other explanations can all help assess causation. One report can identify a possible signal but usually cannot establish its cause or frequency on its own.2571189101516

Small studies leave some questions open

“Severe” and “serious” mean different things in research. Severity describes intensity. Seriousness includes death, an event that was life-threatening at the time, hospital admission or a longer stay, persistent or significant disability, and congenital anomalies. An important medical event can also be serious if clinical judgement says it jeopardises a person or needs intervention to prevent one of those outcomes. An emergency visit without admission is not automatically serious, although another criterion may apply.

A severe symptom may not meet the formal definition of serious; a serious event need not be called severe in everyday language. A study may report observed harms yet be too small or short to rule out rare or delayed problems. “No serious events observed” does not mean serious harm is impossible. Include treatment stopping due to unwanted effects alongside event counts: looking only at people who remain can make tolerability seem better. Monitoring continues after approval.261114

Read the denominator and the time window

A useful safety result states how many people experienced an event, out of how many, over what period and in which comparison group. Imagine nausea occurred in 12 of 100 people taking a medicine and eight of 100 taking placebo over 12 weeks. The observed difference is four additional people per 100 reporting nausea over that period. A relative figure does not replace those counts.

Check whether the analysis included everyone assigned, everyone who received a dose or only people who completed follow-up. Were symptoms actively asked about? Were repeat events counted once per person or separately? Treatment stopping, missing reports and additional treatment can change interpretation.213

Why post-market signals need context

After approval, more people may use a product for longer and in different circumstances. Spontaneous reports, registries, insurance claims, health records and dedicated studies can reveal patterns.

Spontaneous reports can flag a possible problem, but reporting habits, publicity and missing information affect them. A cluster alone does not establish how often an event occurs or prove causation. Registry and health-record studies can include comparison groups, though their findings depend on how those groups were selected and other possible biases. Regulatory systems may require rapid reporting of serious, unexpected suspected events. A report does not itself prove a product caused the outcome.111210

Relative numbers need an absolute denominator

Relative risk compares the proportion with an event in two groups. It does not show how common the event was to begin with. Imagine a one-year study where two of 1,000 people without treatment and three of 1,000 with treatment have an event. The relative risk is 1.5, but the absolute difference is one additional person per 1,000 over that year.

A headline can make a relative change sound large while hiding a small absolute difference, or the reverse when starting risk is high. Read event counts, group sizes and follow-up together. Check whether the study counts people or repeated events.213

A five-question check

What was observed, and how was it defined? How many people experienced it in each group and over what period? What was the background rate? Were severity, seriousness and treatment stopping reported separately, and how many people still contributed data? What evidence supports a cause, and what could be due to chance, illness or another exposure?

If details are missing, the conclusion may be only that an unwanted event was reported or a signal needs more study. This page does not tell you to start, stop or change a medicine.7892