Adverse Events & Tolerability in Trials

An adverse event table is not a list of proven side effects. It is a record of unfavorable medical occurrences observed under defined collection rules. Some events may be related to the studied intervention; others may reflect the underlying condition, background risk, another exposure, or chance.

FDA’s ICH E2A guidance and current IND safety-reporting resources distinguish an adverse event from a suspected adverse reaction and define “serious” using outcomes such as death, immediate risk of death, hospitalization, significant incapacity, or congenital anomaly.[1][2] Accurate summaries preserve those distinctions instead of collapsing every event into “caused by” or every intense symptom into “serious.”

Adverse event describes occurrence, not causality

ClinicalTrials.gov defines an adverse event for results reporting as an unfavorable medical occurrence temporally associated with research participation, whether or not it is considered related.[4] Depending on the protocol, this can include symptoms, diagnoses, physical findings, or laboratory abnormalities.

Temporal association means the event happened during the specified observation period. It does not establish that the intervention produced it. Causal assessment may consider timing, the nature of the event, alternative illnesses or exposures, relevant history, aggregate patterns, and whether the person making the attribution was blinded to the assigned group.[2][5] Individual attribution can remain uncertain.

The terms separate occurrence, causal attribution, and expectedness rather than forming a single ladder:

Term	Core meaning	Causality required?
Adverse event	Unfavorable occurrence during the defined observation context	No
Suspected adverse reaction	Adverse event for which evidence suggests a reasonable possibility of causation	Some evidence; less certainty than an adverse reaction
Adverse reaction	Adverse event caused by the intervention	Yes, within the applicable framework
Unexpected adverse event or suspected adverse reaction	Not listed, or not consistent in specificity or severity, with the applicable reference safety information	Expectedness itself is separate from seriousness; a suspected adverse reaction still requires evidence suggesting causation

FDA’s IND page provides the operative U.S. definitions and notes that aggregate analyses may be needed for events that are common in the study population.[2] A study summary should state whose attribution was reported, such as the investigator, sponsor, or another prespecified process.[2][5]

Seriousness is not the same as severity

Severity describes intensity, such as mild, moderate, or severe. Seriousness is a regulatory and clinical classification based largely on outcome or required action. ICH E2A explicitly warns that the terms are not synonymous.[1]

A severe headache can be intense without meeting a seriousness criterion. A clinically less intense event can be serious if it results in hospitalization or another qualifying outcome. “Life-threatening” means the participant was at immediate risk of death when the event occurred, not that the event might have caused death if it had been worse.[1][2]

Important medical events may be treated as serious when they jeopardize the participant or require intervention to prevent a listed serious outcome, even without immediate hospitalization or death. The report should identify the criterion rather than using “serious” as a general adjective.

Expectedness, relatedness, seriousness, and severity are separate axes

An event can be:

serious but judged unrelated;
non-serious but severe;
related but expected;
unexpected but not serious; or
serious, unexpected, and suspected to be related.

These combinations have different reporting implications. FDA’s current IND resource describes expedited reporting for serious and unexpected suspected adverse reactions and other findings that suggest a significant risk.[2] A public study summary need not reproduce sponsor reporting procedures, but it should avoid treating an expedited case report as a comparative incidence estimate or treating the absence of such a report as proof of safety.

Collection method shapes the result

SPIRIT 2025 asks protocols to specify how harms are defined and collected, including systematic or targeted surveillance, spontaneous reports, time points, coding, severity grading, grouping, and who performs assessments.[5] A checklist administered at every visit and a single open-ended question are different ascertainment methods, so their event counts should not be treated as directly comparable.

Key ascertainment details include:

which events were actively solicited;
which events depended on spontaneous reporting;
observation start and end dates;
visit frequency and follow-up completeness;
coding dictionary and version;
severity and seriousness rules;
whether assessors knew group assignments; and
how repeated events and worsening conditions were handled.

Some reports limit summaries to treatment-emergent adverse events, meaning events that emerge during treatment after being absent pretreatment or worsen relative to the pretreatment state. The pretreatment rule and the handling of pre-existing conditions should be stated. “Treatment-emergent” describes a timing and classification rule; it does not by itself establish causation.[7]

Different collection methods can produce different event frequencies even when the underlying experience is similar. Cross-study rankings are weak when surveillance, definitions, exposure time, and populations differ.

FDA’s current ICH E6(R3) guidance places reliable trial results and participant protection within quality responsibilities spanning the trial lifecycle.[3]

Count participants and events separately

One participant may experience several events. “Thirty events” does not mean “thirty participants,” and the number of participants with at least one event does not show the total burden or recurrence.

CONSORT 2025 recommends reporting, for each group, participants at risk, deaths, withdrawals due to harms, participants with at least one harm, and event counts where appropriate.[6] ClinicalTrials.gov separates all-cause mortality, serious adverse events, and other adverse events in posted results.[4]

ICH E9 says the overall safety and tolerability set is usually defined as participants who received at least one dose, but reports may use other analysis sets. The denominator for each result, the number actually observed, and group-specific follow-up should be stated; randomized, exposed, assessed, and analyzed counts should not be assumed to be equal.[7]

Exposure time also matters. Ten events over ten person-years is not directly comparable with ten events over one hundred person-years. For recurrent events or unequal follow-up, participant proportions alone may omit important information.[6][7]

Stopping study treatment, withdrawing from the study, and being lost to follow-up are different events. A participant who stops treatment may still contribute follow-up, while withdrawal or loss can remove later safety observations. Report reasons and timing by group, whether follow-up continued after treatment discontinuation, and the amount of missing harms data; otherwise low event or discontinuation counts may partly reflect reduced observation rather than better tolerability.[6][7]

Zero observed events does not establish zero risk

Rare events may not appear in a small or short trial. If no serious adverse event occurred in a group, the study observed zero under its sample size and follow-up conditions; it did not prove that the true risk is zero. Confidence intervals or other uncertainty measures help show what event rates remain compatible with the data.[7]

Randomized trials are often designed around an efficacy outcome and may have limited ability to detect uncommon harms. CONSORT notes that trials cannot reliably detect all rare differences and that harms reporting is frequently incomplete.[6] Longer follow-up, larger datasets, and other evidence sources may change the picture, but each source has its own design limitations.

Comparative data are stronger than isolated percentages

An event percentage in one group is difficult to interpret without the comparator, baseline condition risk, observation window, and ascertainment. An apparent numerical imbalance may be due to chance, while a similar percentage can conceal differences in exposure time or event severity.

For each important harm, look for:

event definition and coding category;
participants at risk in each group;
participants affected and total events;
absolute risk difference and relative effect when appropriate;
confidence interval;
observation duration and exposure;
missing follow-up; and
whether the analysis was prespecified or selected after review.

FDA’s ICH E9 guidance treats safety and tolerability evaluation as dependent on the drug, development stage, known information, data collection, and appropriate descriptive and inferential methods.[7] Harms analyses often examine many terms, grades, time windows, and subgroups. Unadjusted p values may be useful as exploratory flags, but a nominally significant result can arise by chance across many comparisons, and a non-significant result does not establish equal safety or the absence of clinically important harm. Prespecification, any multiplicity procedure, effect estimates, confidence intervals, and low-event-count imprecision should remain visible.[7]

What “tolerability” can and cannot mean

ICH E9 defines tolerability as the degree to which a participant can tolerate overt adverse effects.[7] In a trial report, that judgment may be informed by adverse-event type, severity, onset and duration, discontinuations, exposure duration, and other safety variables.[7] There is no single universal threshold that converts those observations into “well tolerated.”

A phrase such as “generally well tolerated” should be unpacked. How many participants discontinued because of adverse events? Were severe or serious events observed? Was follow-up long enough? Were events actively collected? Were vulnerable groups excluded? Was exposure comparable between groups?

Low discontinuation does not prove absence of harm. Participants may continue despite events, investigators may manage events within the protocol, or the study may be too short to observe delayed outcomes. Conversely, discontinuation may reflect inconvenience or unrelated illness rather than a toxic effect.

Study findings do not authenticate another material

Harms data apply to the defined study intervention, formulation, manufacturing context, population, procedures, comparator, and follow-up. A shared ingredient name does not establish that a separately sourced research material has the same identity, composition, impurities, stability, exposure, or risk profile. Neither a low event count nor a regulatory safety report supports a product-quality claim.

Important limitations

Adverse event means temporal occurrence, not proven causation.
Seriousness, severity, expectedness, and relatedness are distinct classifications.
Collection method and surveillance intensity affect which events are found.
Participant counts, event counts, denominators, exposure, and time windows are not interchangeable.
A treatment-emergent classification describes emergence during treatment after pretreatment absence or worsening relative to the pretreatment state; it does not prove causation.
Treatment discontinuation, study withdrawal, loss to follow-up, and missing safety observations are not interchangeable.
Nominal harm p values require prespecification, multiplicity, precision, and low-event-count context.
A non-significant comparison does not establish equal safety or rule out clinically important harm.
Zero observed events does not prove zero risk, especially in small or short studies.
“Well tolerated” is a conclusion that requires its underlying observations and limitations.
Trial harms cannot be transferred to an unverified material or used as medical guidance.
This page explains research reporting and does not provide medical advice, treatment guidance, dosing, preparation, or administration instructions.
These limitations should be considered within the broader Serum Research evidence-review framework.

Frequently asked questions

Does every adverse event come from the studied intervention?

No. An adverse event is an unfavorable occurrence in the defined observation context and does not require a causal relationship.

What makes an adverse event serious?

Seriousness is based on outcomes or actions such as death, immediate risk of death, hospitalization, significant incapacity, congenital anomaly, or an important medical event that threatens those outcomes.

Is a severe adverse event automatically serious?

No. Severe describes intensity; serious is a separate classification based on outcome or required action.

What is a suspected adverse reaction?

It is an adverse event for which evidence suggests a reasonable possibility that the studied intervention caused the event.

Why can two studies report different adverse-event rates?

They may differ in population, follow-up, exposure, definitions, visit frequency, active questioning, coding, missing data, and comparator risk.

Does zero serious adverse events mean the intervention is safe?

No. It means none were observed under the study’s sample size, follow-up, definitions, and collection procedures. Uncommon or delayed risks can remain uncertain.

Does a trial’s tolerability finding apply to a catalog material with the same name?

No. The trial does not verify another material’s identity, composition, quality, or comparability, and its event profile cannot be transferred.

Primary and official sources

Review and corrections

Sources checked: September 1, 2026. Questions or corrections can be sent through the Serum Research contact page.

Editorial responsibility: Serum Research prepared and maintains this educational page. No independent scientific, statistical, regulatory, or clinical review is claimed. See how Serum Research reviews evidence or return to the Research Peptide and Compound Library.

Are you over 21?

Adverse Events, Serious Adverse Events, and Tolerability