Adverse Events and Safety Outcomes

Overview

Safety analysis describes:

• treatment exposure;
• adverse events;
• serious adverse events;
• discontinuations;
• laboratory abnormalities;
• deaths;
• events of special interest.

Safety outcomes are often multiple, recurrent and rare.

The analysis should distinguish the number of participants affected from the number of events.

Why this matters

A participant can experience several headaches but counts once in the participant incidence.

Comparing raw event counts can be misleading when exposure time differs.

Absence of statistical significance does not establish safety, especially for rare serious events.

When to use this framework

Use it for phase I–IV trials, pragmatic trials and post-authorisation studies.

The safety analysis population often includes participants who received at least one dose, analysed by treatment received, but this must be distinguished from the efficacy estimand.

Research question and variables

Key variables include:

• treatment received;
• first and last exposure dates;
• event term;
• seriousness;
• severity;
• relatedness;
• onset and resolution;
• recurrent event count;
• discontinuation;
• person-time.

Data structure

Safety data levels

simulated data

Level	Example
Participant	Any headache during follow-up
Event	Three headache episodes
Exposure	84 treatment-days
Severity	Mild, moderate, severe
Seriousness	Regulatory serious criteria

Severity and seriousness are not synonyms.

Statistical assumptions

Participant incidence

For an event:

Risk difference is:

$$RD=p_1-p_0$$

Report event counts alongside participant risks when recurrence matters.

Exposure-adjusted incidence rate

When follow-up differs:

$$IR= \frac{\text{number of events}} {\text{person-time at risk}}$$

Exposure-adjusted rates require clear risk windows and recurrent-event assumptions.

They should not replace participant incidence automatically.

Worked example

Worked analysis in R

safety <- read.csv(
  "data/part-13-safety-data.csv"
)

risk_difference <- function(
  event,
  group
) {
  p1 <- mean(
    event[group == 1]
  )
  p0 <- mean(
    event[group == 0]
  )

  se <- sqrt(
    p1 * (1 - p1) /
      sum(group == 1) +
    p0 * (1 - p0) /
      sum(group == 0)
  )

  c(
    risk_control = p0,
    risk_intervention = p1,
    difference = p1 - p0,
    lower = p1 - p0 -
      1.96 * se,
    upper = p1 - p0 +
      1.96 * se
  )
}

rbind(
  headache = risk_difference(
    safety$headache,
    safety$group
  ),
  nausea = risk_difference(
    safety$nausea,
    safety$group
  ),
  serious_ae = risk_difference(
    safety$serious_ae,
    safety$group
  )
)

Selected analysis output

The intervention group has slightly higher simulated headache and nausea risks.

Serious adverse events are uncommon and confidence intervals are wide.

Visualising safety effects

Multiplicity

Safety analyses often contain hundreds of terms.

Formal adjustment can be useful for predefined safety outcomes, but routine p-value screening of every adverse-event term is misleading.

Prioritise:

• prespecified events of special interest;
• absolute risks;
• severity;
• timing;
• biological patterns;
• consistency across data sources.

Rare events

Individual trials are usually underpowered for rare harms.

Evidence may require:

• pooled trials;
• meta-analysis;
• pharmacovigilance;
• registries;
• long-term follow-up.

Zero observed events does not imply zero risk.

Interpreting the output

Presenting the results

Suggested reporting

Use CONSORT harms guidance together with CONSORT 2025 [1].

ICH E3 describes clinical-study-report safety displays [2].

Broader recommendations for safety planning and current trial reporting complement the harms-specific guidance [3], [4].

Common mistakes

Knowledge check

What is participant incidence?

Answer and reasoning

The proportion of exposed participants experiencing at least one event.

Why report total events as well?

Answer and reasoning

One participant can experience recurrent events, which participant incidence conceals.

Does p greater than 0.05 establish safety?

Answer and reasoning

No. Rare-event estimates may be highly imprecise.

When are exposure-adjusted rates useful?

Answer and reasoning

When observation or treatment exposure differs meaningfully across groups.

What should be prioritised statistically?

Answer and reasoning

Prespecified events of special interest, absolute risks, uncertainty and clinically coherent patterns.

Key take-home messages

References

1. Ioannidis JPA, Evans SJW, Gøtzsche PC, O’Neill RT, Altman DG, Schulz K, et al. Better reporting of harms in randomized trials. Ann Intern Med. 2004;141(10):781–788. https://doi.org/10.7326/0003-4819-141-10-200411160-00009
2. International Council for Harmonisation. ICH E3: structure and content of clinical study reports. Geneva: ICH; 1995.
3. Crowe BJ, Xia HA, Berlin JA, Watson DJ, Shi H, Lin SL, et al. Recommendations for safety planning, data collection, evaluation and reporting. Clin Trials. 2009;6(5):430–440. https://doi.org/10.1177/1740774509344101
4. Hopewell S, Chan AW, Collins GS, Hróbjartsson A, Moher D, Schulz KF, et al. CONSORT 2025 statement. BMJ. 2025;389:e081123. https://doi.org/10.1136/bmj-2024-081123