Stefano Garzon

Research · Oct 23, 2026

Assessment and reporting of harm in clinical trials

Concept

An adverse event is any untoward occurrence during treatment, whether or not a causal relation is suspected. It occurs in both arms and is compared between them. An adverse effect is a noxious response for which a causal relation is at least a reasonable possibility. Harm is the sum of adverse effects. Severe measures intensity; serious measures medical consequence: death, life threat, initial or prolonged hospitalization, disability, congenital anomaly, or need for intervention to prevent harm.

Benefit has a prespecified outcome and a sample size calculated for it. Harm rarely does. The number of possible events is large, many are unknown at the start, and each has frequency, severity and duration.

Strengths and limits of the trial

A randomized trial has a comparable control group, allows events of interest to be defined in advance, reduces assessment bias when blinded and follows participants systematically.

There are four limits. The population is selected and lower risk; older people, patients with comorbidities and users of other drugs are often excluded. Power is low for rare events: about 3,000 participants are needed to observe one case with 95% probability if the true incidence is 1 in 1,000. Duration is short for late effects. Effects not looked for are not found: without ECGs, arrhythmia goes undetected; without HbA1c, diabetes does.

Venning reviewed 18 important adverse reactions; clinical trials played a key role in identifying 3. Among new FDA boxed warnings in 2009, 76% came from spontaneous reports.

Collection: volunteered or solicited

Open questions (“any health problems?”) capture unexpected events and tend to record the more intense ones. Structured questionnaires record more events in a standard way. In AMIS, direct questioning raised the proportion of complaints in both groups, and the aspirin-placebo difference appeared with both methods. For sensitive symptoms the gap is larger: sexual dysfunction with antidepressants was reported by 46.5% with a specific scale, against 1% to 2% by spontaneous report in fluoxetine trials.

Classification and time

MedDRA is the standard dictionary. Its roughly 20,000 preferred terms fragment a single problem: depression can be coded under 22 terms. Counts per term become small and a real difference can be diluted. Aggregating at higher levels or with standardized queries (SMQs) requires clinical judgment and should be described.

Timing matters. In AMIS, stomach pain with aspirin peaked early and then declined. In DCCT, intensive glycemic control caused early retinal worsening but reduced long-term progression of retinopathy; stopping on the early finding would have led to the opposite conclusion.

Analysis

Reporting

Ioannidis and Lau assessed 192 trials: severity of clinical adverse effects was adequately defined in 39%, and 46% gave reasons for discontinuation due to toxicity. The CONSORT harms extension (2004) has ten recommendations, among them defining each event, describing how it was collected, giving denominators and presenting absolute risk per arm. Pitrou and colleagues, in 133 trials from 2006, found 47.4% with no information on withdrawal due to adverse events and 32.3% with restricted reporting, for example only the most common events.

A useful severity hierarchy is: permanent discontinuation, permanent dose reduction, temporary reduction, complaint without change.

Common errors when reading a paper

What to check when reading a trial

  1. How events were collected: open question, questionnaire, tests.
  2. Which events of special interest were prespecified and how they were defined.
  3. Denominators and absolute risk per arm, including serious events.
  4. Dropout for adverse events, dose reduction and discontinuation, by group.
  5. Duration of exposure and timing of event onset.

Related topics: absolute and relative effect, multiple comparisons, meta-analysis.

References