Clinical Study Connect

Plain-Language Clinical Study and Ingredient Research

Adverse Events Are Reported as a Table: What to Compare First

posted on

By ClinicalStudyConnect.com Research Desk

The short answer: Before you compare any percentages in an adverse event table, compare three things across the study groups. First, the number of participants who were actually assessed for harms in each group (the denominator). Second, the absolute count of participants affected in each group. Third, the time frame over which harms were collected. If any of those three differ between groups, or are missing, the percentages cannot be read at face value.

This page explains why that order matters, what the standard tables on ClinicalTrials.gov contain, and how to use a short harms reading card when you open a study’s results.

What an Adverse Event Table Is Reporting

An adverse event is not the same thing as a side effect caused by the intervention. In trial reporting, adverse events are recorded because they happened during the study period, whether or not anyone thinks the intervention caused them. ClinicalTrials.gov describes its adverse event tables as summarizing events that were temporally associated with the research, whether or not considered related to participation.

That matters for reading. A table of adverse events is a record of what was observed in each group. The comparison between groups, not any single row, is where the information about the intervention sits.

On ClinicalTrials.gov, results records present harms as three tables:

  • All-cause mortality: deaths from any cause in each group. This table is required for studies with a primary completion date on or after January 18, 2017.
  • Serious adverse events: events meeting the serious definition, such as death, a life-threatening event, hospitalization or a longer hospital stay, significant incapacity, or a congenital anomaly.
  • Other (not including serious) adverse events: non-serious events reported above a frequency threshold that the study chooses, anywhere from 0 to 5 percent within any group.

The ClinicalTrials.gov definition of “serious” is based on what happened to the participant, such as hospitalization or a life-threatening event. Current trial reporting guidance treats seriousness and severity as separate ways of grouping harms, so check which one a table is using.

Compare First: Who Was Counted in Each Group

Every harms table has a “number at risk” or “number assessed” for each group. This is the denominator: the participants who could have had the event and were checked for it. It is not always the same as the number enrolled.

Here is an illustrative example, not taken from any real study:

  • Group A: 12 participants with headache out of 200 assessed = 6 percent
  • Group B: 12 participants with headache out of 100 assessed = 12 percent

The raw counts match, but the risk in Group B is twice as high because the denominator is half the size. Reading the counts without the denominators would hide that. Reading the percentages without checking the denominators would hide how few people sit behind them.

Also check how the at-risk population was defined. ClinicalTrials.gov asks studies to explain how the number of participants assessed was determined. Compare that number with how many were randomized to each group. If they differ, look for the explanation.

Then Compare Absolute Counts, Not Just Percentages or Ratios

Once the denominators are clear, look at the actual number of participants affected in each group before you look at any relative comparison.

Another illustrative example:

  • Intervention group: 6 of 200 participants with an event (3.0 percent)
  • Comparison group: 3 of 200 participants with an event (1.5 percent)

A relative comparison says the risk doubled. An absolute comparison says the difference was 1.5 percentage points, or 3 additional participants per 200. Both statements are true. Neither is complete alone. Reporting guidance for randomized trials (CONSORT Harms 2022) recommends presenting both absolute and relative effect sizes for harms, with a measure of precision such as confidence intervals.

Two more count-related checks:

  • Participants versus events. ClinicalTrials.gov tables report the number of participants affected. The number of events is optional. Ten events could mean ten people with one event each, or one person with ten events. If a paper reports only events, you cannot tell how widely a harm was spread across participants.
  • Small numbers move a lot. When counts are in single digits, one or two participants can change a percentage substantially. Treat differences built on very few events as uncertain, and look for a confidence interval.

Check Duration Before Trusting Any Comparison

A percentage only means something alongside the period it covers. Five percent of participants over two weeks is a different finding than five percent over two years.

ClinicalTrials.gov requires a time frame for adverse event collection and advises studies to state it from the participant’s perspective, such as “8 weeks after participant received first dose,” rather than a vague “end of study.” Look for that time frame first. If it isn’t stated, the table cannot tell you whether harms were counted over the same window in each group.

Duration becomes a bigger problem when groups were followed for different lengths of time. This happens when participants in one group stop treatment or leave the study earlier. Methodological research published in Trials notes that standard contingency-table comparisons can mislead when follow-up times vary and differ between groups. Some studies respond by also reporting exposure-adjusted incidence rates, which divide by total follow-up time rather than by headcount. Those rates have their own limits too. The same research shows their validity depends on how risk is distributed over time, so they are best read as one piece of the picture, not a correction that settles the question.

The practical reading step: if one group was followed for noticeably longer, it can accumulate more recorded events for that reason alone. Look for whether the authors accounted for it.

How the Harms Were Collected

ClinicalTrials.gov asks studies to state whether adverse events were collected through systematic assessment, such as a questionnaire given to every participant, or non-systematic assessment, such as events participants mentioned on their own.

This changes what the numbers can support. CONSORT Harms 2022 advises caution with relative or absolute risk comparisons for non-systematically assessed harms, because those outcomes were not actively determined. If two studies of the same intervention used different collection methods, their adverse event rates may not be directly comparable.

What a Missing Row Does and Does Not Mean

An event missing from the “other adverse events” table does not mean it never happened. It may have fallen below the study’s reporting threshold, which can be up to 5 percent. Check the stated threshold before concluding anything from an absence.

Reporting guidance addresses this directly. CONSORT Harms 2022 asks authors to report zero events explicitly when no harms were observed. It also asks them to explain why any harms outcomes were omitted and where that data can be found. A published paper that summarizes harms in a sentence, with no table, gives you much less to work with than a full results record.

The Harms Reading Card

Use this card in order when you open an adverse event table in a paper or on ClinicalTrials.gov. Each step depends on the one before it.

  1. Groups: Are the group labels descriptive (for example, “placebo” or a named dose), and does the description say what each group received, how much, and for how long?
  2. Denominators: What was the number at risk in each group, and how was it determined? Does it match the number randomized, or were participants excluded?
  3. Time frame: Over what period were harms collected? Was it the same window in every group?
  4. Collection method: Systematic (actively asked) or non-systematic (volunteered)? Was a standard coding dictionary used, and is it named?
  5. Absolute counts: How many participants were affected in each group, in whole numbers?
  6. Participants versus events: Is the table counting people or occurrences?
  7. Serious events and deaths: What do the serious adverse event and all-cause mortality tables show for each group, even when the counts are small?
  8. Threshold: What frequency threshold was used for non-serious events, and could an event of interest sit below it?
  9. Relative and absolute comparisons: If a ratio is reported, can you also see the absolute difference and a confidence interval?
  10. Discontinuations: How many participants stopped the intervention because of harms, in each group?

If you cannot answer steps 2 and 3, stop there. Any percentage further down the table is uninterpretable without them.

What Remains Uncertain

  • A table shows only what the study could see. Harms that are rare, or that take longer to appear than the study’s time frame, may not show up in any single trial’s table. An empty row is a statement about this study’s size and duration, not a guarantee.
  • Reporting is still uneven. The CONSORT group updated its harms reporting guidance in 2022 because the 2004 version had not been consistently applied. Do not assume a published trial reports every item on the card above.
  • Relatedness ratings are judgments. Some studies rate whether each event was “related” to the intervention. These are investigator assessments. An adverse event record by itself does not establish that the intervention caused the event.

The Next Practical Step

If a paper’s harms section is thin, look up the study’s ClinicalTrials.gov registration number (these begin with “NCT”) and check whether a results record with the three standard tables is posted. The registry record may include denominators, time frames, and thresholds that the paper left out.

Important Limits of This Page

This article explains how adverse event data are reported in clinical studies. It is educational only and does not assess the safety of any specific product, supplement, or medication. It is not medical advice. A study’s adverse event table cannot tell you how a treatment will affect you personally. Do not start, stop, or change any medication or supplement based on a study table. Talk with a physician or pharmacist, who can weigh study findings against your own health history. If you think you are having a serious reaction to anything you have taken, seek medical care promptly.

Sources