Explanation

Readers need information about the harms as well as the benefits of interventions to make rational and balanced decisions. Randomised trials offer an excellent opportunity for providing harms data, although they cannot detect differences in uncommon or rare harms between treatment groups. The existence and nature of adverse effects can have a major impact on whether a particular intervention will be deemed acceptable and useful. Not all reported adverse events observed during a trial are necessarily a consequence of the intervention; some may be a consequence of the condition being treated. Nevertheless, they all need to be reported.

 

Many reports of randomised trials provide inadequate information on harms. A comparison between harm data submitted to the trials database of the National Cancer Institute, which sponsored the trials, and the information reported in journal articles found that low grade adverse events were under-reported in journal articles. High grade events (Common Toxicity Criteria grades 3 to 5) were reported inconsistently in the articles and the information regarding attribution to investigational drugs was incomplete [474]. Moreover, a review of trials published in six general medical journals in 2006 to 2007 found that while 89% of 133 reports mentioned adverse events, no information on severe adverse events and withdrawal of patients owing to an adverse event was given in 27% and 48% of articles, respectively [475]. In a later review of 196 randomised trials of invasive pain treatments published in six major journals, 76% provided the denominators for analyses on harms and 85% reported the absolute risk per arm and per adverse event type, grade, and seriousness, and presented appropriate metrics [476].

 

For non-systematically assessed harms, reporting can be more complex as the information is not standardised. A common approach is to code the event declared by participants. Authors should report the coding system used, whether coding was prespecified in the protocol, in the statistical analysis plan, or post hoc, and whether coding was performed by researchers blinded to the treatment allocated. In addition, there is a risk of under-reporting and selective non-reporting of harms particularly for non-systematically assessed harms. A reanalysis of individual participant data from six randomised trials of gabapentin found evidence of important harms that were not disclosed in the published reports but identified after data sharing and reanalysis [477]. Sharing of de-identified individual participant data may be needed to be able to adequately synthesise this information, for example for inclusion in a systematic review [231, 235, 236].

 

Authors should report for each group, the number of participants at risk, the number of deaths, the number of participants withdrawn due to harms, the number of participants with at least one harm event, and the number of events, if appropriate. Where appropriate, the estimated effect size with its precision (such as 95% CIs) should be reported including both absolute and relative effects for binary outcomes. It is important to separate the reporting of systematically and non-systematically assessed harms. Systematically assessed harms should be reported even if zero events were identified. It should also be clear whether the authors are reporting the number of participants with at least one harm event or the number of events per unit of time at risk and whether recurrent events were included. The number of participants withdrawn because of harms should also be reported for each group. Finally, results should be reported for all harms. We strongly discourage the use of thresholds or criteria to select which harms should be reported. All harms could be detailed in supplementary materials.

 

We recommend reporting the results in a table with the results for each trial arm [478]. More detailed information can be found in the CONSORT statement extension for harms, which was updated in 2022 [20].