Describing rare and serious harms of interventions

Tags: Poster
C Reeves B, Herxheimer A, Wells G, Gyte G

Background: Individuals, and health care professionals advising them, ('consumers' or 'users') need evidence about rare and serious harms, as well as the benefits, of interventions to help them make informed decisions. Evidence about rare and serious harms comes rarely from randomised controlled trials (RCTs); such evidence is sometimes reported from custom-designed cohort studies, but more usually from routine clinical or administrative databases or longitudinal case series of treated patients, or striking case reports. The applicability of RCT evidence, if available, may be weak because RCTs often exclude people most at risk of harms from a new intervention. In other studies, general data quality may be poor and ascertainment of harms is often uncertain; when alternative interventions are compared, the estimated relative risks are susceptible to confounding.

Objectives: To describe key items of information required when reporting rare and serious harms of interventions.

Methods: The list of key items of information proposed has emerged from consideration of a review of evidence about place-of-birth and, specifically, the possible harms of giving birth at home.

Results: There is only one small RCT (n=11) comparing home birth vs. hospital. Calculation of the relative risk of harms to the baby or mother from home vs. hospital birth depends on using data from non-RCTs combined with parameters which are imprecisely estimated in the literature (e.g. the percentage of women planning to give birth at home who are transferred to hospital, and the percentage of home births that are unintended). The evidence reviewed could not answer key questions that women may have, e.g. the risk of a harmful event if giving birth at home and requiring transfer to hospital during labour. This example highlights that the absolute risk of a harmful outcome is probably the most important information for individuals (and, potentially, other users). How this absolute risk varies with different interventions may be unimportant compared with the perceived benefits of an intervention. The estimate of absolute risk should be qualified by describing the population(s) from which it was obtained, information about the characteristics of individuals that influence this absolute risk, and limitations that might compromise the validity of the estimate.

Conclusions: Absolute risks of rare and serious harms in defined populations give users key information they require to make decisions about whether or not to have an intervention. These items of information are simple to understand and are likely to be available. This approach avoids debates about the validity of comparative effects derived from non-RCTs or by other indirect methods. However, it requires methods for assessing the quality of ascertainment of key events and other limitations of primary studies, and standardised methods for pooling event risks, and the effects of key person-level characteristics on event risks, across studies.