Background: There are many issues one needs to consider when examining adverse effects or adverse events (AEs) in a meta-analysis. We plan to address these issues and present some sensitivity analyses on them using two systematic reviews conducted by our research group.
Objectives: 1) To summarize and present possible solutions to issues regarding the analysis of AEs. 2) To analyse some of these differences with data from actual systematic reviews.
Discussion: The issues we will address include: - Quality of reporting adverse events and effects (i.e. table versus text, exact numbers versus generalities). - Use of non-controlled (or observational) studies in the analysis. - Use of exact methods to combat problems of non-normality when events are rare. - Considerations when analysing adverse events (events that could be caused by an intervention) as opposed to adverse effects (effects that are known to be caused by an intervention). - Non-independence (i.e. total number of event by group, not by patient) - Sensitivity analyses that can be conducted to deal with the above issues.
Methods: Two large systematic reviews conducted by our research group will be examined with respect to methods of analysing AEs. One review looked primarily at adverse effects and included all observational studies in the analysis of event rates. The other review looked solely at adverse events and thus included only RCTs in the primary analysis. Each of these analyses will be redone with or without the observational studies and compared to the original results. We will also compare results in the two reviews of sensitivity analyses that include or do not include studies that have various reporting methods and determine the effect of using exact methods (where applicable) to calculate our result.
Results: The results of this study will be available and presented at the Cochrane Colloquium, 2004 in Ottawa.
Conclusions: We hope to shed some light on some of the issues regarding the analysis of both adverse events and adverse effects, using quantitative data where appropriate to highlight the issues.