An Exploration of Evidence Synthesis Methods for Adverse Events
thesisposted on 21.03.2012, 16:56 by Fiona Claire Warren
Adverse events following the use of medical interventions are a major source of concern for patients, healthcare professionals and pharmaceutical companies. Therefore, evidence synthesis of potential adverse events are very important in determining whether an association exists, and the strength of such an association. It is also desirable to be able to quantitatively balance potential harms against the benefits of the intervention. However, standard statistical techniques for meta-analysis are often unsuitable when applied to datasets where the primary intervention is an adverse event. A review of standard meta-analysis methods, including Bayesian methods, is conducted. The specific challenges of meta-analysis in relation to adverse events datasets are described, with some of the main areas of contention being sparsity of events, subgroup analysis, class effects with regard to drug interventions, and issues related to time factors within the individual studies. Methods used in existing meta-analyses where the primary outcome is an adverse event have also been reviewed; this demonstrates the methods already used in this field and highlights some of their limitations, and where the methods could be extended. In the light of the reviews of methods and previous meta-analyses, four case-studies are performed. The first uses data from GlaxoSmithKline to investigate a potential relationship between paroxetine and suicidality, using many of the standard methods for comparison purposes. The second uses individual patient data for a time-to-event analysis of anti-TNF drugs used for rheumatoid arthritis. This clinical example is extended by the use of Mixed Treatment Comparisons for within-class comparisons, and to assess the effect of dose. Finally, a harm-benefits model is used to assess the interplay of risk of endometrial cancer against breast cancer recurrence for tamoxifen users. These models present novel ways of analysing adverse events data and demonstrate some of the difficulties in their use.