Discussion

This study has shown that the strength of evidence used to support the withdrawal of products relied on information recorded from spontaneous reporting together with other pharmacoepidemiological and clinical research studies. All decisions to revoke marketing authorisations described in this study utilised spontaneous case reports, except for the withdrawal of ximelagatran.8 Spontaneous case reporting remains central to pharmacovigilance as it allows for the detection of suspected adverse drug reactions and significant safety signals. However, spontaneous reporting should be used in conjunction with other pharmacovigilance methodologies such as observational studies and clinical trials to further evaluate any such safety signals due to the potential for over and under-reporting, confounding, bias and misclassification. The RCT is commonly assumed to be the gold standard for clinical research9 but adverse drug reactions may not become evident until a medicine is used in the larger population, post-authorisation, where patients are more likely to have comorbidities and other risk factors. The outcome of pharmacoepidemiology studies and the conduct of systematic reviews, even with their limitations particularly bias, confounding and misclassification, can provide useful additional evidence on the nature and frequency of adverse drug reactions. As a direct country comparison, Clarke et al4 stated that six products were withdrawn in the UK in between 1999 and 2001, namely grepafloxacin (marketed for 2 years), cisapride (12 years), pumactant (6 years), cerivastatin (4 years), droperidol (30 years) and levacetylmethadol (4 years). The median length of time these drugs remained on the UK market was 5 years (IQR 4–10.5). Droperidol was available for approximately 30 years as it was marketed before the licensing system began in 1971 (personal communication from MHRA (2013)). In contrast, only three products were withdrawn from the UK market between 2009 and 2011; orciprenaline, sibutramine and rosiglitazone marking a 50% reduction in the number of drug products withdrawn over the same time period. The median length of time these products remained on the market was 11 years (IQR 10.5–30). Although the length of time taken to withdraw a drug product appears to have doubled (mainly due to the length of time orciprenaline was available, 49 years) over the 10-year period, the evidence used to aid the decision-making has been heightened. Spontaneous case reporting was used to support the case for withdrawal in five of the six products between 1999 and 2001, while only one RCT was included as evidence.4 In this study, spontaneous case reporting and RCTs were shown to have contributed to all three withdrawn products, which suggests an increase in the level of evidence used to support decisions for drug product withdrawal from the market.

The findings of this study are in line with those of the study conducted by Paludetto et al6 in 2012 who observed a shift in the nature of the safety data used as justification for the withdrawal of products in France in the time periods 1998–2004 and 2005–2011. One explanation the authors gave for this change was the implementation of the European Clinical Trial Directive in 2004 and the subsequent analysis of clinical trial safety data by pharmacovigilance experts. The evolution of pharmacovigilance legislation over the past decade is likely to have contributed to the availability of additional data through the conduct of post-marketing safety surveillance allowing regulators and companies to include such information in the drug products safety portfolio. It is difficult to ascertain whether the introduction of RMPs for newly licensed medicinal products has influenced the length of time a drug with safety concerns remains on the market as only two drug products included in this study were marketed after 2005. Rimonabant and sitaxentan were licensed in 2006 and removed from the market in 2008 and 2010, respectively. At the time of authorisation, information was known about the psychiatric side effects of rimonabant10 and liver toxicity associated with sitaxentan11 and the benefit-risk balance deemed to be favourable. However, in the case of rimonabant, as data became available when used in the real-life clinical setting, the risk of obese and overweight patients developing psychiatric side effects was double that of patients taking placebo.10 The marketing authorisation holder voluntarily removed sitaxentan from the market and cancelled ongoing clinical trials as a result of two cases of fatal liver injury.11 During the time period studied, an additional 12 drugs were withdrawn in some but not all EU countries. In some cases, it was not feasible to establish the reasons for the withdrawal, for example, astemizole has been withdrawn from the UK, France and Spain, but information about its availability in other EU countries is difficult to establish. Another example, the NSAID nimesulide, was withdrawn in Spain, Finland, Belgium and Ireland but is still available in 15 EU member states. The EMA had a favourable opinion of the risk-benefit balance after a review of spontaneous reports of epidemiological studies and other published studies in 2010.12 This was in spite of evidence of an increased risk of hepatotoxicity compared with other NSAIDs, although restrictions are in place to limit the number of liver-related side effects. Ultimately, it would seem that there is a disparity in opinion of risk-benefit within the EU between decision-makers at the EMA and various regulatory authorities.

The main limitation of this study was the impact of various language barriers across the EU. Websites of competent authorities are written in their local language and resource constraints did not allow for translations. Only information from the websites of those competent authorities where English and French are spoken locally or an English or French version is available was utilised. This may also have had a bearing on the studies included as evidence in this research as again only English and French publications were included. After a website search, competent authorities and pharmaceutical companies were contacted in order to obtain further information on drug product withdrawals. The response rate from competent authorities was 70% and from pharmaceutical companies 55%. Again, this may have been due to a language barrier problem, but it may also be due to a lack of available information on drugs withdrawn for several years or an unwillingness to share information. Competent authorities or companies may also not have a complete list of products removed from the market. Over time, the way in which drug products have been licensed in the EU has changed considerably. It is relatively easy to determine how products licensed centrally were withdrawn from the market. However, for drugs that have been licensed for a considerable number of years, in countries that have recently joined the EU, and for those that have multiple drug names, it can be very difficult to obtain conclusive, consistent and clear information. The effect of different drug names may have led to potential exclusions of study evidence from our PubMed search strategy as the INN of each drug was used rather than an exhaustive drug name list. Owing to time constraints, it was not possible to conduct a full investigation into why drug products had been removed from the market in some but not all countries and to find detailed information about the decision-making process.

It would be useful to investigate the impact of the introduction of RMPs in a future study. This could be carried out as an investigation of how results from postmarketing studies, conducted as part of RMPs, aid in the assessment of any safety issues. In theory, the introduction of RMPs should expedite post-marketing drug safety decisions.

This study has shown that the strength of evidence has improved during the past 10 years. Indeed, the results show an increased use of case–control studies, cohort studies, RCTs and meta-analyses as justification for the withdrawal of a marketed product. Previously, only one-third of decisions used evidence from observational studies or clinical trials4 but this study showed that they had contributed to decision-making in almost two-thirds of withdrawals. Spontaneous case reports remain the most significant method of pharmacovigilance.