Integrating evidence from multiple sources to evaluate post-approval safety: an example of sildenafil citrate and cardiovascular events.

OBJECTIVE

Recent high-profile medicine withdrawals have highlighted the complex decision-making process that regulators, pharmaceutical companies, prescribers, and patients must undertake in determining whether a drug has an appropriate benefit-risk balance. Our objective was to analyze the utility of different drug safety data sources and methods, using the experience of sildenafil citrate (Viagra) and post-approval concerns about its potential association with cardiovascular (CV) events (i.e., myocardial infarction [MI] and death) as a case study.

METHODS

We evaluated safety data from three sources: the standard passive surveillance system (i.e., spontaneous reports filed to Pfizer Inc), pooled clinical trial data, and a prospective observational cohort study, the International Men's Health Study (IMHS).

RESULTS

More than 28 000 spontaneous reports were received in the first 7 years after approval. Between 2001 and 2005, the proportion filed by persons other than healthcare professionals (61%) was approximately double the proportion averaged across five other drugs from the manufacturer's safety database. CV events and/or deaths represented 22.0% of reports, and 23% of reported deaths were medically unconfirmed reports made by persons other than healthcare professionals. In contrast, MI and all-cause mortality rates for sildenafil from both the pooled clinical trial data and the IMHS were similar to placebo, despite differences in methods and populations.

CONCLUSIONS

These results suggest that passive surveillance may generate apparent signals of risk, as was the case with sildenafil and CV events. However, to adequately assess the benefit-risk profile of a drug, these signals must be evaluated via other data sources such as clinical trial and epidemiologic studies, as the apparent signal was not supported by more rigorously collected data. Our post-marketing analysis was unable to examine all potential influences of spontaneous reports, and the study data sources (although large for erectile dysfunction studies) were not designed to exclude small CV risks.